Most finance professionals use AI the same way every day. They open a chat, type a question, read the answer, and close the tab. The work evaporates. Tomorrow they type the same question again.

That is the search-box trap. It feels productive. It compounds nothing.

The fix is a system. The advantage in finance comes from converting your proprietary data, your methodology, and your controls into a governed system that executes repeatedly. A prompt is a keystroke. A governed system is an asset.

This matters especially in finance, because a financial agent that can act needs three things before it does: governed access to data, encoded operating rules, and explicit authority boundaries. Give it access without rules and controls and you have not built infrastructure. You have built a fast way to make an unlogged mistake.

The tools people list as an AI stack are usually a pile. Interfaces, operating patterns, reusable instructions, cost tricks, and controls, all thrown together as if they were interchangeable steps. They sit at different layers and do different jobs.

Here is the actual stack: four layers that do different jobs. They are logical layers, not a chronological build sequence.

Layer 1: Access. How the agent reaches your data

An MCP server exposes a system as a set of tools the model can call through a standard interface. A browser agent is the fallback for the systems that never shipped an API or an export. Access is the connection between the model and your systems.

Be precise about what this buys you. MCP standardizes connectivity. It does not create universal memory. The client still has to connect and authenticate, the tools still need clear descriptions and scoped permissions, and the server still needs maintenance when the underlying schema or auth changes. What you get is real but bounded: any authorized agent can reach the same governed interface without you re-assembling context by hand every time.

Payments example: a governed interface to your ledger, positions, and reserve balances. Any authorized workflow can then check settlement exposure or reserve coverage against a live source rather than a pasted snapshot. What matters is that access is governed and repeatable.

Browser agents belong here too, but treat them as a last resort. They are brittle, they break when a page changes, and pointing them at a source that restricts automated access is a terms-of-service and data-rights problem, not a clever hack. Use them to navigate authorized web workflows when no stable API or export exists.

Layer 2: Method. How the agent does the work your way

Access without method just moves raw data around. Method is where your analytical process lives. This is skills plus structured schemas and deterministic calculation.

A skill encodes your method. Your inputs, your steps, your output schema, and your review criteria, captured once and reused. Be honest about the limit. A skill does not clone your judgment. It encodes your procedure and your standards so routine cases run consistently and difficult exceptions reach you already framed.

The other half of method is refusing to let the model do arithmetic it should not. Anything that must be exact, a fee calculation, a reserve ratio, an interest accrual, a settlement amount, runs as deterministic code against a fixed schema. The model orchestrates and explains. It does not freehand the numbers.

Payments example: a reconciliation-exception skill. It carries how you classify a break, what evidence it pulls, the schema it returns, and the threshold that forces a human look. The categorization is consistent because the method is encoded, not re-improvised each run.

Layer 3: Execution. How the work actually runs

Execution is the operating pattern. Sub-agents, scheduled runs, batch processing, and prompt caching all live here. These are about throughput and cost, not about what the agent knows or is allowed to do.

Sub-agents parallelize independent work. The key word is independent. Split a job into slices that do not depend on each other, run them at once, then reconcile the results through a common schema and a review step. Parallel agents on interdependent work create rate-limit pressure, reconciliation headaches, and quality drift. Use them to fan out extraction across many payout files or many merchant records, then converge.

Scheduled runs turn a workflow into recurring production. This is the layer people describe as no human in the loop, and in finance that phrase is a red flag. Unattended research and monitoring are straightforward. Unattended decisions and transactions are acceptable only inside pre-approved limits, with deterministic controls and escalation when an exception, threshold, or material change occurs. The correct design runs autonomously inside defined authority boundaries and stops for human approval when it reaches an exception or consequential decision.

Prompt caching and batch processing are cost levers, not durable memory. Caching reuses stable prompt prefixes across repeated calls to reduce input cost and latency, but retention is temporary and provider-specific, typically lasting minutes to an hour. Batch processing trades immediate completion for lower cost and higher throughput. Use it for delay-tolerant jobs, not interactive workflows.

Payments example: a scheduled cross-border payout-failure analysis that runs overnight, fans out across the day’s failed payouts with sub-agents, and lands a categorized report by morning. It recommends. It does not re-attempt payments on its own.

Layer 4: Control. Why a finance leader can actually trust it

This is the layer the hype skips, and it is the one that separates automation from infrastructure. Without it you do not have a financial system. You have a script with confidence.

Control is more than one feature. Hooks are part of it: deterministic checks that fire on an event and fail closed when they do not pass. A hook does not guarantee correctness, and judgment-based review stays probabilistic, but a hook can block a run when a hard rule is violated. That is worth a lot.

The rest of the layer is the unglamorous list finance already knows by heart: scoped permissions and managed secrets, data lineage and freshness checks, evaluations that measure whether the workflow still performs, audit logs that reconstruct exactly what the agent saw and did, monitoring with retries and alerting, explicit human approval at the points where value moves, and version control on every rule and prompt. None of this is exotic. All of it is the difference between something you demo and something you would put near a real balance sheet.

Payments example: an agent that can flag a treasury sweep or a merchant-risk case runs inside scoped permissions, logs every input and decision, checks the freshness of the balances it read, and requires a human to approve anything that actually moves money or changes a limit. The intelligence is a component. The control stack is the product.

Control is shown as a layer, but it is not a final step. It has to govern every layer from the start.

What to build first

The common advice is to build the custom MCP server first. For most finance teams that is backwards. It optimizes for architecture before the workflow has proven it deserves any.

Build a skill first. Take one repeatable workflow, encode its inputs, method, output schema, and review criteria, and run it manually until it produces consistent results. This is the cheapest way to prove value and the fastest to show it.

Schedule that workflow second. Turn the proven skill into recurring production, with logging, retries, delivery, and a human-approval boundary wherever a decision has consequences.

Build the MCP server third, when manual data retrieval has become the actual bottleneck. Now the connection earns its maintenance cost, because a workflow that works is waiting on the data.

Skill, schedule, server. Method before plumbing. Prove the work, then make it run, then remove the friction.

The real divide

AI use in finance will become standard. The gap will be between people who treat it as a chat and people who treat it as governed infrastructure.

The chat user retypes the context every morning and hopes the answer is right. The infrastructure builder encodes the method once, runs it on a schedule, wraps it in controls, and can prove what happened. Same model underneath. Completely different leverage, and only one of them is defensible when someone asks what the agent did and why.

Financial agents will get access to data, and soon after, authority to act. The firms that win will be the ones that gave them encoded rules and hard boundaries first. Start with one skill and one scheduled run, wrapped in real controls. That is the whole beginning. Everything else compounds from there.

The views expressed here are my own and do not represent those of any employer or organization. This is commentary and analysis, not investment, legal, or tax advice.

Most payments commentary is opinion about rails the writer has never touched. I wanted the opposite. So I built three things on a testnet: a USDC escrow contract, an AI agent that decides whether to release or refund, and a cross-chain transfer that moves USDC between networks with no bridge. Everything below ran on test networks with test dollars. That is a deliberate rule, not a limitation. You do not learn programmable money by moving real money before you understand the failure modes.

Here is what building it taught me that reading about it never did.

The escrow contract: money that releases on a condition, not on trust

The core piece is a state machine. The escrow moves through four stages: open, locked, released, refunded. A depositor funds it with USDC. Once funded, it locks. From locked, it can only go to released, where funds go to the beneficiary, or refunded, where funds go back to the depositor. There is no fifth path. That constraint is the product. The money cannot move in a way the rules did not allow.

The interesting part is not the happy path. It is the gotchas that only appear when you actually deploy.

USDC does not just move because you tell it to. The contract cannot pull a depositor’s dollars unless the depositor first approves it to do so. Approve, then transfer. Miss that step and the deposit reverts. This two-step approval model is the single most common thing newcomers get wrong, and it is invisible until it fails.

USDC transfers can fail quietly. A naive contract assumes a transfer succeeded and moves on. A correct one checks the result or uses a safe-transfer wrapper. Assume success and you can mark money as sent that never left.

USDC has six decimals, not eighteen. One dollar is 1,000,000 in the smallest unit. Get the decimals wrong and you move a millionth or a million times what you meant to. That is not a rounding error. That is a wire to the wrong universe.

Release and refund need reentrancy protection. The moment a contract sends money out, a malicious token or contract can try to call back in before the first call finishes. Guarding those functions is not optional. It is the difference between an escrow and a drain.

None of this is exotic. All of it is the boring, unglamorous discipline that separates a demo from something you would trust with real value. Programmable settlement is easy to describe and unforgiving to build.

The AI agent: the model is the easy part, the guardrails are the product

The second piece is an off-chain agent that decides the escrow’s outcome. It reads the agreement terms and the evidence, delivery proof or a dispute, and returns a verdict: release or refund, with a reason. Then it calls the contract.

This is where agentic finance gets real, and where it gets dangerous. A few rules made the difference between a useful tool and a liability.

The agent outputs strict JSON, never prose. A verdict is a machine instruction, not an essay. If the output is not valid structured data with an explicit action, nothing fires.

There is a human confirm step before any transaction moves money. The agent recommends. A person approves. That is not a lack of ambition. It is the only responsible default when a wrong decision is irreversible and on-chain.

When the agent is uncertain, it does nothing and escalates. No confident guess on a coin flip. Uncertainty routes to a human, not to a transaction.

Every decision is logged to an audit trail before execution. If an agent moves value, you need to reconstruct exactly what it saw and why it chose. Reasoning without a record is not governance.

The lesson is blunt. The language model is the easy part. Anyone can get a verdict out of a model. The hard, valuable work is the guardrails: strict output, human confirmation, uncertainty handling, and an audit trail. In agentic settlement, the model is a component. The control stack is the product.

Cross-chain with no bridge: burn on one side, mint on the other

The third piece moves USDC from one network to another. Most people assume that means a bridge, which locks your coin on one chain and hands you a wrapped copy on the other. Wrapped copies carry bridge risk. If the bridge is hacked or the locked reserve fails, the copy is worthless.

There is a cleaner path. Circle’s cross-chain transfer protocol burns the USDC on the source chain and mints native USDC on the destination chain. No wrapper. No locked reserve to trust. Canonical USDC on both sides. The flow is four steps: approve the transfer, burn on the source with the destination and recipient specified, wait for a signed attestation that the burn happened, then submit that attestation to mint on the destination.

Two things surprised me in practice. The attestation is not instant. You poll and wait, sometimes minutes, and the mint fails if you submit before the signature is ready. Patience is part of the protocol. And the recipient has to be encoded in a specific format for the burn call, a detail that reverts the whole thing if you skip it.

The takeaway matters beyond the code. Native burn-and-mint is a genuinely better settlement primitive than wrapped bridging. The dollar that arrives is the real dollar, not a claim on a locked one. For anyone thinking about cross-chain treasury movement, that distinction is the whole ballgame.

What building teaches that opining does not

You can read a hundred posts about programmable money and still miss the things that decide whether it works. That USDC needs approval before it moves. That transfers fail silently. That six decimals will humble you. That the attestation delay is a feature, not a bug. That the hard part of an AI agent is not the intelligence but the restraint. These are not opinions. They are the texture of the actual system, and you only feel them when your own transaction reverts.

That is the case for building even when you have no intention of shipping a product. It changes what you can say with authority. It turns a payments take into a payments understanding.

The bottom line

Programmable USDC settlement is real and it is buildable today. An escrow that releases on a condition, an agent that adjudicates under strict guardrails, and cross-chain movement without bridge risk are not future concepts. They run now, on testnets, with the failure modes fully exposed. The technology is ready enough to learn on. The discipline, approvals, decimals, reentrancy, human confirmation, audit trails, is what separates a demo from something you would trust with a real balance sheet.

One rule stands above the rest. Testnet first, always, and keep a human in the loop on anything that moves money. The point of building on test networks is to make every expensive mistake for free.

Everything described here was built and run on test networks with test funds. It is an educational build, not production software, and not financial, legal, or security advice.

The views expressed here are my own and do not represent those of any employer or organization. This is commentary and analysis, not investment, legal, or tax advice.

In late June 2026, a group calling itself the Open Standard Consortium announced Open USD, a dollar stablecoin backed by more than 140 named partners, including Visa, Mastercard, BlackRock, and Coinbase. The pitch is simple and loud: no minting fees, shared reserve yield, and a partner-governed alternative to the single-issuer models of Tether and Circle. The reaction was loud too. Circle’s stock fell sharply on the news. The number everyone repeated was 140. The number that matters is closer to two: how many of those partners are actually committed, and how many regulators have signed off on the economics.

This is not a prediction that OUSD wins or fails. It is a balanced read on what the model gets right and what it has not yet answered.

What OUSD is actually proposing

Strip away the branding and OUSD is an attempt to change who keeps the money a stablecoin generates. A dollar stablecoin holds reserves, usually short-term Treasuries, and those reserves earn interest. In the single-issuer model, the issuer keeps that interest. Tether and Circle built large, profitable businesses on exactly this spread. OUSD proposes to distribute that reserve yield across its member network instead of retaining it at one company, and to let partners mint and redeem without fees.

That is a real structural idea, not just a logo wall. If the economics of a stablecoin are shared with the businesses that distribute it, distribution can scale faster than a single issuer could buy on its own.

The opportunities

The first opportunity is distribution. A stablecoin is only as useful as the places it is accepted and the rails it moves on. A single issuer has to earn each integration. A consortium that includes card networks, a large asset manager, and a major exchange starts with distribution most issuers spend years building. If even a fraction of the named partners integrate, OUSD reaches merchants, wallets, and treasuries quickly.

The second opportunity is economics. Shared yield changes the incentive to adopt. A business that today holds idle dollars for no return has a reason to hold a token that returns part of the reserve yield to it. That is a genuine pull, not just a technical feature.

The third opportunity is the conversation it forces. Once businesses see what their idle balances are actually worth, they start asking every counterparty, including their bank, why that cash earns nothing. That question does not depend on OUSD succeeding. It spreads on its own. It pressures deposit pricing across the system, which is why incumbents are paying attention.

The challenges

The first challenge is governance. Consortiums coordinate poorly. Large groups of large companies have misaligned incentives, move slowly, and struggle to make the hard calls that durable infrastructure requires. The history of bank and payment consortiums is full of projects that had every logo and shipped nothing. 140 partners is a distribution asset and a decision-making liability at the same time.

The second challenge is credibility. According to press reports at the time of the announcement, several named companies, including large Korean institutions, said they had not held formal talks or committed to participate. If those reports are accurate, the founding partner list has outpaced the agreements behind it, and that is a credibility problem before it is a technology problem. It raises a fair question about how much of the announced structure is signed rather than aspirational.

The third challenge is regulatory, and it is the one that decides everything. The 2025 GENIUS Act created a federal framework for payment stablecoins and restricted issuers from paying interest, yield, or rewards to holders. OUSD’s shared-yield model sits directly on that line. Paying partners for services rather than paying holders for holding may satisfy the letter of the rule, or regulators may treat it as the same thing wearing a different label. Open Standard has not publicly disclosed which regulated entity issues OUSD, in which jurisdiction, and under whose supervision. Until it does, the no-fees, shared-yield claim cannot be tested against the rules that determine whether member banks can lawfully receive what is economically interest on a payment instrument.

The fourth challenge is the incumbents’ response, and it cuts two ways. JPMorgan and others have asked regulators to tighten oversight, arguing that yield on wallet balances lets crypto firms compete with bank deposits without the capital and reserve requirements banks carry. That is partly self-interested and partly a real regulatory question. Either way, it will shape the rulemaking that OUSD depends on.

The bank counter is already moving

Banks are not only lobbying. Many are building tokenized deposits, their own version of a programmable dollar that pays interest but keeps the balance inside the regulated bank. A tokenized deposit moves a regulated bank dollar around the clock without creating a new instrument for anyone to hold, without a float to redistribute, and without a coin that has to clear a central bank’s view of settlement. For regulated treasury and interbank use, that model has a cleaner legal story than a consortium coin.

This is the real contest. It is not OUSD versus Tether. It is whether value settles in a shared stablecoin governed by a consortium, or in tokenized bank money governed by the institutions that already hold the deposit. Both can be right in different corridors. Emerging markets and crypto-native commerce may favor the stablecoin. The regulated US institutional system may favor the tokenized deposit.

What to watch

Ignore the logo count. Watch four things instead. First, issuer disclosure: which regulated entity, which jurisdiction, which supervisor. Second, the regulatory read on shared yield: whether payments for services are treated as interest under GENIUS and the follow-on rulemaking. Third, real integrations: not announced partners, but live acceptance and volume. Fourth, the bank response: how fast tokenized deposits reach production and whether they interoperate with anything outside their own walls.

The bottom line

OUSD is a serious idea with an unfinished answer sheet. The opportunity is real: shared economics plus consortium distribution could move adoption faster than any single issuer, and it forces a deposit-pricing conversation the whole industry has avoided. The challenges are also real: consortium governance is hard, the partner list has already cost the project credibility, and the entire economic model depends on a regulatory question that has not been answered in public.

The winners here will not be decided by who has the most logos. They will be decided by who discloses the cleanest legal structure, ships real volume, and survives the rulemaking. Watch the answer sheet, not the announcement.

The views expressed here are my own and do not represent those of any employer or organization. This is commentary and analysis, not investment, legal, or tax advice.

It is 9:40 on a Tuesday. A payments operations team at a large global bank needs EUR 200 million in Frankfurt by 11:00. The bank has dollars. It does not want to source a short euro loan for ninety minutes of use. So the desk uses the FX market as a funding market: exchange dollars for euros now, agree the reverse leg, make the payment, and unwind the exposure inside the day.

That is the practical heart of synthetic FX.

Most people hear “FX” and think currency view: euro up, dollar down, yen weak, sterling strong. That is not how much of bank treasury uses it. In treasury, FX is often funding plumbing. It is a way to put the right currency in the right account at the right time without carrying that currency as cash all day.

Two ideas sit at the center of that plumbing. Synthetic FX is how a bank creates the economics of a currency position without taking a simple outright cash position or standalone local-currency loan. Intraday FX is the discipline of managing that currency funding by the hour, not just at the end of the day. The two are converging because the banking day is losing its clean cutoff.

