Agentic payments: when AI starts buying everything - who controls money, contracts and commerce?

Written by Reid A. Winthrop, Managing Partner, Winthrop Law Group, PC

For most of its short public life, artificial intelligence has been a very capable assistant. It drafted the memo, flagged the anomaly, summarised the deposition. It advised; people decided. That line is now blurring. The newest AI systems do not just recommend a purchase - they make it: sourcing suppliers, negotiating terms, arranging financing and releasing payment inside limits their owners set in advance. The industry calls this “agentic AI”. Strip away the label and it means something simple and consequential: software has moved from helping companies decide to acting for them.

Agents that negotiate, insure and pay in real time: the makings of autonomous commerce

Source: Teamblockchain

This shift is arriving at the same moment money itself is becoming programmable. Banks and card networks are building tokenised deposits, regulated stablecoins and payment rails that run around the clock instead of stopping for the weekend. Put autonomous agents and programmable money in the same room and you get something new: commercial infrastructure that can negotiate, insure and settle at machine speed. For the executives, insurers and financial institutions I advise, that is not a science-fiction problem. It is a governance problem, and it is arriving faster than most boards expect.

Start with the money at stake. OpenAI, Google, Microsoft, Anthropic and Meta are pouring capital into agents built to handle real commercial work, not party tricks. McKinsey puts the potential annual contribution of generative AI at between $2.6 trillion and $4.4 trillion across industries. Most of the early attention landed on knowledge work, lawyers, coders and analysts, but the same technology is moving into procurement, treasury, logistics and finance where the dollars are larger and the workflows more mechanical. And unlike the software your company already runs, an agent is built to finish a job end to end. A procurement agent can watch inventory, find a second source when the first one fails, compare prices, run a counterparty check, line up financing and pay the invoice the moment the contract conditions are met. It does not automate a task - it automates the whole desk.

Our payment systems were not built for that. They were built around people, banks and business hours. Software keeps no hours. It can weigh a dozen options at once and act the instant its conditions are satisfied, which is exactly why so much money is now chasing payment rails that behave the same way. Visa, Mastercard, JPMorgan Chase and Citigroup have all announced work on tokenised payments and blockchain settlement. The IMF is tracking it; central banks are building wholesale settlement infrastructure to match. The rails are being laid whether the law is ready. And the law is the part everyone underestimates. Technology can move money in a millisecond, but commerce runs on certainty, on knowing that a contract means what it says and that a court will enforce it if it does not. That is not a soft variable. It is the reason capital chooses one jurisdiction over another. The Law Society reported in 2025 that English law governs more than 40% of cross-border commercial transactions, prized for its neutrality, transparent courts and predictable outcomes. England has been doing the quiet spadework, too. The Law Commission of England and Wales has produced serious analysis of smart contracts and digital assets, concluding, sensibly, that existing law already handles most digital commerce, with a few gaps worth closing.

The United States has been doing the same work, and business lawyers here should know it cold. The Uniform Commercial Code now includes a new Article 12 governing “controllable electronic records” - the rules for control, perfection and priority when the asset changing hands is digital - and states are adopting it. In July 2025, Congress went further and enacted the GENIUS Act, the first federal framework for payment stablecoins, giving banks and issuers something they had lacked for a decade: a statute. Add the SEC, the OCC and the state insurance commissioners to the picture, and the American answer to programmable commerce is taking shape, piece by piece, in the ordinary machinery of commercial law. The next Act moving through Congress right now. The CLARITY Act, the Digital Asset Market Clarity Act, would finally settle the question that has dogged this market for a decade and has resulted in continuing volatility. The question of when a token is a security under the SEC and when is it a commodity under the CFTC, may finally be resolved by regulation. The House of Representatives passed the CLARITY Act in 2025. The Senate Banking Committee advanced its own version in May 2026 on a bipartisan vote. As of this writing, the two chambers are still fighting over the details. Whether it clears this year or next, the direction is unmistakable - Washington is drawing the lines around digital assets one statute at a time.

Insurance is the layer almost no one talks about, and it may be the one that decides how fast any of this scales. Insurance built the modern economy by moving risk off the people who could not carry it and onto institutions designed to. Mortgage insurance opened home lending. Export-credit insurance underwrote global trade. Political-risk cover made foreign investment thinkable. The same logic is now aimed at digital commerce, where new products could stand behind programmable payments, tokenised assets and automated performance. I made a version of this case in an article I wrote in November 2025 - Insured Reality: How Blockchain and Insurance Will Redefine Property Ownership. Tokenised real estate stays a novelty until insurance stands behind it, and only then does it become institutional-grade, financeable, legally enforceable wealth. What was true for property is true for payments. Insurance does not make risk disappear; it puts risk where it can be priced and absorbed, and that is what lets counterparties do bigger, more complicated deals than they otherwise would.