What “synthetic” really means

Synthetic FX is not magic. It is construction. You assemble the currency exposure, funding result, or settlement effect you need from instruments already available to the desk. The exposure is real. The funding cost is real. The settlement obligation is real. What changes is the route used to create it. Three versions matter most.

1. Synthetic funding through FX swaps

This is the workhorse. A bank has dollars and needs euros. Instead of borrowing euros in the cash market, it enters an FX swap: exchange dollars for euros on the near leg, then reverse the trade at a pre-agreed rate on the far leg. For the life of the swap, the bank has euro liquidity funded by dollars.

The important point is simple: an FX swap is a funding trade wearing FX clothes. The difference between spot and forward, the swap points, is not mainly a prediction about where the euro is going. It reflects the interest-rate differential between the two currencies, adjusted by market frictions. When the implied cost of dollars through the swap market diverges from the cost suggested by covered interest parity, that gap is the cross-currency basis.

That is where many people misread the trade. A desk that books this as “an FX position” may miss the real risk. The exposure often lives less in spot FX and more in funding markets. When dollar funding tightens, or dealer balance sheet becomes scarce, the basis can move even if the spot rate barely changes.

In a frictionless textbook market, arbitrage should close that gap. In the real market, the arbitrage consumes balance sheet, credit capacity, collateral, and regulatory capital. Since post-2008 leverage and balance-sheet rules, the trade that should close the gap is itself costly to hold at scale. That makes the basis more than a currency price. It is often a market price for scarce balance sheet, scarce dollar funding, and constrained intermediation capacity.

This matters most around reporting dates. At quarter-end and year-end, banks can reduce balance-sheet usage. Dealer capacity can shrink. The synthetic dollar or euro funding route that looked cheap during the month can become expensive exactly when everyone needs it. The lesson is blunt: do not plan large synthetic funding needs around average conditions. Plan them around the dates when balance sheet disappears.

2. Synthetic crosses

Not every currency pair trades deep and direct. If a bank needs to move between two less liquid currencies, it may not execute the pair in one clean trade. It may build the exposure through the dollar: currency A into dollars, dollars into currency B. The cross is synthetic. The price is constructed.

That means the real cost is not just the screen spread on the final pair. It is the combined cost of both legs, the dollar liquidity in the middle, timing risk, slippage, and settlement risk on whichever leg is thinner. For an operator, the synthetic cross is not a quote to accept. It is a route to test.

3. Tokenized and stablecoin FX

The newest version uses different rails but the same basic idea: create currency movement, currency exposure, or currency settlement without using the full traditional correspondent banking chain. A dollar stablecoin can be swapped for a euro stablecoin. A tokenized deposit can move value across a permissioned network. A blockchain-based rail can make value available around the clock instead of inside bank operating hours.

The operating promise is not the technology. It is liquidity release. Today, banks and payment firms keep pre-funded balances in correspondent accounts around the world so they can settle locally in each currency. That money is trapped. It earns little, consumes liquidity, and exists mainly because cross-border money movement is slow. Tokenized FX promises to reduce that trapped float. That is the prize. Not a prettier ledger. Not a new wrapper. Less idle liquidity sitting in nostro accounts.

But the catch matters. Stablecoin FX is not clean FX. A dollar stablecoin is not the same thing as a dollar deposit. A euro stablecoin is not the same thing as a euro deposit. Each carries issuer risk, reserve risk, redemption risk, legal risk, and peg risk. A stablecoin pair therefore has its own basis: the gap between the token price and the underlying fiat it claims to represent.

Tokenized deposits are different again. They are bank liabilities in tokenized form. They may be better suited for regulated treasury and interbank settlement, but they still depend on network rules, legal finality, interoperability, and participant balance sheets. The point is not that tokenized rails remove FX risk. They do not. They change where the risk sits.

Synthetic dollars are debt that does not look like debt

Step back from the desk and the system-level issue becomes larger. Every synthetic dollar raised through an FX swap or forward creates a future dollar payment obligation. Economically, that looks like dollar borrowing. Accounting-wise, it does not show up like a normal dollar loan.

That is why the synthetic-dollar market matters. Large amounts of dollar funding sit in FX swaps, forwards, and currency swaps rather than visible on-balance-sheet debt. The obligation is real, but it is not reported in the same place as ordinary debt. That creates two problems.

First, rollover risk is hidden in plain sight. Much of this funding is short-dated. It works only if the market can keep rolling. In calm markets, that looks routine. In a dollar squeeze, everyone tries to roll at once. The basis moves. The cost of synthetic dollars jumps. The trade that looked like plumbing becomes the stress point. March 2020 showed this clearly. The problem was not just that currencies moved. The deeper problem was that dollar funding through FX swaps became more expensive and less reliable when global institutions needed dollars at the same time.

Second, the backstop may sit outside the borrower’s home jurisdiction. When private dollar funding breaks, the Federal Reserve can provide dollars to other central banks through swap lines. Those central banks can then lend dollars into their local banking systems. That matters because a non-US bank’s synthetic-dollar book can depend, in extremis, on access to a dollar facility controlled by another jurisdiction. That is not a reason to avoid the trade. It is a reason to understand the dependency before stress arrives.

Intraday FX: the day loses its cutoff

For a long time, the treasury day had a shape. Payments moved. Trades accumulated. Positions were monitored. Books were squared. End-of-day snapshots mattered because the end of the day mattered. That rhythm is breaking.

Real-time payment systems, instant settlement expectations, longer operating windows, and tokenized rails all push banks toward more continuous liquidity management. The issue is not that every system becomes gross or that netting disappears. Netting still matters. CLS still matters. RTGS systems already exist. The issue is that more flows now require usable liquidity at specific moments inside the day.

That changes the job. A bank can be flat at end-of-day and still be deeply short a currency at 11:00 a.m. A desk can show little overnight exposure but still run major intraday peaks. End-of-day risk reports can miss the very liquidity pressure that determined whether payments were made on time.

This is why intraday liquidity regulation exists. BCBS 248 was built around the idea that banks need to monitor intraday liquidity positions, payment flows, settlement obligations, available collateral, and stress scenarios. The regulator’s point is obvious: a bank that survives the close can still fail during the day. Intraday liquidity is not free. It uses collateral. It uses credit lines. It uses balance sheet. It creates operational risk. It creates client risk if payments are delayed. It creates reputational risk if a bank cannot move money when it said it could.

That is where intraday FX becomes important. The intraday FX swap is a funding instrument measured in hours rather than days. A bank can borrow one currency against another inside the same business day, then unwind when the liquidity need passes. The market is still developing, but the direction is clear: liquidity is being priced below the overnight tenor.

That is a big change. It means the old curve is no longer enough. If money can move continuously, the market needs a price for liquidity at 10:00, 12:00, 14:00, and 16:00. An intraday funding curve starts to form under the floor of the traditional overnight market. Synthetic FX is the toolset. Intraday FX is the operating discipline.

Where the two meet

The relationship is clean. Synthetic FX gives treasury the instruments to create currency funding without carrying every currency all day. Intraday FX tells treasury when, how much, and for how long that funding is needed.

Return to the Tuesday example. At 9:40, the bank needs euros for an 11:00 payment. It funds the position synthetically through an FX swap. At 13:00, another flow leaves the bank short dollars for two hours. It creates a short-dated synthetic dollar position against the liquidity it holds elsewhere. By the close, the major legs are unwound. The end-of-day book looks calm.

But the day was not calm. It was a sequence of temporary funding needs, temporary synthetic positions, settlement deadlines, collateral constraints, and pricing decisions. That is modern treasury. Not one balance at the end of the day. A moving liquidity problem by currency, account, rail, legal entity, and time zone.

The upside: treasury can become a P&L line

Most firms treat intraday liquidity as a cost. That is incomplete. A bank with spare balance sheet, strong dollar access, and good intraday systems can lend liquidity synthetically to banks that need it. The same FX swap that funds your own payment can be reversed to monetize someone else’s shortage.

This is not free money. It consumes limits, collateral, operational capacity, and balance sheet. But the commercial logic is clear. When liquidity becomes scarce by the hour, the institution that can see it, price it, and move it fastest has something to sell. That is the shift. Treasury is no longer only defending the balance sheet. In the right institution, treasury can monetize the balance sheet.

What operators should watch

Synthetic does not mean free. A synthetic position still consumes credit lines, collateral, settlement capacity, operational bandwidth, and balance sheet. Treat it as leverage from nowhere and the limit call will arrive before the theory does.

The basis is not just an FX number. It is often a funding signal. It can reflect dollar scarcity, dealer balance-sheet constraints, collateral terms, reporting-date pressure, and the cost of intermediation. If the basis moves, the spot chart may not explain why.

Reporting dates matter. Quarter-end and year-end can change dealer capacity and synthetic funding costs. Calendar risk is real. Build it into the plan before the market charges you for forgetting it.

Settlement risk has not disappeared. PvP settlement reduces the core Herstatt problem, but not every currency, counterparty, tokenized rail, or settlement window sits inside robust PvP protection. Outside that protection, one leg can still move before the other. The newer the rail, the more important the legal and operating details become.

Stablecoin FX adds a basis. A dollar token and a euro token do not only express USD/EUR. They also express confidence in issuers, reserves, redemption, market depth, and legal enforceability. A tokenized FX price is not just an FX price.

Automation is no longer optional. Treasury managed by the hour is too fast for manual ticketing. Routine liquidity movements will need rules, systems, limits, alerts, and automated execution. Humans will still manage judgment, exceptions, stress, and client priority. But the operating model has to match the speed of the instruments.

The day used to have an end. Positions squared, books closed, reports printed, and everyone went home. For a bank moving currency across real-time payment systems, global clients, longer settlement windows, and tokenized rails, the day no longer ends cleanly. Synthetic FX is how you fund that day. Intraday FX is how you survive it.

The views expressed here are my own and do not represent those of any employer or organization. This is commentary and analysis, not investment, legal, or tax advice.

Settlement lag is not a law of physics.

Some delay is useful. Netting reduces liquidity demand. Cutoffs create control points. Batch windows give operations teams time to catch errors. Legacy rails were built around those constraints for a reason.

But the delay is not free. Every extra day between execution and final settlement creates cost. It ties up liquidity. It creates in-flight exposure. It forces reconciliation. It turns settlement into a credit event, even when nobody calls it one.

That is the real case for atomic settlement. Not hype. Not “blockchain fixes payments.” Not a cheaper wire. A balance sheet problem.

What atomic settlement actually means

Atomic settlement means both legs of a transaction settle together, or neither settles.

In securities, that is delivery-versus-payment. The asset and the cash move together. In FX, that is payment-versus-payment. One currency does not settle unless the other does too. In commercial flows, it means funds release only when the agreed condition is met: delivery confirmed, invoice matched, asset transferred, or entitlement recognized.

Three things matter. Simultaneity: both legs settle as one event. Finality: settlement is legally recognized and cannot be unwound through the ordinary payment process. Conditionality: the transaction executes only when the agreed condition is satisfied. If those three conditions are not true, do not call it atomic settlement.

The technology matters less than the guarantee. Stablecoins, tokenized deposits, wholesale central bank money, and shared ledgers can all support atomic models. But they are not the same. The issuer, settlement asset, redemption right, legal finality, and rulebook matter. “Atomic” describes the settlement logic. It does not eliminate credit, legal, liquidity, or operational risk by itself.

Why firms still use slower rails

The reason is not stupidity. The old rail wins because the cost of switching is obvious and the cost of staying is buried.

Integration costs are visible. Legal review is visible. New controls are visible. Training is visible. New vendor risk is visible. The cost of doing nothing is spread across treasury, operations, risk, capital, and client service. Nobody sees the full bill unless someone adds it up.

That bill has three main parts. Liquidity drag: cash and collateral sit in buffers because the rail cannot guarantee both legs at the same time. Settlement exposure: a firm may pay before it receives, which creates principal risk, counterparty exposure, and capital cost. Operational drag: failed trades, payment investigations, reconciliation breaks, exception queues, and manual repair all exist because expected settlement and actual settlement drift apart.

Staying put is not free. It is a recurring tax.

What atomic settlement buys

The first benefit is principal-risk reduction. If both legs settle together, one party does not pay while waiting for the other side to perform. In FX, true PvP atomic settlement can reduce Herstatt-style risk because neither currency leg completes unless the other completes too.

The second benefit is faster liquidity reuse. Atomic settlement does not magically make liquidity needs disappear. In some cases, it increases intraday funding pressure because transactions settle gross and in real time. But once a transaction settles, the funds are final and reusable. The same unit of liquidity can turn faster. The gain is not “less money needed” in every case. The gain is higher velocity, cleaner certainty, and less trapped buffer.

The third benefit is cleaner operations. A shared settlement record reduces the number of places where records can diverge. It does not remove all exceptions, but it removes many of the conditions that create them.

The fourth benefit is real-time position visibility. If settlement is final and visible as it happens, treasury does not have to wait for end-of-day files to know where cash and collateral sit.

The fifth benefit is programmability. Conditions can move into the settlement flow itself. Release on delivery. Release on match. Release on asset transfer. Release on rule satisfaction. That changes the control model. You stop sending money first and proving later that it should have gone.

The part most pitches skip

Atomic settlement changes the risk profile. It does not remove risk. It reduces settlement risk and some credit exposure. It can reduce trapped liquidity. It can reduce manual reconciliation. But it creates or increases other risks.

Liquidity risk: you need the right funds in the right place at the exact time of settlement. Operational risk: controls must run before settlement, not after. Legal risk: finality must be recognized in the relevant jurisdiction and rulebook. Oracle risk: if settlement depends on external data, the data source becomes part of the risk model. Error risk: if settlement is final, mistakes move faster too.

This is the trade. You exchange delayed uncertainty for immediate precision. That is usually better, but only if the control stack is built for it.

Gross settlement creates a real constraint

The old system often relies on netting. Many transactions collapse into one smaller settlement obligation. That is efficient for liquidity. Atomic settlement often moves closer to real-time gross settlement. Every leg may need to be funded in full at the moment it settles. That can increase intraday liquidity demand.

So the serious question is not: “Does atomic settlement reduce liquidity?” The serious question is: “Does faster liquidity reuse offset the higher funding precision required?” Sometimes yes. Sometimes no. It depends on flow shape, concentration, timing, currency, counterparty mix, and available liquidity-saving tools.

Any serious atomic settlement design needs queuing, prioritization, liquidity-saving mechanisms, partial netting, and gridlock resolution. If the rail has no answer for those, it is not ready for institutional scale.

Settlement lag is a hidden credit line

When one party pays today and receives in two days, that party has extended credit for two days. It may not be priced that way. It may not be called a loan. But economically, that is what it is.

Remove the lag and the hidden credit line disappears. For weak counterparties, that is pure risk reduction. For strong counterparties, it may expose an uncomfortable truth: some relationships were subsidized by settlement delay. Once the delay is removed, the credit has to be explicit and priced. That is why faster settlement can feel more expensive to some participants. It removes a hidden subsidy.

Why incumbents resist it

Incumbents do not resist atomic settlement only because of technology. They resist because settlement lag supports existing economics. Some revenue sits in float. Some sits in exception handling. Some sits in correspondent chains. Some sits in balance-sheet spread. Some sits in internal P&L that nobody wants to shrink.

That does not make incumbents irrational. It makes them economically rational. A new rail project can pass the technology test and still die because it threatens a revenue line inside the institution sponsoring it. Name that early. Otherwise the project will be framed as an implementation issue when it is really a business-model issue.

The cold-start problem

A settlement network is only valuable if others use it. That creates a brutal adoption problem. Early users pay the highest switching cost when the network has the lowest utility.

The way through is closed-loop first. Start where you control both sides of the transaction. Internal treasury movements. Intercompany funding. Internal FX. Collateral movements between related entities. Controlled client corridors. Repetitive flows with known counterparties.

Closed-loop flows let you test legal finality, funding rules, controls, exception handling, liquidity demand, and operational governance before asking the market to join. Do not start with the hardest open-network problem. Start where the economics are visible and the adoption dependency is low.

A simple corridor example

Take a USD/EUR corridor that moves $100 million a day and settles on T+2. Under the old model, two days of value may sit in flight. At steady state, that can mean up to $200 million of settlement exposure across the window. Atomic PvP settlement can reduce that exposure sharply because both currency legs settle together.

The liquidity answer is more nuanced. If the flow can be supported with a $20 million intraday buffer that turns repeatedly through the day, then the firm may free a large amount of liquidity relative to the old model. At a 4 percent cost of funds, every $100 million of reduced required funding capacity is worth roughly $4 million per year. But the actual number depends on peak intraday demand, not average daily volume.

That is the analysis that matters. Model the real flow. Measure the peak. Add the buffer. Price the funding. Then compare it with today’s settlement exposure, capital cost, and operational expense.

The business case usually comes in this order: principal-risk reduction, then capital and exposure reduction, then liquidity velocity, then operational savings, then new transaction types made possible by all-or-nothing settlement. The last point matters most. The real prize is not a cheaper payment. It is transactions that become possible because no party has to sit exposed between legs.operational savings, then new transaction types made possible by all-or-nothing settlement. The last point matters most. The real prize is not a cheaper payment. It is transactions that become possible because no party has to sit exposed between legs.

The question that flips the sale

The weak pitch says: believe in a new rail. The stronger pitch says: measure what the old rail costs.

How much liquidity sits idle because settlement is delayed? How much exposure exists between payment and receipt? How much capital supports that exposure? How many people manage breaks created by timing gaps? How much revenue depends on delay? How much client value would exist if settlement were final, conditional, and real time?

Atomic settlement is not magic. It is not live everywhere. It does not eliminate risk. It changes the risk from counterparty timing risk to liquidity, control, legal, and data risk. That trade is still worth taking seriously.

Settlement lag was never free. It was just familiar.

The views expressed here are my own and do not represent those of any employer or organization. This is commentary and analysis, not investment, legal, or tax advice.

The demos work.

That is why the hard problem is no longer the demo.

JPMorgan, Citi, Fnality, Partior, and other institutional players have shown that regulated money, liquidity, and settlement instructions can move faster, more continuously, and with more programmability than legacy rails allow.

That matters.

But most bank-led tokenized deposit models still prove something narrower than an ecosystem. They prove controlled movement inside a known perimeter. Money moves from one account to another account inside the same institution, or inside a tightly governed network where the operating model is still highly contained.

That is useful. It is not yet a rail.

A tokenized deposit does not become strategically interesting because one bank can move its own liabilities faster. It becomes interesting when a token issued by one bank can settle against a token issued by another bank under governance that both banks, their clients, their regulators, and their operating systems can trust.

The ecosystem starts at counterparty #2.

That is where every hard question appears. Whose ledger is canonical? Whose compliance check binds? Who owns the loss if one leg fails? What law governs finality? How is liquidity funded? Who resolves disputes? Who can reverse, freeze, or reject a transaction? Who has operational control at 2:13 a.m. on a Sunday when something breaks?

A smart contract does not answer those questions. Governance answers them.

That is why tokenized deposits are not mainly a ledger problem. They are an ecosystem problem.

Money ecosystems have different physics

Software ecosystems and money ecosystems look similar from a distance. They are not the same. Three rules matter.

1. Balance sheets are conserved

A software platform can subsidize growth with cheap digital supply. A money network cannot.

Every tokenized deposit is someone’s liability. Every unit in circulation sits on a balance sheet. Every early-network subsidy has a capital cost, a liquidity cost, or a risk cost.

You cannot growth-hack settlement money the way a consumer platform growth-hacks users. That is the first constraint most fintech writing misses.

Tokenized-deposit adoption is rate-limited by who is willing to lend their balance sheet to the early network before the network is valuable. That is not a marketing problem. It is a capital problem.

2. Networks have two stable states

Most network-effect businesses have two natural states. Dead: nobody joins because nobody else is there. Alive: everyone joins because everyone else is there. The hard part is crossing the gap between the two.

In payments, that gap is unusually wide. A failed software platform leaves abandoned code. A failed money rail strands liquidity, burns trust, and leaves regulatory scar tissue. The next attempt becomes harder because the failed attempt becomes evidence.

That is why pilots are easy and ecosystems are hard. The pilot proves the mechanism. The ecosystem requires enough credible participants to make the mechanism worth adopting.

3. Trust compounds slowly and collapses quickly

Most platforms can survive a bad user experience. Money systems cannot survive many bad settlement experiences.

One failed leg, one frozen token, one unclear loss allocation, or one weekend operational incident can cause real-volume operators to leave and not come back.

This is the rational core inside what outsiders call bank conservatism. Settlement systems are not rewarded for looking innovative. They are rewarded for not creating unexplained loss. The governance has to be designed for the tail event, not the average day.

Cold start is a governance problem

The hardest participant to recruit is not the first user. It is the second bank.

The first bank builds the rail because it controls both ends. It captures the internal efficiency. It can justify the build.

The second bank faces a different decision. It is asked to connect to another institution’s infrastructure, accept shared rules, expose its clients to a new operating model, and absorb integration cost before the network has full value. The benefits only become compelling once the third, fourth, and fifth banks are also there. So everyone waits.

This is the penguin problem. The penguins crowd the edge of the ice. Everyone is hungry. Nobody wants to jump first because the first mover takes the risk. A protocol does not solve that. A credible governance model does.

The early network needs someone willing to fund the activation energy: onboard participants, seed liquidity, absorb coordination cost, define loss rules, create operating standards, and accept worse economics until the network exists.

That is why shared-infrastructure models matter. Bank consortia, regulated-liability networks, shared settlement platforms, and central-bank-money settlement experiments are not just technical projects. They are attempts to distribute the cold-start cost so no single participant has to jump alone. Whether they clear the hump remains open. But they are aimed at the right problem.

The treasurer does not care until counterparty #2 exists

Corporate treasurers will not re-plumb workflows for a rail that reaches one bank. They already manage multiple banking relationships. They already operate through treasury workstations, ERPs, payment hubs, and bank portals. They already live with complexity.

A single-bank tokenized deposit may improve liquidity movement inside that bank. That is useful. But it does not create a reason to change the operating model.

The treasurer cares when the rail solves a real cross-bank problem:

• trapped liquidity

• settlement timing

• failed payment uncertainty

• collateral movement

• weekend funding gaps

• delivery-versus-payment risk

• payment-versus-payment risk

• cash visibility across institutions

• programmable controls across counterparties

That requires counterparty #2. Without it, tokenized deposits are an internal efficiency tool. With it, they can start becoming infrastructure.

Keystone, dominator, or utility

The strategic question for banks is not whether they can issue tokenized deposits. It is what role they want to play in the ecosystem. There are three basic postures.

A keystone builds shared infrastructure, lowers the cost of participation, and lets others create value on top. A dominator tries to own the loop, control the participants, and internalize the economics. A utility supplies settlement capacity to whoever owns the customer relationship.

Banks naturally default toward dominator. That instinct is understandable. Banks are regulated, risk-managed institutions. They want control. They want known counterparties. They want bounded exposure.

But dominator logic creates closed loops. Closed loops create limited networks. Limited networks become supplier infrastructure for someone else.

The keystone strategy is harder but more powerful. It requires the bank to make the ecosystem easier for others to join, even when that means sharing some economics.

The most underrated keystone move is not branding. It is compliance infrastructure. KYC, AML, sanctions screening, transaction monitoring, legal-entity identity, data-sharing rules, wallet controls, and permissioning are duplicated across institutions. Every participant rebuilds the same machinery. Every onboarding process creates friction.

An actor that creates reusable compliance primitives for the network reduces everyone’s cost of joining. That actor becomes structurally central without owning every transaction. Compliance-as-infrastructure sounds like a back-office function. It is not. It is an ecosystem wedge.

Multihoming caps the rail moat

Here is the uncomfortable law for any bank issuing a token. Operators multihome.

A corporate treasurer already works across multiple banks. A payments platform already routes across processors. A treasury workstation already abstracts banking relationships behind one interface. A payment hub already thinks in rules, limits, cutoffs, fees, currencies, and counterparties.

Once a new rail sits behind the same interface, the operator does not pledge allegiance to the rail. The operator routes. That does not mean integration is free. It is not. But once routing is abstracted, adding one more rail becomes a marginal decision, not a strategic commitment.

That caps the value of being only the rail. It increases the value of being the routing layer. The bank funds the infrastructure. The orchestration layer captures the choice. That is the value-capture problem banks need to face directly.

The durable value is in the connective layer

If multihoming caps the rail, where does durable value live? In the connections.

The important layer is not only the token. It is the interoperability layer around the token:

• common messaging

• ISO 20022 data alignment

• shared identity

• compliance primitives

• liquidity standards

• settlement rules

• exception handling

• dispute resolution

• ERP and TMS integration

• controlled programmability

Standards are not glamorous. They are where power accumulates. Whoever defines the connective layer shapes how everyone else plugs in. Whoever controls the operating standard becomes harder to route around than whoever issues the largest token.

The same is true for composability. Operators do not buy “instant settlement” in the abstract. They buy workflows that work better:

• release payment only when goods arrive

• move collateral when a margin call hits

• settle FX legs atomically

• sweep idle balances under policy

• enforce approval rules automatically

• reconcile without manual repair

• reduce trapped cash

• shorten funding gaps

That is the product. The token is the mechanism. Banks tend to protect the mechanism. Operators pay for the product.

The value creator may not be the value capturer

This is the core strategic inversion. The bank that builds the rail may create the value. The platform that owns the operator relationship may capture it.

Treasury workstations, ERPs, payment hubs, PSPs, and orchestration platforms sit closer to demand. They decide which rail gets used, when, and why. That makes them powerful. The issuing bank becomes one settlement option among many unless it also controls the operating layer around the client.

This is not theoretical. It is how payments already work. The merchant does not manage every acquiring rail directly. The merchant uses an orchestrator. The treasurer does not manually optimize every banking channel. The treasurer uses a TMS, ERP, bank portal, or payment hub.

The more rails proliferate, the more valuable abstraction becomes. That means banks face a sharp choice. They can build keystone infrastructure: open enough to attract participants, useful enough to create standalone value, governed well enough to earn trust, and integrated deeply enough to matter in daily operations. Or they can defend closed loops and become regulated suppliers to whoever owns the workflow. There is not much comfortable ground between those two positions.

A short playbook for corporate treasurers

Do not integrate a rail because it is new. Integrate it only if it solves an operating problem today. Ask:

• Does it work with more than one counterparty?

• Does it improve liquidity availability?

• Does it reduce failed-payment risk?

• Does it integrate with my existing TMS or ERP?

• Does it produce better data?

• Does it improve controls?

• Does it reduce manual repair?

• Does it give me routability, or does it create another closed dependency?

Instant settlement is not enough. Routable, controlled, programmable settlement is the test.

For bank product owners

Decide early whether you are building a keystone or a utility. The architecture follows the decision.

If you want to be a keystone, spend your first subsidy dollar on counterparty #2. Spend the second on shared compliance and operating standards. Spend the third on integration into the systems clients already use. Do not over-invest in the token and under-invest in the ecosystem. The token gets attention. The ecosystem gets adoption.

For platforms and PSPs

Stay rail-agnostic. Your advantage is abstraction. Every new rail should become another routing option behind your interface. Your moat is not one rail. Your moat is the decision layer that compares rails, applies rules, manages risk, and optimizes economics. Banks may build the expensive settlement infrastructure. The platform that controls routing controls demand.

What to watch

Ignore most pilot announcements. Pilots prove that the mechanism can work. They do not prove that the ecosystem will form. Watch for five things instead.

First, counterparty #2. The meaningful signal is not a bank moving its own liabilities faster. The meaningful signal is settlement between genuinely separate institutions under shared governance.

Second, loss allocation. Who bears the loss if one leg fails, a token freezes, a participant defaults, or an operational error occurs? If that answer is vague, the network is not ready for real volume.

Third, compliance reuse. If every participant must rebuild every control from scratch, adoption will be slow. If compliance primitives become reusable, the ecosystem can scale.

Fourth, interoperability standards. The durable value is in the connective layer: messaging, identity, data, liquidity rules, settlement finality, and exception handling.

Fifth, native workflow integration. A rail that sits outside the treasurer’s actual operating environment is a demo. A rail embedded into the TMS, ERP, bank portal, or payment hub becomes usable infrastructure.

The bottom line

The technology is ready enough to expose the real problem. The real problem is not whether tokenized deposits can move. They can. The real problem is whether they can move between institutions under governance that operators trust, regulators accept, and platforms can route.

That starts with counterparty #2. Without counterparty #2, tokenized deposits remain a faster closed loop. With counterparty #2, they can start becoming a rail.

Build for the second bank, or build a very fast database and call it a revolution.

Banks did not go on-chain for speed. They went to keep commercial bank money behaving like one dollar.

A dollar in a bank account looks simple because the system works hard to hide how strange it is.

To a customer, a dollar is a dollar. A Chase dollar, a Citi dollar, a Wells Fargo dollar, and a BNY dollar all spend the same way, and nobody checks which bank’s balance sheet minted the balance before accepting it. That sameness feels like a law of nature. It is closer to an engineering achievement.

It comes from deposit law, clearing rules, Fed settlement, supervision, liquidity requirements, and decades of plumbing built so that thousands of separate private bank liabilities behave like one public currency. Economists call the result the singleness of money. It is the quiet miracle of commercial bank money, which is where most households and businesses actually experience the dollar.

Then banks started putting deposits on-chain, and the miracle stopped being automatic.

That sounds like a routine technology upgrade. It is not. Once a deposit becomes a token, the old guarantee loosens. A tokenized JPMorgan deposit is a JPMorgan liability. A tokenized Citi deposit is a Citi liability. A tokenized HSBC deposit is an HSBC liability. Each one can live on a different ledger, follow different rules, and sit inside a different walled garden. The token says dollar. The system still has to prove it is money.

That is what the June 5 announcement was actually about. A group of large banks, including JPMorgan, Bank of America, Citi, Wells Fargo, BNY, PNC, TD, Truist, U.S. Bank, and HSBC, backed a shared tokenized deposit network to be operated by The Clearing House, with a target launch in the first half of 2027. The press release used the usual vocabulary: programmability, 24/7 settlement, interoperability, richer data.

Read it the other way. The banks had begun building branded versions of on-chain commercial bank money, which threatened to fragment the dollar into bank-specific pieces, and they needed a common utility to make those pieces clear against each other at par again. The title is only half a joke. They did not break par in the live system. They started breaking the on-chain dollar into branded fragments, then moved to reassemble it before the fragmentation could harden.

This is not a payments-speed story. It is a singleness-of-money story.

The dollar is not the token

A tokenized deposit gets described as a bank deposit on a blockchain. That is accurate and almost useless, because it points at the wrong object.

The token is not the money. The money is the legal claim on the bank, the right to redeem at par, the settlement rule that stands behind it, and the confidence that another institution will treat it as the same dollar when it moves. Crypto learned this the hard way. USDC is spoken of as one stablecoin, but it lives across many chains, and moving it still relies on bridges, mint-and-burn mechanics, liquidity, and trust assumptions. The user sees a dollar. The infrastructure sees fragments held together by effort.

Banks walked into the same trap. JPMorgan issues JPMD. Citi runs Citi Token Services. HSBC runs a Tokenized Deposit Service. Each works cleanly inside its own bank, and that is exactly the problem, because corporate treasury does not stop at one bank. A real company runs multiple banking relationships, multiple legal entities, multiple jurisdictions, and multiple liquidity pools. A token that only settles cleanly inside one institution is not money at scale. It is a closed-loop product, however well built.

So the interesting thing TCH is building is not a blockchain. A blockchain vendor has not even been selected yet. The thing being built is a rulebook and a settlement bridge: common rules and a connectivity layer that lets one bank’s tokenized deposit clear against another’s without a bespoke integration every time. The chain is the part everyone can see. The rulebook is the part that does the work.

Par has always come from the Fed, not the banks

Here is the point most coverage skips, and it is the one that makes the rest make sense.

The reason a Chase dollar equals a Citi dollar today is not goodwill between banks. It is that both ultimately settle in central bank reserves. When real value moves between banks, it nets down to a transfer of Fed balances, and central bank money is the asset that no commercial bank can debase. Par is manufactured upstream, at the Fed, and the banks inherit it.

That is why TCH is the operator and not a fintech. The Clearing House already runs CHIPS and RTP, the rails that clear and settle interbank dollars, with CHIPS alone moving roughly two trillion dollars a day. The new network is explicitly designed to plug tokenized deposits back into those rails. In other words, the banks are not inventing par. They are extending the existing par mechanism, central bank settlement, onto new ledgers, and wrapping it in a tokenized front end.

If you want to know whether a tokenized deposit is really money, ask one question. Does it settle in central bank money, or does it settle on its own chain and merely promise to be worth a dollar? Everything else is interface.

Banks did not do this because payments were too slow

The lazy explanation is that banks want faster payments. It is not wrong. It is just shallow.

The US already has RTP, FedNow, CHIPS, and ACH. Corporates are not lying awake begging for a fifth rail with a nicer logo. The real driver is balance-sheet defense, and the threat is not the stablecoin payment, it is the stablecoin balance.

A dollar that sits in an operating account funds the bank. The bank lends against it, anchors the treasury relationship around it, and cross-sells liquidity, payables, receivables, FX, working capital, cards, custody, and cash management on top of it. A dollar that moves into a stablecoin wallet stops doing all of that for that bank. The systemic version of this is softer than the slogans suggest, because stablecoin reserves mostly sit in Treasury bills and in deposits at a handful of banks, so the money does not leave the banking system so much as concentrate and change tenor. But the competitive version is brutal and specific. The deposit leaves your bank, even if it lands at someone else’s.

So the question underneath the whole initiative is not who moves a dollar fastest. It is who owns the operating account when dollars become programmable. Tokenized deposits let a bank tell a corporate client: you can have 24/7 movement, programmable settlement, and richer data without leaving the regulated deposit system. That is not a product pitch. It is a funding strategy wearing a product’s clothes.

The yield advantage is real and oversold

There is a legal edge hiding in the GENIUS Act. Payment stablecoin issuers are barred from paying yield to holders. A bank tokenizing a deposit faces no such bar, because a deposit can pay interest, and JPMD is already structured to be able to bear interest. On paper, banks get to offer programmable dollars that also pay, and issuers do not.

Do not lean too hard on it. Corporate operating cash does not mainly choose between a stablecoin and a deposit. It chooses between a deposit and a money market fund, and money funds pay yield while banks pay close to nothing on operating balances. So the honest version of the moat is not yield. It is legal finality, counterparty treatment, auditability, sanctions screening, ERP integration, and the fact that the instrument already fits inside relationships the treasurer maintains. The yield asymmetry helps at the margin and gives banks a reason to keep lobbying to close the wallet-rewards loophole. It is not the reason a treasurer picks a bank token.

And the stablecoin still owns a flank the banks cannot easily take. USDC and USDT are embedded across exchanges, wallets, fintech apps, and emerging-market corridors. Bank tokens are credible where institutional trust and compliance decide the deal. Stablecoins are usable across the open internet. The market will not crown one winner. It will split by use case, and both sides already know which half they hold.

The chain choice tells you what banks actually want

JPMorgan’s behavior is the clearest read on intent. JPMorgan has moved JPMD onto Coinbase’s Base network for institutional clients and has also announced phased native issuance on Canton, a network built for institutional finance with privacy controls. That pairing is not hedging. It is a statement that banks want two things at once: public-chain distribution and institutional privacy.

Crypto treats transparency as a virtue. Corporate treasury treats it as a liability. Settlement flows leak working-capital pressure, supplier concentration, acquisition activity, and liquidity stress, and no CFO wants competitors reading that off a public explorer. The institutional ideal is shared and programmable infrastructure that stays private where it has to. So the banks are not going crypto-native in any ideological sense. They are taking the parts they need, distribution and programmability, and wrapping them in permissioning, privacy, and compliance. They do not want transparent money. They want programmable money with controlled visibility.

That phrase matters more than it looks, because controlled visibility is the actual product.