Stack these layers - AI, programmable money, digital identity, insurance and enforceable commercial law - and you are not looking at five separate trends. You are looking at the floors of a single building. As it goes up, the work that today takes a procurement team, a treasury desk, an underwriter, a lawyer and a payment processor may increasingly run through software operating inside rules the company sets in advance.

When AI starts paying AI

Here is the part that should get a board’s attention: transactions may start originating from software instead of people. Not employees hunting for suppliers and signing off on wires, but agents doing that work continuously, inside guardrails their organisations define. Think about how a purchase happens today. Purchasing, legal, treasury, compliance and finance each take a turn - vetting the supplier, papering the deal, sizing the risk, letting the money go. Agentic systems compress those handoffs into one motion: monitor inventory, compare suppliers, price the financing, arrange the cover and pay once the conditions are met.

The infrastructure is already here. The banks and networks named above are building the rails, and regulated stablecoin regimes are now live across the United States, the European Union, Hong Kong and the United Kingdom. What was theoretical two years ago is operational today.

Every business makes and receives payments, so the implications run well past efficiency. International trade still moves on documents, reconciliation and trust between strangers. Instruments such as the letter of credit exist to bridge that trust gap - useful, but slow and expensive. Pair programmable payments with digital identity and insurance, and much of that choreography can be verified digitally and settled the moment obligations are confirmed. The same logic reaches project finance, supply-chain finance and infrastructure, where an agent can keep a constant eye on financing costs, premiums, commodity prices and market conditions before it commits. Meanwhile, insurance runs through all of it. The mortgage-insurance playbook, broaden a market by underwriting its risk - is being rewritten for tokenised assets, automated obligations and machine-generated transactions. Even the supervisors see it coming: the International Association of Insurance Supervisors has been pressing regulators to keep their frameworks moving at the pace of the technology.

Autonomous agents also raise a question I find genuinely interesting: what happens to money itself? People pick a payment method out of habit, geography or regulation. An agent will not. It will optimise - for transaction cost, settlement speed, liquidity, interoperability and legal certainty - and it will switch the instant a better option appears. That turns money into a competition for machine preference, not just human loyalty. The winning form of digital money may be whichever one serves autonomous commerce best, retail habits notwithstanding. And that leaves policymakers with questions they have barely begun to answer:

· How do you supervise payment flows that software generates on its own?

· Who is liable when an AI negotiates a binding deal inside its approved parameters and the deal goes wrong?

· What happens to anti-money-laundering regimes when machine-generated payments run into the billions per day?

· How does a central bank even measure monetary activity when a growing share of commerce settles in programmable money moving across multiple blockchains?

Governments are not answering those questions the same way, and the divergence is the story. The European Union’s AI Act takes the cautious route, demanding real human oversight wherever an AI system makes consequential legal or financial decisions. The United Kingdom is going principles-first, with the Financial Conduct Authority feeling out how to supervise autonomous agents without smothering them. The United States is betting on the market for AI itself - the current administration would rather let the technology mature than write a comprehensive rulebook, even as Congress builds the digital-asset plumbing piece by piece, stablecoins first, market structure next. Reading the same tea leaves, the US Consumer Bankers Association has argued that developers, banks, networks, merchants and technology providers should build the guardrails - consumer protection, governance, operational controls - before regulation catches up. Three bets, one experiment:

1. Europe regulates before scale,

2. America innovates before regulation, and

3. Britain tries to split the difference.

Step back from the noise about jobs and productivity, and a deeper shift comes into view. Every economy rests on three things: contracts created, risk gets moved and payments get made. Let software do all three, negotiate the contract, place the insurance, send the money and you have changed the architecture of commerce itself, not just its user interface.

History has a lesson here worth remembering. Financial leadership has almost never gone to whoever had the fastest technology - it went to whoever paired good technology with trusted law, sound institutions and deep, liquid markets. Venice, Amsterdam, London, New York, each won because it was safe to do business there, not merely fast. As autonomous commerce grows up, those old foundations will matter as much as the algorithms. Which is why the questions facing governments, central banks and financial institutions are now strategic, not technical. When agents optimise for liquidity, settlement speed and legal certainty instead of national loyalty, which currencies will they prefer? Which jurisdictions will they choose to call home? If software becomes the largest buyer of goods and services on the planet, how does a regulator watch the money, run monetary policy and keep the system stable? And if billions of machine-to-machine transactions clear every day, are today’s payment systems, rules and commercial codes remotely fit for the job?

The next chapter of finance may be written less by people choosing how to pay than by machines deciding whom to buy from, how to contract, when to settle and which money to trust. The institutions that see this early will not just modernise their payments. They will help write the rules for an economy in which the busiest participants are not people at all.