The real product is not the payment, it is the visibility

Most coverage treats tokenized deposits as a payment instrument. Payments are the door, not the room.

The prize is the programmable operating account, and the moat around it is who gets to see the workflow. Once a client holds tokenized deposits, the bank can wrap logic around the balance: automated liquidity sweeps, programmable escrow, conditional supplier payments, delivery-versus-payment for securities, intraday credit, FX triggers, and audit-ready data trails. A payment fee is thin. An operating account is strategic. A programmable operating account that the bank can observe in full is the strategic asset, because the institution that sees the cash flows can price credit, FX, fraud, and liquidity risk better than anyone who cannot. That is why “richer data” is not a feature bullet. It is the foundation of an information advantage that compounds over time. The bank does not just move the money. It reads the business around the money, and that asymmetry is the moat.

What it quietly costs the banks

This is not a free upgrade for the people building it, which is worth saying out loud because the press releases will not.

Round-the-clock programmable settlement compresses float. Banks earn real money today on the lag between instruction and final settlement, and on intraday balances that sit still. Instant settlement shrinks both. So the banks are partly cannibalizing an existing revenue line to protect a larger one, the operating relationship, against stablecoins that would take the whole thing. That is a rational trade. It is also a tell that this is defense, not opportunism.

There is a second, deeper cost they cannot engineer away. Control and composability pull in opposite directions. Stablecoins are powerful because they plug into the entire open financial stack, permissionlessly. The more a bank walls its token for compliance, privacy, and supervision, the safer it is and the less composable it becomes, which weakens the network effect that makes money useful. Banks cannot have maximum control and maximum reach at the same time. Where they land on that curve will decide how much of the market they actually win, and no rulebook resolves the tension. It only chooses a point on it.

Why now, when nobody is demanding it

Bank of America’s head of global payments said clients are not “beating down the door” for tokenized deposits, and the candor is the most useful line in the whole announcement. So why build a shared network for a product with no line out the door?

Because the build is an option, not a response to current demand. It is an option on a future where 24/7 markets, tokenized securities settlement, and machine-driven commerce make par-stable programmable bank money load-bearing rather than experimental. The clearest version of that future is agentic. AI agents will increasingly initiate payments, and agents need money with hard, native guardrails: spend limits, identity inheritance, reversibility, and policy enforced at the ledger. Stablecoins do not inherit bank-grade controls by default. A tokenized deposit can carry identity, limits, permissions, and compliance policy inside the banking relationship from the start. If autonomous commerce becomes real, the rail that wins is the one that lets an agent move value fast and lets a compliance officer sleep. Banks are building ahead of that demand on purpose, because infrastructure that takes years to stand up has to exist before the wave, not after.

What can still go wrong

Adoption is the first risk, and the BofA quote is the proof. Treasury teams switch rails when the workflow is clearly better, the accounting is clean, and the integration is worth it, not when banks issue a press release.

Fragmentation is the second. A US bank network is not global interoperability, and a shared rulebook does not automatically reconcile every chain, jurisdiction, and currency.

Liquidity is the third. If pools form around individual bank tokens rather than the network, the system recreates the very fragmentation it was built to cure.

Stress is the fourth, and it is the one that actually matters. Par is trivial on a calm Tuesday. The test is whether one bank’s tokenized deposit still redeems at par against another’s when that bank is under pressure, liquidity is tight, and clients want out. Par survives stress in the bank system because deposits sit inside a prudential perimeter: supervision, liquidity rules, access to central bank settlement, and FDIC insurance where applicable, so the unavoidable design question is whether the token sits inside that perimeter. If the tokenized deposit represents an insured claim and settles in central bank money, par holds when it counts. If it does not, par is a market convention, and conventions are the first thing to break under stress.

Indifference is the fifth. Banks may care about defending deposits more than clients care about tokenizing them, and defensive infrastructure only becomes important if it solves a client problem rather than a balance-sheet anxiety.

What to watch

Watch settlement, not speed. If tokenized deposits remain tied to established bank settlement rails and ultimately to central bank settlement, the banks have a credible path to rebuilding par on new ledgers. If they settle on a private chain that merely promises a dollar, they have not.

Watch the operating account. If these tokens show up inside treasury portals, ERP integrations, and liquidity dashboards, they are serious. If they stay demos for innovation teams, they are theater.

Watch the perimeter. Insurance status and central-bank settlement are the two things that decide whether par survives a bad day.

Watch privacy. If institutions keep choosing privacy-enabled networks over transparent ones, that confirms corporate money needs auditability without public exposure.

Watch the corridors. Inside the regulated US system, bank tokens have the stronger institutional case. In emerging markets and crypto-native commerce, stablecoins still have the stronger distribution case.

The one-line version

Banks did not put deposits on-chain because the world needed a faster rail. They did it because stablecoins threatened to pull balances out of their accounts, and because each bank’s private answer to that threat created a new problem: too many branded dollars.

The Clearing House initiative is the move to fix that before it sets. The banks fragmented the on-chain dollar into bank-specific pieces, and now they need the rulebook, and the central bank settlement behind it, that makes the pieces behave like one dollar again.

The rulebook is the product. Par is the test.

The views expressed here are my own and do not represent those of any employer or organization. This is commentary and analysis, not investment, legal, or tax advice.

It is tempting to read Antom’s Card Revenue Booster as the moment AI in payments turned commercial. It is not. Adyen marketed revenue optimization in 2016. Checkout.com shipped AI-driven Intelligent Acceptance in 2023. Stripe sells Authorization Boost as a revenue product today. Machine-learning routing, retries, and false-decline recovery have been sold as growth, not just protection, for years.

So Antom is not the start of a category. It is evidence that previously separate optimization capabilities are being packaged and governed as one product. The interesting question is not when AI-driven authorization optimization became commercial. It has been commercial for at least a decade. The question is what it is becoming. The answer is a unified decision layer that governs the authorization lifecycle, and that is a harder thing to build and a more defensible thing to own.

What Antom is actually showing

Card Revenue Booster coordinates five things in one decisioning layer. Routing, adaptive messaging, authentication, retry, and credential lifecycle management. The list is ordinary. The word that matters is coordinate.

These capabilities have shared a platform before. Adyen’s 2016 RevenueAccelerate already combined message formatting, routing, account updating, and selective retries. Antom’s claim is narrower. A common decision layer can coordinate them from a shared view of transaction state, rather than present them as separate optimization features tuned in isolation. That distinction matters because the decisions are not independent.

A routing choice does not change the issuer. The card’s BIN identifies the issuer. Routing changes the path, data, authentication context, cost, and latency through which the issuer receives the request, and those factors can change the issuer’s response. Authentication context changes the authorization outcome, because issuers apply different logic depending on what came before. Retry timing either recovers a payment or trips issuer velocity and fraud controls and converts a good customer into a flagged one. Credential work upstream, network tokens and account updater services, can prevent a decline before any recovery is needed.

These effects run into each other. Optimizing each step alone produces a locally good, globally mediocre result. The shift Antom represents is treating the authorization lifecycle as one connected problem rather than a stack of point tools. That is the real change, and it is architectural, not commercial.

Stop framing it as fraud versus recovery

The lazy framing pits fraud prevention against revenue recovery as if they pull in opposite directions. They do not. They are the same optimization problem viewed from two ends.

The objective is not the highest approval rate. A higher approval rate is trivial to manufacture and often destroys value, because it waves through fraud, drives chargebacks, and earns issuer distrust that depresses tomorrow’s approvals. The objective is expected contribution margin. That is approved good revenue, minus processing cost, minus fraud loss, minus disputes, minus retry and scheme fees, minus the customer friction that quietly kills conversion and retention.

But margin is not observable at authorization. Fraud losses, disputes, refunds, returns, and retention arrive days or months later. The engine therefore optimizes a proxy under delayed and incomplete labels. That makes governance, calibration, and post-transaction measurement as important as the model itself.

A good engine maximizes that number. It approves more genuine transactions while holding fraud, chargebacks, and issuer penalties inside tolerances. Fraud control and recovery are two constraints on one margin function, not two products on opposite sides of a ledger.

A Transformer predicts. A payment system has to decide.

Antom leans on a Transformer model, and the architecture fits, because issuer behavior is sequential and path-dependent. But pattern prediction is necessary, not sufficient. Predicting that a declined transaction might succeed on retry is not the same as choosing the action that makes it succeed.

That is a causal decision problem. After a decline the system must choose whether to reroute, retry, alter message fields, invoke authentication, swap in a network token, or stop. Each action changes the outcome and the cost. Getting this right requires constrained decisioning, controlled experimentation, guardrails, and counterfactual measurement. The standard is not proving why any individual issuer approved a transaction. No provider can usually see inside the issuer’s decision. It is demonstrating incremental lift across a randomized holdout, after accounting for fraud, disputes, cost, and customer friction.

A model that only predicts will happily take credit for recoveries it did not cause. A system that decides has to measure what it actually moved. That gap is where most “AI-powered” payment claims quietly fall apart.

The moat is not the model

No serious provider has a moat because it “uses AI.” Model architectures diffuse in months. The defensible asset is privileged data joined to the ability to act on it.

That means issuer-response and outcome data at scale, deep merchant history, network-token access, direct authentication controls, multi-acquirer routing options, and a closed loop from action to measured outcome. Network-token access alone is not a moat, because most major processors offer it. The defensible asset is coverage, data quality, merchant-level outcome data, execution rights in the authorization path, experimentation capability, and the trust to make automated decisions there. Generic intelligence without control of the payment flow optimizes nothing. Control of the flow without rich data optimizes blindly. The provider that holds both can learn from every transaction and act on what it learns. That compounding loop is the moat, and it is far harder to copy than a Transformer.

What a merchant should demand

The buyer’s job is to refuse the dashboard. A before-and-after chart proves nothing, because approval rates drift with traffic mix, issuer policy, and season.

Demand a holdout. A randomized holdout is the cleanest baseline, and Checkout.com already publishes performance against exactly that. The holdout should be randomized within comparable cohorts, not merely across total traffic, or traffic mix will contaminate the result. Then demand the cut. Results by issuer or BIN cohort where sample size permits, country, currency, card type, credential type, transaction type, authentication treatment, decline reason, and first attempt versus retry, reported alongside fraud rate, dispute rate, and net margin. Separate gross authorization lift from incremental profitable revenue, net of processor fees, network costs, fraud losses, chargebacks, and the optimizer’s own fee. They are not the same number, and the difference is where merchants get fooled.

The deeper governance question is whose P&L the engine is optimizing. A processor can improve its own economics by favoring a route, token program, or acquiring path that is not the merchant’s best net outcome. The merchant’s real contribution margin also depends on data the processor may not hold, including settlement, fulfilment, refunds, returns, chargebacks, fraud losses, and customer lifetime value. The merchant should see the objective function, set hard constraints, and keep override rights. A payment control layer without auditability is outsourced discretion.

Where this lands

Authorization optimization is not a new place to compete. It is an old one that is becoming visible and strategic, and the basis for winning it is shifting.

The provider that can prove incremental profitable approvals, not merely a higher authorization rate, becomes genuinely hard to replace. “Fraud AI” and “recovery AI” will remain standalone buying categories. But they are insufficient descriptions of the leading payment platforms now taking shape. The product taking shape is a governed payment decision engine. A shared layer of models, policy rules, experiments, and merchant-set constraints coordinating decisions across the lifecycle, measured against a holdout, optimized for margin rather than approval rate.

Antom is one signal of that engine arriving. It will not be the last, and the ones that win will be the ones that can prove what they caused.

The views expressed here are my own and do not represent those of any employer or organization. This is commentary and analysis, not investment, legal, or tax advice.

When you deposit cash at a bank, you become a creditor of the bank.

When you hold a security in custody, the premise is different. Properly segregated custody assets should not be the custodian’s property. They should not sit on its balance sheet. They should not be available to its creditors if the custodian fails.

That is the real difference between banking and custody.

Custody is bankruptcy-remoteness.

That sounds technical until something breaks. Then it becomes everything.

If records are clean, assets are segregated, and ownership is traceable, the client owns the asset. If records are broken, assets are commingled, or entitlement is unclear, the client may be fighting over a claim.

Segregation is the line between owning your asset and owning a claim against a failed institution.

That is the back office at its most important: the legal and operational machinery that decides whether ownership can be proven when it matters.

You usually do not own the share directly

There is another fact most investors never think about.

In the US street-name model, the investor is usually not listed directly on the issuer’s register.

The investor is the beneficial owner. The broker records the investor’s position. DTC records the broker’s position. Cede & Co., DTC’s nominee, appears as the registered owner on the issuer’s books.

So when a public equity trade settles, the issuer’s register usually does not change.

The chain changes.

That is the point. The market is not constantly rewriting millions of company share registers. It is maintaining a layered chain of claims.

The investor owns the economic interest. The broker records the client claim. The depository records the intermediary claim. The custodian reconciles the client’s version of reality against the market’s version of reality.

Custody is not storage of your asset.

It is maintenance of your claim.

Finality means the chain agrees on who holds what. That agreement is the asset.

The depository holds the security. The custodian makes it usable.

A depository is market infrastructure. It holds securities electronically and moves them through settlement systems.

A custodian is investor infrastructure. It handles settlement coordination, asset servicing, cash, corporate actions, tax, reporting, reconciliation, and exceptions.

The difference looks simple until you ask what the investor actually needs.

Did my trade settle? Was the right account credited? Did cash move? Did the instruction match the counterparty’s instruction? Was the dividend entitlement correct? Was withholding tax applied properly? Did the position reconcile against the depository? Is the client statement reliable?

The depository helps prove that securities exist and can move.

The custodian proves that the client’s position is usable.

That is the overlooked value. Custody is the operating system around ownership.

Clearing guarantees the market trade. Custody proves the client result.

There is another institution sitting between execution and settlement: the central counterparty.

In US equities, NSCC steps into the middle of the trade. It becomes buyer to every seller and seller to every buyer. Once the trade is validated, the clearing house guarantees settlement, supported by member margin, netting, and a mutualized risk framework.

That distinction matters.

The custodian does not guarantee the whole market by itself. The clearing house absorbs counterparty default at the market level. The custodian makes sure the client record, cash movement, securities position, entitlement, tax, and reporting line up with that market guarantee.

Clearing turns bilateral counterparty risk into centralized market risk.

Custody turns the cleared market result into a client-owned position.

Both are needed before a trade becomes real.

DvP is institutionalized distrust

Delivery versus Payment is usually explained as simultaneous exchange of securities and cash.

Correct, but too polite.

DvP exists because markets cannot rely on goodwill.

Without it, one party can pay cash and not receive securities. Or deliver securities and not receive cash. DvP removes that one-sided exposure. Either both legs settle, or neither settles.

This is one of the deepest ideas in market infrastructure: trust is replaced by mechanics.

Securities settlement uses DvP. FX uses payment-versus-payment. Lending uses collateral. Corporate actions use record-date entitlement. Reconciliation uses agreement between internal and external books.

These are not rituals. They are controls built around a hard assumption: incentives, records, systems, and timing will not always line up.

Cash penalties can make settlement fails more expensive. They do not eliminate the exposure. If the penalty is lower than the economic benefit of failing, the fail can remain rational.

Mature markets do not assume clean behavior. They design around failure.

The simple processes are where firms get hurt

The most dangerous custody work often looks simple from the outside.

Take a cash dividend.

ABC Ltd. announces USD 1.00 per share. The client holds 70,000 eligible shares. Withholding tax is 15%. Net payment should be USD 59,500.

The math is easy. The control process is not.

The custodian has to verify the announcement, record date, ex-dividend date, payable date, currency, client account, eligible position, restrictions, withholding treatment, payment status, and reporting output.

A dividend is not one event. It is an entitlement decision.

Who owned what, in which account, on which date, under which tax treatment, with what net payment?

That is where errors happen.

Dividend-tax fraud showed what happens when the market cannot prove who owned the dividend at the right moment. Traders used rapid share movements around dividend dates to create multiple tax refund claims against one economic dividend. The CumEx Files investigation estimated these dividend-tax schemes caused tax losses of about EUR 150 billion between 2000 and 2020.

That is not a side story. It is the dark version of corporate action processing.

The timing that looks like plumbing becomes the attack surface.

A wrong entitlement is an operational error. A manufactured entitlement is a heist.

Securities lending separates economics from control

Securities lending has the same pattern.

On paper, it is simple. A borrower needs securities. A lender has securities. The custodian arranges the loan. The borrower posts collateral. The securities come back later.

In reality, the business depends on disciplined reuse.

Collateral must exceed the value of the loan. It must be marked to market. Margin has to be called when values move. Exposures have to be monitored daily. Loan agreements need to be enforced. Returned securities and released collateral have to line up.

The custodian is often selling more than administration.

In many lending programs, the agent indemnifies the lender against borrower default, usually after collateral is applied. That means the custodian is not just matching two parties. It is absorbing a tail risk the lender does not want to carry directly.

The visible transaction is the loan. The real product is the guarantee that reuse will not become an unsecured loss.

There is another wrinkle.

Lending separates two things investors assume are inseparable: the economics of a share and the right to vote it.

If shares are on loan over a voting record date, the borrower holds the vote while the lender keeps the economic exposure. A share can be economically owned by one party and politically controlled by another.

Reuse does not only touch the security. It can touch the ballot.

Faster settlement moved the risk

The move to T+1 settlement made one thing clear: faster is not automatically safer.

Shorter settlement reduces time exposure. It also compresses the repair window.

There is less time to affirm the trade, fix mismatches, source cash, recall lent securities, complete FX, and solve exceptions before the deadline arrives.

The FX problem is especially sharp.

US equities close at 4pm Eastern. CLS, the key utility that reduces FX settlement risk through payment-versus-payment, has a 6pm New York cut-off for T+1 FX trades. Miss the window, and the currency leg may settle outside CLS, with more settlement risk.

A reform designed to reduce risk in equities can push cross-border investors back toward one-sided exposure in FX.

Risk does not vanish. It moves.

There is another cost hiding underneath. Overnight processing was not just legacy laziness. It gave firms time to net, repair, fund, and optimize. Compress the cycle and more activity moves into intraday liquidity. Speed is paid for in cash, systems, staffing, and exception capacity.

The same applies to securities lending. If a fund sells shares that are out on loan, the shares must be recalled fast enough to settle. Earlier recalls can reveal trading intent. Failed recalls can create settlement fails. Manual recall processes become market risk.

The front office thinks it made a trading decision. The back office determines whether that decision can physically settle.

Tokenization does not remove custody

Tokenization makes the same point in sharper form.

A tokenized security does not become useful just because it exists on-chain. It needs legal standing, entitlement treatment, settlement rules, collateral treatment, custody arrangements, and a way to reconcile the token record against the authoritative securities record.

That is why the depository still matters.

In DTC’s tokenization model, the tokenized entitlement is connected to securities held within the existing DTC custody framework. The digital representation can move on approved blockchain infrastructure, but its credibility depends on the underlying market infrastructure.

Custody did not disappear. It became the legal anchor that makes the token enforceable.

The most interesting detail is not the token. It is the mismatch between records.

In early tokenization frameworks, the same economic exposure may be recognized differently across traditional and tokenized infrastructure.

That is the future in one sentence.

The asset moves faster. The records multiply. Someone has to decide which record is authoritative and who pays when two records disagree.

Reconciliation is the market’s truth function

The clean workflow is not where custody earns its keep.

The value shows up when something does not match.

A trade fails. Instructions are unmatched. A depository position differs from the internal book. A cash balance breaks against the bank. A corporate action entitlement is unclear. Collateral coverage falls. A token record and a depository record diverge.

This is where operations stops being process and becomes judgment.

The custodian has to isolate the break, find the source, escalate early, communicate clearly, and prevent a small mismatch from becoming a client problem.

It also has to know that not every fail is an accident.

Sometimes failing is cheaper than settling. If the security is scarce and the borrow cost is higher than the penalty, the fail becomes an economic choice. The counterparty is not confused. It is financing itself by not delivering.

So reconciliation is not only finding mistakes. It is identifying when a break is being used as a strategy.

If the books do not reconcile, the client does not know what it owns.

If the client does not know what it owns, everything downstream becomes suspect: performance, liquidity, risk, tax, collateral, compliance, and reporting.

Most of the market talks about alpha. Custody protects the factual basis on which alpha is measured.

The real product is certainty

Sometimes the client experiences certainty because the custodian advances it operationally.

Under contractual settlement or contractual income practices, a custodian may credit the client on the expected settlement date or payable date before the underlying cash or securities have actually arrived. The client sees the trade or income on time. The custodian manages the timing gap and reserves the right to reverse if the market does not deliver.

That is custody in one move. Turn market uncertainty into client certainty, then manage the residual risk behind the curtain.

Strip all of this down and custody solves one problem in many costumes.

The ownership chain. The FX cut-off. The dividend entitlement. The securities lending recall. The token versus the depository record.

The same question keeps returning. Which record is true, and who is liable when two records disagree?

New rails do not retire that question. They multiply the number of records that have to agree.

That is why custody sits in the quiet center of capital markets.

It is not where the trade idea begins. It is not where the headline innovation usually appears. It is rarely where the story gets told.

But it is where the market becomes enforceable.

The front office creates the trade. The custodian proves what the market actually owns.

The views expressed here are my own and do not represent those of any employer or organization. This is commentary and analysis, not investment, legal, or tax advice.

In June, three companies shipped the same idea without saying so out loud.

Coinbase unveiled a card built on Cardless. Visa expanded its work with Bridge and Lead Bank. Rain added a rewards layer on top of its issuing stack. Each move pointed at one structural change: card programs are starting to originate in tokenized dollar balances instead of bank deposits.

This is not a crypto story. It is a settlement story. And it is the most important one in payments right now.

What actually changed

For a decade, the card sat on top of a deposit. You funded an account at a bank, the bank held the balance, and the network moved value against it. Stablecoins were a parallel system that lived in apps most people never opened.

Three shifts collapsed that separation.

First, merchants stopped caring about the source of funds. A merchant wants value that clears and reconciles on time. Whether it started as fiat, credit, or a tokenized dollar is now an implementation detail, not a business decision.

Second, the stablecoin went invisible. The user never sees the token. They tap a card. The settlement rail underneath runs on tokenized dollars, and the experience is identical to any other card. Invisible infrastructure is the sign that a technology has stopped being a feature and started being plumbing.

Third, issuance unbundled. Rain is the clearest example. Rewards, branding, and settlement flexibility now detach from the core issuing stack. A company can assemble a card program from parts instead of buying the whole thing from one issuer. That is how a market scales.

Why the banks are pushing back

The banking lobby sees the threat clearly. Their fight is happening inside GENIUS Act rulemaking, and their word for it is deposit displacement.

The mechanics are simple. A stablecoin backed by safe assets functions like cash for the user. It sits outside deposit insurance and outside the central bank backstop. So the user treats it like money, but the bank cannot treat it like a deposit. That asymmetry matters because deposits are the raw material banks lend against. Money that leaves the deposit base does not just move. It stops working the way the banking system needs it to work.

The lobby is not wrong about the risk. They are arguing about who controls the float and who carries the systemic role when spending and settlement move off the bank balance sheet.

The real question

Strip away the announcements and one question remains. Are payments still anchored in banks, or are spending, credit, and settlement being rebuilt from the chain upward?

The honest answer is that both are true at once. Banks still hold the reserves behind the major stablecoins. Lead Bank sits inside the Visa and Bridge structure for a reason. The chain is not replacing the bank. It is repricing the bank’s role from owner of the rail to provider of one input.

That is a worse position than owning the deposit, and it is a better position than being cut out. The banks that understand the difference will partner early and set terms. The banks that wait will take whatever terms the issuers and networks hand them.

What to watch

Rain, because unbundled issuance is the mechanism that turns this from a handful of pilots into a market.

American Express, because its ACE Developer Kit verifies agent intent before execution, which is the trust layer agentic commerce needs before money moves on its own.

The GENIUS Act rulemaking, because the deposit displacement fight will decide how fast tokenized balances are allowed to behave like the money users already think they are.

The card moved on-chain this month. The argument about what that means for banks just started.

The views expressed here are my own and do not represent those of any employer or organization. This is commentary and analysis, not investment, legal, or tax advice.

Most bankers meet stablecoins with one fear: deposits leave, the wallet takes the customer.

The fear is half right. The coin is not the threat. Losing the funding, the conversion, the data, and the settlement around it is.

Two decisions sit underneath, and banks conflate them. Customer economics and balance-sheet economics are separate. You can keep the customer and still lose deployable funding.

Scope: regulated payment stablecoins, not algorithmic tokens.

What the fear gets right

• A dollar that becomes a stablecoin can leave your balance sheet.

• For USDC, most reserves sit in a government money-market fund. Circle reported about $10.36B at banks against $77.13B total reserves as of March 31, 2026, roughly 13%.

• The originating bank can lose the deposit. The funding reappears elsewhere in the system, not necessarily at you.

• Deposit migration can change lending capacity, most for banks that lose low-cost operating balances. The aggregate credit effect depends on where reserves land.

Take the fear seriously. Then get past it.

The two decisions banks conflate

• Customer economics. Do you keep the account, the identity, and the payment relationship.

• Balance-sheet economics. Do you keep deployable funding.

• Stablecoin capability protects customer primacy only when you keep the operating account, identity, funding, conversion, redemption, or payout relationship. It can still create funding drag.

• A stablecoin balance can support repeated payment, FX, payout, and reconciliation events. You capture value only where you own a monetizable service around those events.

The real question is who operates the customer’s money and which layers you keep.

Why a bank should want in

• Cross-border that works better. Correspondent banking runs 2% to 7% all-in over 3 to 5 days. Stablecoin rails settle the value-transfer leg in seconds. That leg is the slow, expensive one.

• New flows banks were never in. Agent payments are an emerging opportunity, mostly low-value software-to-software transactions. Reported data suggests AI agents made about 140 million micropayments over nine months, mostly in USDC. The bank opportunity is not an uncontrolled agent wallet. It is a governed operating account with spend limits, merchant rules, approval policies, and reconciliation.

• Defense. If you do not offer it, a fintech will, and it owns the customer.

Fast transfer is not commercial completion

• Moving the token in seconds is the easy part.

• The recipient still needs sanctions clearance, FX, local payout, reconciliation, beneficiary confirmation, error recovery, and support.

• The bank that solves those layers earns the margin.

• The low corridor pricing you see quotes the transfer leg. It assumes someone else eats the last mile.

The objections, answered

• “We lose deposits.” You lose them to someone else’s coin. Offer your own product and keep the relationship. Accept that funding may still move.

• “The token is not insured.” A properly structured tokenized deposit can retain ordinary deposit treatment, including applicable FDIC coverage limits. A payment stablecoin does not receive pass-through FDIC insurance through its reserve account.

• “We cannot compete on yield.” GENIUS limits issuer-paid interest. It does not end third-party reward economics. Coinbase rewards are the live example. Compete on utility.

• “Stablecoins drain credit.” They can shift funding away from lenders that depend on transactional deposits. A tokenized deposit keeps the claim on a lending balance sheet.

Why issuing a coin is not the win

• Circle proves the point against Circle. Circle incurred about $907.9M in Coinbase-related distribution costs in 2024. The lesson is not that distribution always beats issuance. It is that distribution can claim a very large share of reserve economics.

• The pure issuer is rate-sensitive. Reserve income was 99.1% of Circle’s 2024 revenue and 96.0% in 2025. Earnings stay materially exposed to reserve rates, even though lower rates also cut some distribution payments.

• Minting is commoditizing. Regulated issuance is not. Stripe’s Bridge won a conditional OCC trust charter in February 2026. Reserves, custody, and supervision are the moat, not the mint.

A coin with no acceptance does not move.

Where a bank makes money

Own the layers that hold the customer, not the coin.

• The operating account. The customer’s primary money.

• The off-ramp. Payees want fiat. Conversion charges spread, float, and fees. Own the payout and you keep that.

• Compliance infrastructure. Win on KYC and KYB, wallet screening, sanctions monitoring, custody, transaction policy, and payout approval. Issuer-level freeze authority sits with the issuer, not you.

• Reserves and custody. Even without issuing, you can hold the backing.

You do not need every layer. You need durable control of the ones that keep the account, the data, and the fees.

Read the card networks correctly

• Visa reported a roughly $7B annualized stablecoin settlement run rate for issuers and acquirers. That is settlement, not merchant acceptance.

• Mastercard added stablecoin settlement and agreed to acquire BVNK, which is investment in infrastructure, not proof the rail is absorbed.

• The read: card networks are integrating stablecoin settlement. That reinforces the value of acceptance, acquiring, and fiat conversion. It does not remove the need for merchant distribution.

The tools, and when to use each

• Tokenized deposit. Stays your liability, keeps paying yield, supports lending. JPMorgan positions JPM Coin for both programmable and cross-border payments.

• Stablecoin. Moves across wallets and platforms that never onboard to your bank. Best for open and agent payments.

• Tokenized deposits trade open participation for permissioned eligibility. That constrains reach. It does not prevent programmable or cross-border use.

Note the border: GENIUS raises the compliance bar for foreign stablecoins. Whether USDT can meet the pathway is unresolved.

Match the play to the flow

• Known corporate counterparties. Tokenized deposits and programmable liquidity.

• Supplier and contractor payouts. Operating account, conversion, FX, and local payout.

• Merchant settlement. Acquiring, prefunding, stablecoin settlement, and reconciliation.

• Agent payments. Controlled wallet, policy engine, limits, and audit trail.

• Open-network payments. Stablecoin access where you own onboarding or off-ramp.

Where GENIUS stands

• Congress established a regulated pathway for payment stablecoins.

• The OCC and FDIC proposed issuer, reserve, and custody standards. Treasury, FinCEN, and OFAC proposed AML and customer-identification rules. These are proposed, not final.

• The Act takes effect January 18, 2027, or 120 days after final rules, whichever is earlier.

• Final rules will decide which issuer structures, controls, custody models, and foreign-issuer arrangements can operate in practice.

The decision

• The choice is not whether to issue a stablecoin.

• It is which parts of the customer’s money flow you keep, and whether you accept funding drag to keep them.

• Own the account, the conversion, and the payout, and you keep the customer and the fees.

• Cede those layers, and someone else operates your customer’s money.

Stablecoins will not eliminate banks. They can shift funding, customer data, conversion economics, and settlement control to firms that operate the flow better.

The views expressed here are my own and do not represent those of any employer or organization. This is commentary and analysis, not investment, legal, or tax advice.

The issue is no longer whether banks can send compliant ISO 20022 messages. The issue is whether they keep the structured data inside those messages, collect more of it at the source, and turn it into products treasurers will pay for.

On 22 November 2025, the Swift cross-border coexistence window closed. For in-scope CBPR+ interbank payment instructions, ISO 20022 MX over FINplus is now the network standard. MT103 and MT202, the workhorses of correspondent banking for a generation, are gone from in-scope CBPR+ payment-instruction traffic.

Fedwire cut over on 14 July 2025. T2 went in 2023. CHAPS has mandated LEIs for in-scope financial-institution payments and purpose codes for FI and property payments since 1 May 2025, through channels within CHAPS Direct Participants’ control. From November 2027, the Bank of England expands mandatory purpose codes to all CHAPS payments within that controlled-channel scope.

The major high-value payment rails can now carry richer data.

That does not mean banks are keeping it.

The Problem: Banks Are Mapping Rich ISO 20022 Data Back Into Old MT-Shaped Systems

When an ISO 20022 MX message arrives, many downstream systems still expect old MT-shaped data.

So banks take the practical route. They map the new message back into the old model so payment engines, sanctions filters, reconciliation tools, and reporting systems do not break.

That solves the deadline problem.

It creates the commercial problem.

Swift’s own translation guidance shows the issue. MX-to-MT mapping can create missing, truncated, or non-translatable data wherever MT is too short or has no equivalent field. From 1 January 2026, in-flow translation also became chargeable for payment-instruction messages. So the workaround is now both lossy and metered.

A pacs.008 can carry structured remittance data of up to 9,000 characters. It can carry ultimate-debtor and ultimate-creditor information. It can carry LEIs, purpose codes, and a UETR that gives chain-level traceability, although not data completeness.

If that data gets squeezed into MT-shaped downstream models, especially Field 70-style remittance constraints, the bank stays compliant but loses the fields with the highest commercial value.

The problem is simple.

Banks upgraded the message format. Many did not upgrade the data model.

ISO-Native Banks Will Have Better Data Than Banks Still Operating Through Legacy Models

This creates a two-tier market.

ISO-native institutions can process and retain the structured payload.

Banks that rely on legacy mapping stay tied to old data models and may need to re-ingest or reconstruct the data later, if they can.

That difference matters because structured payment data is not just operational metadata. It can improve cash-flow forecasting, reconciliation, counterparty visibility, fraud detection, sanctions screening, liquidity planning, and working-capital analytics.

The bank that preserves the data has a product advantage.

The bank that strips it out has only completed a compliance project.

The Data Has to Be Captured Before the Payment Hits the Rail

The richest field in an ISO 20022 message is only useful if the bank collected it in the first place.

A structured remittance block has to be captured at pain.001, at the host-to-host channel, inside the ERP or treasury integration, or during onboarding.

An ultimate-creditor LEI has to be captured before the message is sent.

A purpose code has to be requested at the source.

The rail cannot enrich what the channel never collected.

This is why the November 2026 structured-address mandate matters. Swift reported in March 2026 that approximately 65% of payment messages still contained unstructured addresses. The mandate requires at least town and country in dedicated fields and removes unstructured-only postal addresses from Swift cross-border payments.

That is not a minor formatting change.

It forces banks, corporates, onboarding teams, payment channels, and treasury systems to clean up the data at the source.

Payments Does Not Own All the Data Capture Points

This is where many banks will miss the value.

The payment rail sits in payments.

The data capture point sits in cash management channels, onboarding, ERP integrations, treasury interfaces, and client service workflows.

That means ISO 20022 data monetization is not only a payments project.

It is also a channel project, an onboarding project, a product project, a legal project, and a sales project.

If those teams do not share ownership, the data leaks before it becomes a product.

Network Position Determines Data Quality

Structured data is only valuable if it is captured before it degrades.

Data can degrade across legacy systems, local mappings, bilateral flows, and any leg where the rich payload was never collected.

The first-leg bank has an advantage.

The on-us bank has an advantage.

The bank that owns the corporate channel has an advantage.

A correspondent bank three hops away may receive the payment, but not the same usable data.

For a large global transaction-banking network, this is the real advantage. The data is not valuable because it is secret. It is valuable because only the institution close enough to the client and the payment source can capture it cleanly and preserve it intact.

Coverage matters only if the bank’s data model keeps what the network captures.

Where Banks Can Monetize ISO 20022 Data

The product is not the payment itself.

The product is what the bank can show the treasurer after it processes the payment.

The clearest monetization surface is the reporting layer: enriched camt.053 and camt.054 statements, APIs, dashboards, alerts, forecasts, and workflow tools built from structured payment data.

1. Cash-Flow Forecasting

Purpose-coded, counterparty-resolved inflows and outflows are better forecasting inputs than balance snapshots.

A treasurer does not want more rows of data. The treasurer wants better visibility into what cash is coming, what cash is leaving, when it will settle, and why.

2. Counterparty and Supply-Chain Visibility

Ultimate-party data can expose customer, supplier, and counterparty concentration that a corporate may not see cleanly inside its own ERP.

This matters for supplier risk, customer exposure, working-capital planning, and dependency analysis.

3. Working-Capital and FX Timing

A bank that sees settlement rhythm across a payable or receivable book can help clients time liquidity, funding, and FX more precisely.

That is not generic reporting.

That is pricing liquidity against real payment behavior.

4. Compliance and Entity Resolution

Structured party fields can improve screening precision.

The same entity-resolution layer used for internal sanctions, AML, and fraud controls can also become a client-facing service, if legal, consent, and data-use controls are designed correctly.

The Bigger Prize Is Labeled Settlement Data

Properly permissioned and joined to outcomes, counterparty-resolved payment data becomes training data.

It can improve models for forecasting, fraud, liquidity, credit, and client behavior.

But this only works if the bank designs consent, lineage, retention, residency, and purpose-limitation controls from the start.

This is where banks have an advantage over many challengers.

Challengers can copy features.

They cannot easily recreate decades of labeled settlement behavior across a large transaction-banking network.

The Regulatory Calendar Creates Commercial Deadlines

The mandate schedule is not just a compliance calendar.

It is a product roadmap.

November 2026 removes unstructured-only postal addresses from Swift cross-border payments. Hybrid and fully structured formats remain.

November 2027 extends mandatory purpose codes across all CHAPS payments within DP-controlled channels.

MT101 treatment from November 2026 shows the direction of travel. Multiple-instruction MT101s reach end of life and are rejected, while single-instruction MT101s pass through contingency conversion with added validation and fees.

For relevant CBPR+ reporting flows, the migration continues through 2027, with coexistence ending in 2028.

Each date raises the minimum standard for payment data quality.

It does not create clean semantic data by itself. But it does create windows where banks can build ahead of the market and monetize each step-change as it arrives.

Why Banks Will Still Miss the Value

The blocker is not XML.

The blocker is ownership.

• Payments operations can preserve the data, but product needs the right to package it.

• Legal needs to approve secondary use under residency and purpose-limitation rules.

• Sales needs a paid client narrative.

• Channels need to capture the fields at source.

• Onboarding needs to collect better data.

• Technology needs to persist native MX before mapping.

Most banks will struggle because the capability sits across too many teams and no single owner has the full revenue mandate.

Without that operating model, ISO 20022 becomes a richer archive, not a richer business.

What Banks Should Build Now

1. Preserve Native MX Before Mapping

Persist the native MX message, including structured remittance data and ultimate-party fields, before any downstream mapping.

Truncated fields do not reliably come back inside the bank’s own data estate.

2. Capture Structured Data at the Source

Push structured collection into onboarding, host-to-host channels, ERP integrations, treasury portals, and corporate payment workflows.

The rail cannot enrich data that the channel never requested.

3. Build a Queryable Structured Payment Store

Treat the structured payment store as a product layer, not a passive log.

Design it for entity resolution, counterparty analytics, remittance search, auditability, consent, lineage, and jurisdiction-level controls.

4. Launch Paid Treasurer-Facing Products

Start with one or two products that a treasurer can understand and buy.

• Cash-flow forecasting.

• Counterparty exposure.

• Working-capital timing.

• Reconciliation intelligence.

• Compliance-grade entity resolution.

Give the product a price, a sales story, and a P&L owner outside pure payments operations.

The Migration Was Not the Finish Line

If ISO 20022 is treated as a compliance event, the migration becomes a one-time cost.

If it is treated as a structured-data event, the same work can become a data franchise.

The rails are already capable.

The open question is whether banks keep the structured data, collect it earlier, and turn it into products before legacy systems strip the value away.

Disclaimer: This post is for general informational and educational purposes only. It is not legal, regulatory, investment, compliance, or operational advice. References to ISO 20022, Swift, CBPR+, CHAPS, Fedwire, T2, and related migration timelines are based on public information and should be validated against official guidance before use in production, regulatory, or compliance decisions.

In May 2026, a single platform processed $7.2 trillion in tokenized repo transactions. That is roughly $362 billion a day, up 220 percent from a year earlier. It did not run on a meme chain. No token went to the moon. The platform belongs to Broadridge Financial Solutions, a company most people outside back-office finance have never heard of, best known for mailing you the proxy statement you throw away every spring.

I find this the most interesting story in digital assets right now, precisely because it is so understated. It cuts against almost everything the tokenization narrative has promised for a decade, and it points at where the value actually lives.

The idea

For ten years, the pitch for tokenization has been about access and issuance. Put real-world assets on a blockchain. Fractionalize the Mona Lisa. Let anyone, anywhere, own a slice of a Manhattan skyscraper or a private equity fund. The implicit promise was that putting an asset on-chain would, by itself, summon liquidity.

For the most part, it did not happen. BlackRock’s BUIDL, one of the largest tokenized money market funds, has spent most of its life with a small holder base, often fewer than a hundred addresses depending on the chain and data source, heavily concentrated, with a single address at times controlling roughly a third of the fund. Many tokenized assets, once issued, simply sit there. Even executives at the banks building these platforms have conceded the awkward truth that tokenization does not equal liquidity. An SEC commissioner went further in a May 2025 speech titled “Tokenization: Our Field of Dreams?”, arguing that the settlement cycle is a design feature rather than a bug. The existing clearing system already nets away the vast majority of trade obligations, on the order of 98 percent, and the United States moved to next-day settlement in 2024 with exactly zero blockchain involved.

So if issuance is not the unlock, what is?

Velocity, not supply

The Broadridge repo platform answers that question, and the answer is plain. It does not create new assets. It makes existing ones move faster.

Here is the mechanic. In a normal repo trade, a bank pledges collateral, usually Treasuries, in exchange for short-term cash. The collateral has to be located, instructed, and delivered through custodians, often overnight at minimum, even if the actual cash need is only three hours long. The asset has to be moved or re-recorded through custodial and settlement accounts. This is slow, it ties up balance sheet, and it forces banks to hold expensive buffers of liquid assets against the timing mismatches.

On a shared ledger, the collateral is locked in place and ownership transfers by smart contract. The bond never moves. A bank can borrow for exactly three hours and unwind automatically. The same security can be pledged, returned, and re-pledged many times in a single day.

That sounds like a small convenience. It is actually a new product. Liquidity has always been sold in coarse blocks, overnight at the finest. Once money can be lent for three hours and returned on a schedule, a term structure starts to form below the level of a single day. That starts to make intraday funding explicitly priceable: a price for an hour of cash, distinct from the price of a night of it. As that market matures, the interesting question is not whether the bond is tokenized. It is what gets priced once the smallest practical funding unit shrinks.

This is the part worth internalizing. Tokenization, done usefully, is a velocity play, not a supply play. You are not minting more Treasuries. You are turning the same Treasuries more times. After 2008, the velocity of collateral in the financial system actually fell, by some estimates from around three turns to under two, as post-crisis rules pushed everyone to hoard high-quality assets. The genuine promise of tokenized collateral is reversing that, quietly, inside the regulated system, for institutions who care about basis points of balance-sheet cost rather than vibes.

A detail worth flagging, which Broadridge’s own team has described publicly, is the time-zone trade. On a 24/7 ledger, an Asian institution can fund an American counterparty’s position outside US market hours, when the traditional books are closed. The time zone stops being a wall and becomes an opportunity. That is a real edge, and it has nothing to do with speculation.

There is a shadow to this, and it is worth naming, because almost no one selling the efficiency does. The post-2008 collapse in collateral velocity was not only fear. Regulators wanted high-quality assets to sit still. The same Treasury pledged through a five-link chain is the same Treasury that disappears five ways when one link breaks. Velocity and fragility can be the same number read in two directions. A ledger that turns collateral faster can also rebuild dense chains of dependency. The visibility may improve, but the dependency still has to be watched. Faster collateral is not just an efficiency story. It is also a supervision story. The same mechanism that creates efficiency can also create systemic risk. That does not make the technology bad. It means the honest case for tokenized collateral is not “faster is better.” It is “faster is better until it is not, and someone has to watch the chain.” Whoever watches the chain owns both the franchise and the supervisory burden.

The bottleneck was rarely the token

The people building this have, by their own public account, largely stopped talking about issuance. Broadridge’s product chief framed it well at a conference this spring: success is no longer defined by whether you can issue a tokenized asset, but by whether that asset can move through existing financial systems without friction. The goal is for tokenization to be a replacement for inefficiency, not an additional layer on top of it.

That reframing matters, because it explains why most tokenization projects die. They build a beautiful island. A lone tokenized bond on a private chain, disconnected from the custodians, settlement systems, and cash rails the rest of the market uses, is a museum piece. The graveyard is full of these. The Australian Securities Exchange wrote off about a quarter of a billion Australian dollars on a scrapped blockchain rebuild of its clearing system, and trade-finance consortia like TradeLens and we.trade folded after failing to reach durable commercial scale.

There are still two genuinely hard problems nobody has fully solved. The first is the cash leg. Tokenizing the bond is easy. Settling the cash side on-chain, with legal finality, is hard, because most money still lives in legacy central-bank plumbing. U.S. policy has shifted sharply against a central-bank digital currency, including a 2025 executive order barring federal agencies from establishing, issuing, or promoting one. That pushes the American cash leg toward private money: deposit tokens, regulated stablecoins, or systems that hold central-bank money behind the scenes. The second problem is legal. A late-2025 SEC no-action letter let the main US depository offer a tokenization service, but explicitly said it would not assign those tokenized entitlements collateral or settlement value for its own risk-management and settlement calculations. This is a reminder that a token representing an asset is not always the same as legally owning the asset.

These are not reasons tokenization fails. They are the actual work. And they sit deep in the plumbing, which is exactly why the serious progress is being made by infrastructure firms and not by token projects.

The moat hiding in the proxy statement

Which brings me to the part of this story I cannot stop thinking about.

Almost everyone in this race is fighting over settlement. JPMorgan has its single-bank platform. The DTCC is launching a tokenization service later this year. Goldman Sachs has moved to spin its platform out into an industry-owned utility, a structure that concedes, at least implicitly, that a bank-owned rail has a harder path to broad trust. They are all competing on the same leg of the trade.

Broadridge owns a different leg that almost nobody talks about: governance. It already runs proxy voting for more than 80 percent of US shares. And here is the insight. A tokenized share with no voting rights is not really a share. It is closer to a synthetic derivative. For any institution bound by fiduciary duty, that distinction is the whole game. In 2026, Broadridge launched on-chain shareholder voting for Galaxy, the first US public company to use its platform for tokenized-equity governance, with ballots delivered to digital wallets and votes recorded on-chain.

It goes further than that. When Ondo Finance, a major issuer of tokenized stocks and ETFs, wanted to give its token holders a voice in the companies behind those tokens, it did not build its own governance system. It plugged into Broadridge. Holders of more than 250 tokenized securities can now connect a crypto wallet, read the filings, and register a voting preference that the issuer routes to the underlying shares. Notice the shape of that deal. Broadridge became the governance layer for assets it never tokenized in the first place. The whole industry is quietly trying to bolt governance back onto tokens that were issued without it, and the company they are bolting it onto is the proxy incumbent.

Consider what that means competitively. The settlement players can move a tokenized asset. Only the proxy incumbent can make that tokenized asset behave like real equity, with corporate actions and voting attached. That capability sits in the company’s largest business, the communications segment, not the capital-markets one. The moat is not the blockchain. The moat is the decades-old infrastructure the blockchain plugs into.

Blockchain was sold as the technology that would finally remove the trusted intermediary. Ten years on, the intermediaries were not removed so much as entrenched. The settlement utility, the depository, and the proxy incumbent each ended up more central in the tokenized version of the market than in the analog one, because each owns something a token cannot mint for itself: legal finality, custody, the canonical record of who owns what. The trustless system turned out to need trust more than ever, and could only borrow it from the institutions that already had it. You can fork the code. You cannot fork fifty years of issuer relationships or the regulatory consent that lets a firm stand in the middle of a trade. The technology that promised disintermediation has, so far, tended to widen the moats of the incumbents it was meant to displace. Trust, it turns out, is hard to fork.

There is a sharper version of this for anyone watching the company rather than the category. The governance layer sits inside the communications business, not only inside the capital-markets technology story. That is the point: the durable advantage is not always filed under the line item everyone watches.

What I actually take away

I am not here to tell you to buy the stock, and the volume numbers deserve a skeptical footnote. Year-over-year growth is decelerating as the base gets large, and a single platform’s repo figures are not the same as the firm’s much larger, oft-quoted, whole-company trading numbers. Those figures are easy to blur, and they should not be treated as interchangeable.

But the strategic lesson generalizes well beyond one company. In every technology hype cycle, the loudest applications and the durable businesses are rarely the same thing. Tokenization’s loud phase was about issuance and access. Its durable phase is turning out to be about velocity, interoperability, the cash leg, and governance: the underlying plumbing of who owns what, how it moves, and how it settles with finality.

You can watch the shift happen in an unlikely document, a company’s own boilerplate. Broadridge now describes itself as underpinning the daily trading of over $15 trillion in “tokenized and traditional securities.” When a word like tokenized migrates out of the press release and into the single sentence a company uses to define what it is, the category has stopped being a project and started becoming an identity.

The winners are unlikely to be whoever issues the most tokens. They are more likely to be whoever already owns the rails the tokens have to run on. Sometimes the most important infrastructure in finance is the thing you have been throwing in the recycling bin every spring.

Disclaimer: The views expressed are the author’s own and do not represent those of any employer, client, or organization. This article is independent commentary written for educational and informational purposes only. It is not investment, legal, tax, or financial advice, and nothing here is a recommendation to buy or sell any security. The author is not affiliated with, sponsored by, or endorsed by Broadridge Financial Solutions or any other company mentioned, and all trademarks and company names belong to their respective owners. Figures and events are point-in-time, drawn from public sources believed to be reliable as of June 2026, but accuracy and completeness are not guaranteed and details may have changed since publication. The author holds no position in any company mentioned.

Sources

• Broadridge DLR May 2026 volumes: PR Newswire

• BUIDL holder concentration: Gate Learn, CoinDesk

• Crenshaw, “Tokenization: Our Field of Dreams?” (May 12, 2025): SEC.gov

• Time-zone funding and DLR mechanics (Horacio Barakat): finews.asia

• “Move through existing systems without friction” (Germán Soto Sanchez): Consensus Miami coverage

• ASX CHESS write-off: Finextra

• Goldman GS DAP spin-out: Goldman Sachs

• DTCC tokenization service: DTCC

• Broadridge first on-chain shareholder vote: Broadridge

• Ondo Finance proxy voting via Broadridge for 250+ tokenized stocks and ETFs: PR Newswire, CoinDesk

• Broadridge boilerplate (”$15 trillion in tokenized and traditional securities”): Broadridge

If your company holds idle cash, two products are now fighting for it, and most people cannot tell them apart. One earns you a yield. One moves your money. They look identical on a screen and behave like opposites on a balance sheet. Confusing them is about to become an expensive mistake.

Start with a fact that hints at what is going on. In early 2026, Circle’s USYC became the largest tokenized Treasury fund in the world, and Arkham Intelligence data showed roughly 94% of it tied to a single holder, Binance. The second largest, BlackRock’s BUIDL, is more diversified across roughly 100 holders, though its top ten still account for more than 95% of the fund. These are the flagship products of on-chain cash, and their holder bases are strikingly narrow.

That is the tell. This market is real and large, and it is still institutional infrastructure marketed with a retail face. The distinction that explains it is easy to miss, so let me make it impossible to.

There are two kinds of dollar competing for corporate cash right now. They look identical on a screen. They are opposite on a balance sheet.

Tokenized Treasury funds. Tokenized deposits.

The whole thing fits in one line.

The fund earns. The deposit settles.

A tokenized Treasury fund is a fund share. You own a slice of a pool of T-bills, repo, and cash. The token is the record of that ownership. It delivers the Treasury rate, roughly 3% to 4% today depending on the product. There is no FDIC insurance, because it is not a deposit. Its value can move, because it is a fund.

A tokenized deposit is bank money. You hold a 1:1 claim on a regulated bank’s deposit liability. The token represents that claim. Today, its primary product promise is settlement, not yield. It exists to move, instantly, around the clock, inside a control framework your auditor already understands.

Both live on-chain. But one is a tokenized fund share and the other is tokenized bank money. That difference matters more than the chain they run on.

So the question is never “which is better.” It is “what is this cash for.”

The access asymmetry nobody talks about

Here is the part that turns a plumbing story into a power story.

Some tokenized funds have a retail door. Tokenized deposits do not.

A US retail investor who passes onboarding can put $20 into Franklin Templeton’s BENJI. It holds in an app and funds with USDC. A college student can own a slice of a regulated government money fund before lunch. The institutional funds stay gated: BUIDL is for qualified purchasers, and Circle markets USYC as institutional-grade collateral. But the door exists.

Tokenized deposits have no retail door. JPMorgan’s JPMD went live to institutional clients on Base in late 2025 and is rolling onto the Canton Network in phases through 2026. It is permissioned, institutional only, no public access. HSBC went live with its Tokenized Deposit Service in the United States on April 13, 2026, its fifth market after Hong Kong, Singapore, Luxembourg, and the UK. It serves eligible corporate and institutional clients moving cash between treasury centers and subsidiaries. There is no app. There is no $20 entry. A deposit token is a bank liability issued to onboarded institutions, full stop.

The pattern matters. Some yield instruments have a retail door. Deposit tokens have a velvet rope. Some yield products are opening up, while deposit tokens remain reserved for the institutions that already had the best rails.

That is not an accident. It is the shape of the whole market.

The deposit-token field is splitting too

The funds are not the only side picking lanes. The deposit-token camp is already dividing into two philosophies.

• Kinexys (JPMD). Live on Base for institutional clients, moving to Canton in phases through 2026. The bet is public-chain reach and interoperability, JPMorgan’s balance sheet meeting open networks.

• HSBC TDS. Live in the US since April 13, 2026, across five markets and multiple currencies. The bet is a closed, multi-currency corporate network optimized for treasury-center-to-subsidiary flows.

Same instrument, two distribution philosophies. One reaches out to public blockchains. One builds a walled garden for corporates. Both are deposits, so both settle, and neither is trying to be the thing that earns for you. That contest is its own story, and it is worth watching.

The third dollar: the bond commits

There is a third tokenized instrument, and the useful thing about it is that it does not fit the operating-cash binary.

A tokenized bond is a debt security on-chain. You hold a claim to coupons and principal at maturity, with your rights set by the bond documents, not the token. It is an asset you own, like a fund share, but it behaves nothing like one. A tokenized Treasury fund is a diversified pool you can exit most days at a stable value. A bond is a single security with a fixed maturity, a price that moves with rates, and capital you have committed for the term.

So the map is not two instruments. It is three jobs. The deposit moves your cash. The fund parks it. The bond commits it.

This is the most gated corner of the three. Issuance has been mostly institutional and accredited: the European Investment Bank across multiple chains, Siemens on Polygon and a permissioned ledger settled in central bank money, and a wave of bank-led pilots. In December 2025, JPMorgan arranged a $50 million tokenized US commercial paper deal for Galaxy on Solana, settled entirely in USDC and bought by Coinbase and Franklin Templeton. The retail experiments are sovereign, like Hong Kong’s tokenized green bonds and Thailand’s retail G-Token government bonds. McKinsey projects total tokenized financial assets could reach roughly $2 trillion by 2030, with bonds and ETNs among the leading categories.

It also sharpens the asymmetry. The further the dollar travels from cash toward duration, the higher the wall. Yield in a fund has a $20 door. Settlement in a deposit is institutional only. A bond is institutional, accredited, or a government pilot. Access tightens as commitment lengthens.

One trap the wrapper hides. Trading a tokenized bond in seconds is not the same as the bond being liquid. You can move a ten-year obligation instantly and still find no buyer at par the day you need one. Instant settlement is not market depth, and no smart contract is obligated to make you a market. A tokenized bond can feel liquid right up to the moment it matters, which is the moment it stops.

How to actually use them

Cash that sits wants a tokenized fund. Treasury reserves, collateral buffers, the balance that funds nothing this week. Put it in the fund and let it earn. BlackRock structures BUIDL so eligible institutions can use it as derivatives collateral while preserving exposure to fund yield. That is idle cash doing two jobs at once.

Cash that moves wants a tokenized deposit. Treasury center funding a subsidiary. A cross-border liquidity move. A settlement leg that has to land at par the instant it arrives. Put it in the deposit token. It clears in seconds and stays on rails your controllers can sign off on.

Cash you commit wants a tokenized bond. Capital you can lock for a defined term in exchange for a fixed return. Different job, different instrument, and not somewhere to keep money you might need next week.

The treasurer who treats this as a single choice loses. The treasurer who maps cash by job wins.

The deposit’s real killer app is FX

“Settles” sounds dull until you apply it across currencies. The expensive, invisible problem in corporate treasury is cross-currency settlement. To pay a euro supplier, a treasurer pre-positions cash in foreign accounts, works around market cutoff times, and carries risk in the gap between paying out one currency and receiving the other. That gap has a name, Herstatt risk, and it has cost the banking system billions.

A multi-currency deposit-token network goes straight at it. JPMorgan’s Kinexys runs on-chain FX in USD, EUR, and GBP, settling cross-currency payments around the clock, past the close of traditional markets. Ant International used it to move dollars into euros in near real time, outside market hours, through its own treasury platform. The conversion and the payment clear together, so the settlement-risk gap closes, and cash that used to sit pre-funded across three time zones can stay productive instead.

That is the deposit side’s version of capital efficiency. The fund frees idle cash by letting collateral earn while it is pledged. The deposit frees trapped cash by killing the pre-funding and the cutoff times. Same prize, opposite instrument. It is also the clearest answer to the question a treasurer actually asks, which is not “can this token move,” but “can it move my euros at 2am on a Sunday without me parking money in three countries first.”

Why the banks are not panicking

For a year the story was that tokenized funds would drain bank deposits. A fintech could hold operating cash in a fund at 4% instead of leaving it at a bank earning nothing. Real threat. Real flows.

The banks are answering, and the answer is patient. They are not fighting the fund. They are absorbing the use case. JPMorgan has signaled the obvious next move: deposit tokens could eventually become interest-bearing. Then in May 2026 it filed JLTXX, a government money market fund designed for stablecoin issuer reserve assets under the GENIUS Act framework. The bank is positioning to sell the settlement instrument and the yield instrument and keep the relationship either way.

A note on that framework. The GENIUS Act became law in July 2025, but it created scaffolding, not a finished building. Regulators are still working through proposed rules. The direction is set. The details are not.

The part the explainers miss

Six things that are true and rarely said.

One. These funds are not savings products. They are collateral infrastructure. The concentration is the proof. When 94% of the largest fund sits with a single exchange, the real customer is not a treasurer parking idle cash. It is a trading venue and a stablecoin issuer that needs yield-bearing collateral. The money-fund label describes the wrapper, not the buyer. The demand comes from crypto market structure, not corporate treasury, and that shapes every decision the issuers make.

Two. The stablecoin law quietly built the funds’ demand curve. The GENIUS Act bars stablecoin issuers from paying any yield to the people who hold their coins. That yield demand does not vanish. It moves. It moved into BUIDL and USYC. A law written to protect bank deposits became the best growth engine tokenized Treasury funds ever had. Regulators are now moving to extend the ban to exchanges and affiliates, which would push even more yield-seeking cash toward the funds, not away from them.

Three. When deposit tokens pay yield, the fight does not end. It moves to the regulators. To a user, an interest-bearing deposit token and a tokenized fund will look identical. Legally they never converge. A deposit lives inside insurance, reserve, and capital rules. A fund never does. So convergence forces the real question. Is a programmable, transferable, interest-bearing deposit token still a deposit for capital and insurance purposes? That is a Basel and balance-sheet fight, not a product fight, and it is the one that decides who wins.

Four. The prize is capital efficiency, not speed. Everyone sells 24/7 instant settlement. The genuinely new thing is that the same asset can earn yield and sit as posted collateral at the same time, because the chain enforces both states at once. In traditional markets, pledged collateral is dead weight. On-chain, it keeps accruing while it secures a position. I spend a lot of time modeling treasury liquidity, and this is the part that quietly changes the math, because an asset no longer has to choose between earning and being pledged. Settlement speed is the headline. Capital efficiency is the business.

Five. Tokenization did not remove the middlemen. It traded them. The pitch was disintermediation. The reality is substitution. Do not assume the old intermediaries disappear. In many cases, the chain operator, oracle, bridge, custodian, and clearing layer all still matter. If a deposit token settles on a public chain, that chain’s operator now sits in your settlement path. If a fund is 94% one holder, that holder is your single point of failure. Count the intermediaries again. The number may have gone up, and the new ones carry less regulatory weight than the ones they replaced.

Six. Follow the issuer’s incentive and you can predict the roadmap. A bank earns when its deposit token moves, because it monetizes velocity and fees. An asset manager earns when its fund token sits, because it monetizes balance. Same rail, opposite economics. The bank wants your money in motion. The manager wants it at rest. When you cannot tell where a product is heading, follow which of those two incentives is paying for it.

Where the three dollars meet

Run the logic forward and you reach one question. If the deposit, the fund, and the bond all live on-chain, where do they settle against each other?

The answer the largest institutions are converging on is the Canton Network, a privacy-enabled blockchain built for regulated finance. The pitch is specific. A bank issues a deposit token, an asset manager issues a fund share, a clearinghouse settles a Treasury, and all of it clears in a single atomic transaction, in real time, with no party seeing another’s full position.

This is not a whiteboard. JPMorgan is issuing JPMD natively on Canton in phases through 2026. HSBC, whose own deposit service is a closed corporate network, still ran its tokenized deposit through a Canton pilot to test atomic settlement against digital assets, which tells you even the walled-garden players are hedging toward a shared rail. Franklin Templeton put its Benji fund platform on Canton in late 2025, wiring tokenized money funds straight into its collateral network. And in the move that signals where this goes, DTCC, the clearinghouse that settles most US securities, is bringing tokenized Treasuries onto Canton. The old intermediary is not being removed. It is migrating onto the new rail.

Watch which funds show up and which stay away, because the fund side is splitting exactly like the deposit side. Franklin’s BENJI is hedging onto Canton’s permissioned institutional rail while keeping its public-chain footprint. BlackRock’s BUIDL is doing the opposite, spread across multiple public blockchains and plugged into venues like Uniswap. One camp is betting on a private, compliance-first network. The other is betting on open chains and composability. Nobody yet knows which rail wins the institutional money, and that is the contest underneath the contest.

Here is why that matters, and it returns to capital efficiency. Proponents estimate there are roughly $300 trillion of high-quality liquid assets in the world, with only about 10% to 11% actively working as collateral at any moment. The rest sits still because moving it means settlement cycles, batch windows, and cutoff times. A network where cash, funds, and collateral settle atomically and around the clock is an attempt to wake up the other 89%. That is the prize. Not faster payments. Mobilized capital.

So here is the bet. Through the end of 2026 the clean split holds: funds win idle cash, deposits win moving cash, bonds hold committed cash, and no serious treasury team evaluates one without understanding the others. The collision starts when deposit tokens pay yield and all three instruments are clearing on the same network. At that point the question stops being which token you hold and becomes which network you trust to settle it.

What to do with this if you run treasury

Three moves.

First, sort your cash by job, not by account. Money that earns and money that moves are now different instruments, not different sub-ledgers of the same one. Write the policy before the products force the question.

Second, watch the par line. A tokenized deposit promises par because a regulated balance sheet stands behind it. A fund promises yield and a value that can move. In calm markets the difference is invisible. In a stress event it is the only thing that matters.

Third, treat the chain as plumbing, not strategy. The interesting question was never Ethereum versus Canton versus Base. It is whether the dollar you hold is an asset you own or a liability you are owed, and what you need it to do this afternoon.

The hard part is the policy, not the purchase

Buying the token is trivial. Writing the treasury policy around it is the work. A tokenized fund needs limits for NAV movement, redemption timing, collateral eligibility, counterparty concentration, and wallet controls. A tokenized deposit needs limits for bank exposure, network access, jurisdiction, operational permissions, and settlement finality. A tokenized bond adds duration, credit, and secondary-market liquidity to the list. The product is new. The discipline is old. Define the cash job, set the risk limit, then choose the instrument.

The line that ends the debate

Stablecoins taught the market that a dollar could be programmable. These two instruments are teaching the harder lesson. A programmable dollar is still either something you own or something you are owed. Tokenization did not erase that line. It put it on-chain.

The fund earns. The deposit settles. Pick the one that fits the job, then let it run.

Disclaimer: This article is for informational and educational purposes only. It is not financial, investment, legal, tax, or accounting advice, and it is not a recommendation to buy, sell, or hold any product, security, token, or instrument mentioned. Tokenized funds, tokenized deposits, stablecoins, and tokenized bonds carry risk, including loss of principal, and their treatment under law and regulation is still evolving. Product names and examples are used for illustration only and do not constitute an endorsement. Information is believed accurate as of the date of publication but may change, and no guarantee is made as to its completeness or accuracy. Always do your own research and consult qualified professionals before making any financial decision. The views expressed are my own and do not represent those of any employer, client, or organization with which I am affiliated.

Sources and notes on figures. Circle’s USYC overtaking BlackRock’s BUIDL as the largest tokenized Treasury fund: CoinDesk, 13 March 2026, https://www.coindesk.com/markets/2026/03/13/circle-overtakes-blackrock-in-tokenized-treasuries-as-market-hits-record-usd11-billion. USYC supported as yield-bearing off-exchange collateral for Binance’s institutional clients: Circle, https://www.circle.com/pressroom/circles-usyc-now-supported-as-yield-bearing-off-exchange-collateral-for-binances-institutional-clients. USYC product detail: https://www.circle.com/usyc. The USYC and BUIDL holder-concentration figures derive from Arkham Intelligence and rwa.xyz on-chain data as reported in the coverage above. On-chain concentration changes quickly, and all figures are as of publication.

In April 2026, the Linux Foundation launched the x402 Foundation, giving Coinbase’s x402 payment protocol a neutral open-source home with support from Circle, Google Cloud, Microsoft, Stripe, and Visa.

That matters because financial AI is moving past chat.

The next frontier is an agent that can safely touch payments, filings, ledgers, statements, sanctions lists, market data, and approval workflows.

That stack is the tool layer: the APIs, MCP servers, repos, data standards, ledgers, parsers, and controls that let an agent reach a real financial workflow inside proper guardrails.

The firms that win will be the ones whose systems an agent can operate. MCP matters because it gives AI applications one standard way to connect to external tools instead of a custom integration per system.

Here is the stack, by vertical.

The simple rule

A repo is valuable if it sits near one of five choke points.

1. Money movement.

2. Identity and account access.

3. Market and company data.

4. Risk, compliance, and sanctions.

5. Workflow governance.

Everything else is noise.

How to read this

• Repos are open-source code you run, fork, or learn from.

• APIs are hosted services you call in production.

• MCPs are Model Context Protocol servers. They let an agent call a financial system directly.

The MCP layer is early, but it is where the workflow advantage is forming.

1. The Agent Layer (MCPs)

Whoever controls the MCP layer controls how agents reach financial systems. These connect agents to real operations, not novelty.

• Stripe MCP exposes tools for payments, refunds, invoices, subscriptions, disputes, and payment links, hosted at mcp.stripe.com. A Treasury extension lets agents move money, pay bills, and issue cards.

• PayPal MCP is official and live at mcp.paypal.com, with both a local agent toolkit and a remote OAuth server. It covers invoices, orders, refunds, and subscriptions.

• Square MCP is in beta and connects agents to customers, orders, catalog, and payments. It is the SMB commerce option.

• Plaid MCP gives agents access to Plaid developer diagnostics, analytics, Link health, and integration support. The broad account and transaction data still runs through Plaid’s production APIs.

• SEC EDGAR MCP connects agents to filings, financial statements, and insider trading data. It is the diligence workhorse.

• OpenBB MCP serves market and research data into analyst and portfolio workflows.

• Moov Watchman MCP runs sanctions and screening checks as an agent-callable gate.

• Genesis Global MCP governs how agents act inside institutional capital markets applications, with permissions and human approval built in.

Guardrail. Read-only by default. Human approval before any money moves. Agent toolchains can now call payment, refund, billing, and treasury actions, so the controls are not optional. The unlock is not one MCP. It is chaining several safely.

2. Wealth Management

The cleanest agentic use case. Gather accounts, classify holdings, analyze risk, recommend, document suitability, push a human-approved action.

• Plaid Investments pulls balances, holdings, and investment transactions. It is the connective tissue for a household balance sheet. plaid.com/docs

• OpenBB is a credible open financial data layer covering equities, options, macro, and crypto. github.com/OpenBB-finance/OpenBB

• OpenFIGI maps securities to proper identifiers across tickers, CUSIPs, and ISINs. Wealth and asset management break the moment IDs fail to reconcile. openfigi.com

• PyPortfolioOpt implements mean-variance, Black-Litterman, shrinkage, and Hierarchical Risk Parity in Python. github.com/robertmartin8/PyPortfolioOpt

• QuantLib handles bond analytics, curve construction, and fixed-income risk. github.com/lballabio/QuantLib

Do not treat PyPortfolioOpt as a black-box allocator. Weak inputs produce confident, wrong answers.

Build this. An Advisor Control Tower. Pull holdings through Plaid, normalize through OpenFIGI, analyze risk through OpenBB and PyPortfolioOpt, generate recommendations, require human approval, log every source and override.

3. Payments

This is where agentic finance gets real, and dangerous. An agent that reads docs is useful. An agent that can trigger payouts needs hard limits.

• moov-io is one of the most important open-source payments ecosystems in the United States. It maintains ach (NACHA files), achgateway (event-driven ACH run in production), wire (Fedwire), iso8583 (card messaging), fed (routing lookup), and watchman (sanctions). All Apache-2.0, all Go. github.com/moov-io

• Stripe API and MCP cover cards, billing, and agentic commerce. The stripe/ai monorepo holds the agent toolkit. github.com/stripe/ai

• PayPal and Square MCPs cover merchant and SMB payment operations through natural language.

• Plaid Auth, Identity, and Transactions handle account funding, verification, cash-flow underwriting, and fraud screening.

• Blnk is an open-source double-entry ledger for wallets, balances, and reconciliation. github.com/blnkfinance/blnk

Rule. If money moves, there must be a ledger. Do not store balances as casual database fields.

Build this. A Payment Ops Agent. Read Stripe, PayPal, or Square through MCP, verify accounts through Plaid, screen counterparties through Watchman or OpenSanctions, write ledger entries in Blnk, require approval for refunds and payouts.

4. Treasury and Cash Management

The most underrated agent use case in the sector. Treasury is full of repetitive, governed, high-value workflows.

• ISO 20022 and CAMT.053 parsers ingest bank statements in the format adoption is accelerating toward across bank messaging. The EmergentFinancial/iso-20022 Rust SDK covers pain, pacs, and camt. github.com/EmergentFinancial/iso-20022

• Moov Fed and Wire validate routing data and Fedwire messages. github.com/moov-io

• FRED API gives programmatic access to rates, spreads, inflation, and macro series that drive cash and funding decisions. fred.stlouisfed.org

• Circle, Fireblocks, and Coinbase Developer Platform are the digital-asset treasury layer for USDC payments, vaults, wallets, and programmable payouts. developers.circle.com, developers.fireblocks.com, docs.cdp.coinbase.com

• Blnk again, for internal cash accounting.

Guardrail. Autonomous execution stays off until policy, approvals, limits, sanctions checks, and rollback are solved.

Build this. A Liquidity Control Tower. Ingest CAMT.053 statements, pull FRED rates, forecast 7, 14, and 30-day liquidity, compare T-Bills, money market funds, insured deposits, tokenized deposits, and stablecoins, recommend sweeps, require four-eyes approval.

5. Investment Banking

Banking is not short on data. It is short on clean, traceable workflows. The useful agent finds the right filing, extracts the right number, and cites the source.

• SEC EDGAR APIs expose company submissions and XBRL company facts. This is the foundation of public-company analysis. sec.gov

• SEC EDGAR MCP puts filing-backed research and comps in front of an agent. github.com/stefanoamorelli/sec-edgar-mcp

• FINOS Legend is the open data governance platform Goldman Sachs contributed for governed models and lineage. github.com/finos/legend

• FINOS Perspective is the analytics and visualization engine J.P. Morgan contributed, built for large and streaming data. github.com/finos/perspective

• OpenFIGI normalizes entities and securities across the comp set. openfigi.com

Build this. A Deal Diligence Agent. Pull filings through EDGAR, normalize entities through OpenFIGI, extract financials, draft diligence questions, cite every source, push into a controlled analyst workspace. Never publish client-facing work without review.

6. Capital Markets and Trading

The open-source stack here is stronger than most people realize.

• QuickFIX is a core open-source implementation of the FIX protocol, versions 4.0 through 5.0 SP2. FIX is the standard for order messaging and connectivity, and QuickFIX is how many teams run it. github.com/quickfix/quickfix

• FINOS Common Domain Model is a machine-executable model for how products trade and settle across the lifecycle, covering derivatives, repo, and securities lending. github.com/finos/common-domain-model

• FINOS FDC3 is the open standard for financial desktop interoperability, letting trader and banker apps share context and launch workflows. github.com/finos/FDC3

• QuantLib belongs here too for pricing and risk. github.com/lballabio/QuantLib

• Backtrader and Zipline Reloaded handle strategy research and backtesting. github.com/mementum/backtrader, github.com/stefan-jansen/zipline-reloaded

Do not confuse a backtester with production trading infrastructure.

7. Risk, Compliance, and Financial Crime

This is where most financial AI products either earn trust or get killed. An agent without controls is a liability.

• OpenSanctions is an open database of sanctions, PEPs, and related entities, with bulk data and a screening API. github.com/opensanctions/opensanctions

• OFAC Sanctions List Service provides the primary-source datasets and a fuzzy-matching search tool. ofac.treasury.gov

• Moov Watchman runs AML, KYC, and OFAC search through an HTTP server, a Go library, and an MCP server. github.com/moov-io/watchman

• FINOS Morphir captures business logic as data so policy and regulatory rules stay explainable and portable. finos.org/morphir

• FINOS Waltz maps the technology landscape, including application ownership, dependencies, and lineage, for resilience and regulatory response. github.com/finos/waltz

8. Insurance

Insurance is an ideal agentic environment. The work is document-heavy, rules-heavy, and approval-heavy, which is exactly where controlled agents earn their place.

• OpenSanctions screens insureds, vendors, claimants, and counterparties. github.com/opensanctions/opensanctions

• FINOS FDC3 carries context across claims, underwriting, billing, CRM, and policy systems on the desktop. github.com/finos/FDC3

• FINOS Waltz maps application ownership and dependencies for operational resilience. github.com/finos/waltz

• Blnk holds premium, commission, claim payment, and receivable state in a real ledger. github.com/blnkfinance/blnk

• Moov ACH and Wire validate payment files and claim payment workflows. github.com/moov-io

• SEC EDGAR APIs surface public-company insureds, broker research, and carrier intelligence. sec.gov

Build this. A Claims Payment Control Agent. Read claim payment instructions, validate routing details, screen payees, check policy limits, flag exceptions, prepare the payment memo, and require human approval before release.

9. Digital Assets and Stablecoins

No longer a separate category. This is becoming another settlement and treasury layer.

• Circle supports USDC payments, payouts, wallets, and cross-chain transfers. developers.circle.com

• Fireblocks provides vault accounts, transactions, wallet infrastructure, and webhooks. developers.fireblocks.com

• Coinbase Developer Platform covers wallets, payments, onchain data, stablecoins, and agentic wallet workflows. docs.cdp.coinbase.com

• Chainalysis sanctions oracle lets a smart contract check whether a wallet address is sanctioned. go.chainalysis.com

Do not pretend every treasury problem needs a blockchain. Most do not.

The killer stack by vertical

Wealth. Plaid Investments, OpenBB, OpenFIGI, PyPortfolioOpt, QuantLib, FRED, SEC EDGAR MCP, OpenSanctions.

Payments. Stripe, PayPal MCP, Square MCP, Plaid Auth, moov-io (ACH, Wire, Fed, Watchman), Blnk.

Treasury. ISO 20022 parsers, Moov Fed and Wire, FRED, Blnk, Circle, Fireblocks, Coinbase, Chainalysis, OpenSanctions.

Investment Banking. SEC EDGAR APIs and MCP, OpenFIGI, OpenBB, FINOS Legend, FINOS Perspective, FINOS CDM, QuantLib.

Capital Markets. QuickFIX, FINOS CDM, FDC3, Perspective, QuantLib, Backtrader, Zipline Reloaded, OpenFIGI.

Risk and Compliance. OpenSanctions, OFAC, Moov Watchman, FINOS Morphir, FINOS Waltz, Blnk.

Insurance. OpenSanctions, FINOS FDC3, FINOS Waltz, Blnk, Moov ACH and Wire, SEC EDGAR.

The hard truth

Most financial AI demos stop at text. Financial services does not run on text. It runs on balances, ledgers, messages, filings, identifiers, approvals, settlement, and audit trails.

That is why these repos, APIs, and MCPs matter. They are the bridge between AI and real finance.

The next wave will not be “chat with your bank.” It will be agents that read filings, reconcile payments, screen counterparties, monitor liquidity, draft memos, and never move money without controls.

The advantage goes to the teams that build the tool layer. That work starts today.

Sources

• Linux Foundation, x402 Foundation launch: linuxfoundation.org/press

• Stripe MCP and Treasury tools: docs.stripe.com/mcp

• PayPal MCP: developer.paypal.com

• moov-io open-source payments: github.com/moov-io

• Plaid developer docs: plaid.com/docs

• OpenBB: github.com/OpenBB-finance/OpenBB

• FINOS projects (Legend, Perspective, CDM, FDC3, Morphir, Waltz): finos.org

• QuickFIX and the FIX protocol: github.com/quickfix/quickfix

• OpenSanctions: opensanctions.org

• OFAC Sanctions List Service: ofac.treasury.gov

• SEC EDGAR APIs: sec.gov

• Circle, Fireblocks, Coinbase Developer Platform: developers.circle.com, developers.fireblocks.com, docs.cdp.coinbase.com

If this was useful, send it to one person on your team who builds. That is how these lists reach the people who need them.

The views here are my own and do not represent my employer. This article is for informational purposes only and is not legal, compliance, investment, or professional advice. I have no financial relationship with the tools mentioned. Tools and their capabilities change quickly, so verify current documentation before you build.

Wholesale settlement scales by getting out of bespoke, name-by-name credit and leaning on standardized, secured, or central-bank-backed money. Judged by that standard, a single-issuer stablecoin is a step backward: a private workaround that leads at the edge only because the more neutral public settlement option is locked by policy. The question was never deposits versus stablecoins. It is who gets to stand on the floor underneath both.

Most of the stablecoin argument happens at the wrong layer. People debate whether stablecoins beat cards or eat into bank deposits. The question that actually lasts is quieter: what asset does the network settle on. Ask it that way and the popular answer starts to look backward. The instrument everyone treats as the future is, by the logic of how settlement has always scaled, a step in the wrong direction. It leads the market anyway, and why it leads is the interesting part.

Networks scale away from bespoke credit

There is a pattern that runs through the whole history of market plumbing. The systems that scale stop relying on credit underwritten name by name and move toward exposure that is secured, standardized, mutualized, or settled in central bank money.

This is not because credit cannot move money. It obviously can, and most of the plumbing is still credit underneath. The narrower point is that bilateral credit, priced one relationship at a time, does not scale across a network. Each link is its own piece of work: a view on the counterparty, the legal terms, the liquidity, the capital you hold against it. Add one participant and you add that work in both directions with everyone already there, so the relationships grow with the square of the players.

The systems that beat this did not abolish credit. They made it less bespoke. Repo turned unsecured bilateral lending into secured, standardized exposure. Central counterparties took credit that used to sit between two firms and mutualized it behind a shared rulebook. CLS cut principal settlement risk by tying payment in one currency to payment in the other. Even card networks, which get misread here all the time, did not mutualize credit. They standardized the rulebook and the interbank settlement and left cardholder credit with the issuers who actually carry it.

The systems that scale don’t remove credit. They make it secured, standardized, mutualized, or settled in central bank money, instead of bespoke and bilateral.

So where does a stablecoin sit

Hold that pattern up against the instrument everyone is excited about, and the answer is awkward.

A single-issuer stablecoin is the opposite of mutualized. It is concentrated counterparty risk in one private company: its reserves, its banking relationships, its operations, its freeze switch. That puts it among the least mutualized things anyone seriously proposes as settlement money. By the same logic that explains why CCPs, repo, and CLS scaled, a single-issuer token is not a step forward. It is a regression.

Not a fraud, and not useless. A regression specifically against the way wholesale settlement has always scaled, which is exactly why a stablecoin is genuinely useful at the edge and structurally imperfect at the core. The cleaner wholesale endpoint is not any single issuer’s token. It is neutral settlement moving at message speed: tokenized central bank money, or a mutualized tokenized clearing layer that stands in the middle the way a CCP already does. A stablecoin is a privately built stand-in for that.

A fair reader will ask the obvious follow-up: if the stablecoin fails that test, does the tokenized deposit pass it? Not really, and not for a different reason. A tokenized deposit is faster bank credit, a claim on one institution, so it carries the same single-balance-sheet limit. Its edge is elsewhere. It keeps the dollar inside the regulated perimeter, which is precisely why most of traditional finance prefers it and why bank-led settlement is moving first there. So the deposit token wins inside the banking system; it is not the neutral layer underneath it. Two private claims, two different jobs, neither one the floor.

So here is the claim the rest of this piece defends. The stablecoin is a transitional workaround, not the destination. It leads not because it won the design argument but because the more neutral public option is shut. The honest test is whether that survives the hardest facts, and there are four of them.

The two instruments, read through the pattern

A tokenized deposit is bank-credit money in a faster wrapper. Its value still rides one bank’s balance sheet, with that bank’s credit, legal, and interoperability questions attached, so it is not mutualized either. It is a claim on one institution. That is the real gate on bank-to-bank interoperability, which still looks hard to scale, and the block is not mainly technology, it is credit. For two banks to settle in each other’s deposit tokens, each has to get comfortable with the other’s credit, legal, liquidity, and capital treatment, and every bank you add multiplies those bilateral relationships. It is the same scaling problem from before, now sitting inside the regulated perimeter. So the deposit token moves fastest where it never has to cross that gate: in-house and bank-led settlement, where keeping the dollar inside one regulated perimeter fits the existing capital and AML treatment. JPMorgan’s Kinexys is the live proof of that. A stablecoin sidesteps the gate by handing the market one more neutral asset to settle on, which is why it keeps landing as the bridge layer, even though it brings its own issuer, reserve, and redemption risks. Slower as neutral cross-network money, faster as in-house bank settlement.

A stablecoin is a reserve-backed issuer liability built to move like a payment. In most structures the holder does not hold the reserves directly. They hold a claim on the issuer, and the issuer holds the backing. What makes a well-built one feel like neutral settlement is not a security interest in the holder’s hands. It is that the backing is meant to be standardized, high quality, and ring-fenced. But that only goes so far. Better reserves make the single issuer safer; they do not make the exposure mutual. You are still standing on one private balance sheet.

Why the weaker design still leads: the door is locked

If the single-issuer token cuts against the pattern, why does it dominate? Because the better option is shut, by choice, not by physics.

Non-banks reach for a private token because they cannot reach central bank money directly. The floor exists; most of the economy just isn’t allowed to stand on it. And there are really two doors here, both closed on purpose. The wholesale door is master-account access to central bank settlement, which stays policy-gated, as Custodia’s denial and the long fight over TNB show. The retail door is a public digital dollar, which is blocked by executive action and a separate anti-CBDC bill, not by the stablecoin statute itself. Neither is an engineering delay. Both are decisions.

With the more neutral public option closed by decision, a concentrated private substitute wins by default. That tells you something about the door, not about the merits of the token.

The model and the market are not the same thing

The clean, ring-fenced model I just described is where regulation is pushing the category. It is not what leads the market today.

The biggest token by float is Tether, around three-fifths of the market on its own, with Tether and USDC together making up the large majority of stablecoin value. Tether’s published attestations show the bulk in Treasuries alongside allocations to secured loans, gold, and bitcoin. It operates offshore and is not authorized under the EU’s MiCA regime. The instrument carrying most of the volume, in other words, is not the standardized, fully ring-fenced asset the clean story assumes.

That does not weaken the argument. It sharpens it. If the leading token wins while holding gold and bitcoin against demand liabilities, and while direct redemption with the issuer is subject to verification and minimum thresholds that put it out of reach for most retail holders, who reach the peg through exchanges and on-chain liquidity instead, it is not winning on reserve quality or neutrality. It is winning on distribution, and on the fact that nobody has shipped a better-built public option. The market leader is the workaround in its least clean form, and it still leads. That is the locked door at work, not the design.

The law is coming, not here, and it reads like a patch

In the US, as of 2025, there is finally a legal standard aimed at the reserve model, and the tense matters, because a careful reader will catch it. The GENIUS Act was enacted in July 2025. It is not yet in force. It phases in on the earlier of January 2027 or 120 days after the final rules land. The agencies are still writing those rules through 2026, and non-compliant tokens stay sellable through a multi-year transition after that.

Once live, the Act sets a one-to-one reserve model in permitted assets such as cash, insured deposits, short Treasuries, certain repo, and qualifying money-market-fund shares. It requires segregation, limits rehypothecation, bans issuer-paid yield to holders, and carves reserve assets out of the issuer’s bankruptcy estate for the benefit of holders. The exact ranking on a shortfall is still being argued, so the honest line is to claim the carve-out, not perfection.

What is striking is what the carve-out is doing. It does the work a security interest would, without the holder ever touching the collateral. The law is making an unsecured claim on one private issuer behave more like a ring-fenced, secured-style payment claim. What it does not do is make the exposure mutual. Every holder is still standing on one issuer; the statute just makes that issuer safer to stand on. You only write a segregation-and-priority regime for an instrument whose core weakness is that holders are unsecured creditors of a single company. The law is the clearest admission that the workaround needs reinforcing.

Funding transformation, plus some real flight

A stablecoin can look like money walking out of the banking system. Part of that read is too clean, and part of it is true.

The reserves do not vanish. They reshape, landing across short Treasuries, repo, money market funds, and bank deposits. To the extent they recycle back as deposits at the issuer’s reserve banks, this is funding transformation, not exit. A deposit from a household or a company is granular and cheap to fund with. Pool a lot of them into one issuer and the balance that comes back to the reserve banks is larger, more concentrated, more rate-sensitive, and quicker to move. The liquidity rules treat it accordingly: a stablecoin issuer’s balance generally lands in the least favorable bucket, non-operational wholesale funding from a financial entity, with assumed outflows near the top of the scale, against single-digit assumptions for stable retail deposits. Same dollars, worse funding.

Some of this is funding transformation. Some of it is real deposit flight. Being honest about which is which is the whole point.

And the part the comforting version skips: where reserves sit in Treasuries instead of cycling back as deposits, it is not transformation at all. It is genuine funding loss for banks, concentrated on the smaller ones that leaned on those operating balances, plus a new structural bid for T-bills. One more concession a treasurer will force, so I will make it up front: the sticky-retail-deposit comfort is thinner than it sounds for the balances in scope here. Insurance caps at $250k, the large operating balances we are talking about are mostly uninsured, and uninsured balances are exactly what ran from SVB in March 2023. The runnable comparison is closer than the clean story admits.

Convertibility is the real bottleneck, and it is single by design

Among bank deposits, par is propped up by a public bundle you never see: central bank settlement, deposit insurance, supervision, a lender of last resort. That backstop is shared and public, mutual by design in a way one issuer’s reserves are not. It is not unconditional, either. Insurance is capped, support is discretionary, and SVB showed uninsured par can break without intervention. But the contrast holds: a deposit’s par leans on a mutual public net, a stablecoin’s par leans on one private issuer.

And that par is not symmetric. Institutions with direct issuer access can redeem at par. Most everyone else exits through secondary markets, where par depends on depth, spread, and confidence. Underneath sits an arbitrage that somebody has to fund on their own book, capacity that is cheapest when nobody needs it and thinnest when everybody wants out at once.

The bottleneck is not moving the token. It is convertibility. Can holders get back to bank money at par under stress, and through how many doors?

The SVB weekend in March 2023 was the proof, and it proved the thesis as much as the risk. USDC did not wobble because tokens stopped moving. It wobbled because part of the reserve was trapped in a failed bank and primary redemption was choked over a weekend, so the secondary market repriced convertibility. The token-transfer layer worked the entire time. The boundary back to bank money was where the single issuer’s concentration showed up, running through a handful of banking access points. A more mutualized settlement layer would not have the same single-issuer seam. The fragility was not a stablecoin accident. It was the predictable cost of standing on one private balance sheet instead of a shared one.

One tension in my own view

Stablecoins are most useful out at the edges of the system: offshore, in weak-currency economies, and for players who can’t get a bank account. Those edges happen to be dominated by the token with the least clean reserves and the hardest redemption access for ordinary holders, where direct redemption with the issuer is subject to verification and minimum thresholds that put it out of reach for most retail holders, who reach the peg through exchanges and on-chain liquidity instead. So the weak spot is widest exactly where the asset is most useful. That is a real problem, not a footnote.

Programmability cuts the same way. The admin controls that make 24/7 settlement possible are the same ones that allow freeze, blacklist, and seizure. That is a feature to a compliance team and a censorable settlement asset to a treasurer. It is a property of single-issuer control, not of tokenization as such.

Where it goes

The contest was never deposits versus stablecoins, and never which token has the slickest features. It is two old questions asked again with new plumbing. What do you settle on, bespoke credit or standardized, neutral backing? And who is allowed to touch the most neutral settlement asset of all?

The non-bank edge already leans toward standardized, reserve-backed money, and it built a private path to the floor because the public one is shut. Banks and regulators pull the other way, toward the tokenized deposit, because it modernizes settlement without leaving the regulated perimeter. If the public door opens, through tiered settlement access, a synthetic arrangement, or eventually tokenized central bank money, the private substitute loses its structural reason to exist at the wholesale core. Distribution, programmability, and offshore reach would still keep it alive at the edge. If the door stays shut, the workaround quietly hardens into the architecture. Which of these becomes the base layer is not settled, and pretending otherwise would be a forecast dressed as analysis.

Infrastructure has a way of becoming permanent long before anyone decides it should be.

The views here are my own and do not represent my employer or any client. This is general analysis, not investment, legal, tax, or accounting advice, and not a recommendation to buy or sell any asset.