I keep coming back to the same awkward question. If money can already move in a second on a phone, why would a central bank spend years asking whether an AI agent should be allowed to spend it? The short answer is that the European Central Bank is not trying to look fashionable. It is trying to find out, before any launch decision, whether a future digital euro can handle tiny transfers, machine-to-machine billing and app features that current card rails handle poorly or not at all.
That work is now public. Companies, payment firms, fintechs, merchants, public bodies and researchers can apply to join the next stage of the innovation platform. Workshops and prototype builds are scheduled for 2027. A separate twelve-month pilot with thirty-six payment providers is also lined up. None of this is a launch. It is a controlled look at what might be useful if lawmakers finish the legal framework and the Governing Council later says yes.
What The New Digital Euro Test Actually Covers
The announcement splits the research into two tracks. One is exploratory. The other is practical. I like that split more than I expected. Too many public projects collapse workshops and live builds into one messy room. Here the bank is saying, in effect, talk first about AI and machines, then build receipts, multi-party splits and conditional flows that can be shown on a screen.
Workshops On AI Agents And Tiny Payments
The exploration program will look at how AI could interact with a possible digital euro without promising that any of those ideas will ship. That caveat matters. Headlines love the phrase “AI-powered payments.” The bank is more cautious. It wants people in a room in Frankfurt to argue about use cases for consumers, merchants, firms and public services.
Among the topics: agents that talk to payment systems, agents that talk to other agents, micropayments that are too small for some legacy fees, and machine-to-machine exchanges. The bank has not published ticket sizes. It has not published an authorization model. In my view, that is the honest order of operations. First ask whether the idea is even worth the privacy cost. Then worry about the protocol.
Participants are expected to treat privacy, user control and data protection as design constraints, not as a slide at the end of the deck.
Workshops are planned at headquarters in the first and second quarters of 2027. Applications close on 9 November at 17:00 CET. Joint teams are welcome. A merchant plus a processor plus a research lab is the kind of mix the bank appears to want. That is harder to organize than a solo pitch, but it is closer to how a real payment actually happens.
Prototypes, Not Slide Decks
From January through June 2027, the experimentation track asks selected teams to build working proofs of concept in four areas. Electronic receipts. Payments with several senders or recipients. Conditional payments. New features inside payment apps. Teams will file reports. Some may be invited to present. Findings could be published. That last point is easy to skip. Public notes from a central-bank lab can shape vendor roadmaps even if the currency itself is still years away.
- E-receipts that sit next to the payment inside an app, with privacy still in view
- Split or pooled payments when more than two parties are involved
- Conditional execution after agreed checks are met
- App features that go beyond a simple send button
Conditional payments are the piece most likely to be misread. Under the Commission’s draft approach, a digital euro would not be money locked to a shop, a product or a time window. That is not “programmable money” in the restricted-coupon sense. It is closer to “the transfer fires when the contract conditions are satisfied.” I have found that distinction gets lost in social threads within about four comments. It is still the distinction that matters for law and for public trust.
This Is Not The Same Thing As The 2027 Pilot
Easy mix-up. The innovation platform is research. The larger pilot is operations under glass. That pilot is due to start in the second half of 2027 and run about twelve months. It will use a beta version of the digital euro to test real payment functions in a closed setting. Legal tender status will not apply to that beta unit. Staff will try person-to-person sends and purchases with participating merchants.
Thirty-six payment service providers were chosen from more than fifty applicants. The group includes large banks, newer digital banks and processors that already sit on European checkout pages. Nineteen euro-area national central banks will take part with the ECB. Expected scenes include online transfers, offline person-to-person moves, point-of-sale taps, e-commerce checkouts and mobile flows. Merchants for that exercise were still being recruited, with an earlier window for e-commerce and mobile-commerce firms.
| Track | When | What It Tests |
| Innovation workshops | Q1–Q2 2027 | AI agents, micropayments, machine interactions |
| Prototype builds | Jan–Jun 2027 | Receipts, multiparty and conditional flows |
| Operational pilot | H2 2027, 12 months | Beta payments with selected providers |
| Possible issuance | Target readiness 2029 | Only after law and a later Council decision |
Perhaps the most interesting aspect is how ordinary the pilot list looks on purpose. Person pays person. Person pays shop. Phone and till. That is the baseline. The AI work sits beside it, not on top of it. If agents ever spend, they still have to live inside a product people recognize.
Could An AI Agent Pay With A Digital Euro?
Could it? Technically, an agent is just software with permissions. The hard part is not the API. The hard part is who is liable when the agent is wrong, how a user revokes it at 11 p.m. on a Sunday, and how much of the user’s life the agent needs to see in order to be useful. The bank has not said autonomous spending will be a launch feature. It has said the idea is worth a workshop. That is a cooler temperature than the marketing copy some vendors will write around the same news.
Micropayments sit in the same folder. Think of a stream of tiny charges for content, sensor data, parking minutes or machine time. Card economics often break at that scale. A public digital unit with different fee logic could, in theory, make those rails less painful. Theory is cheap. Settlement, dispute handling and fraud still have to work when the amount is smaller than a bus ticket. I would rather see ugly prototypes than elegant white papers on that point.
Machine-to-machine is the flashy cousin. A charger bills a car. A warehouse robot pays a docking slot. A factory tool pays for a certified data feed. None of that requires science fiction. It does require identity for devices, limits, logs and a human who can still say stop. The program is framed as an examination of future possibilities, not a green light for devices with an open wallet.
Privacy Is Not An Afterthought In This Brief
The bank has already talked about cash-like privacy for some digital euro payments, with banks still running identity and anti-money-laundering checks where online rules demand them. That two-speed model is politically delicate. Too much visibility and the public hears surveillance. Too little and crime units walk away from the table. AI makes the tension sharper. An agent that can pay well is often an agent that can infer a lot.
So the brief tells applicants to use privacy-by-design. User control is listed as a requirement, not a slogan. In practice that should mean clear consent, narrow data use, and a way to inspect what an agent did last Tuesday. If a prototype cannot explain a payment in plain language, it is not ready for a public-money experiment. That is my bar. It should be theirs too.
A digital euro is described as retail central bank money used beside notes and coins, distributed by banks and supervised payment firms rather than through ordinary accounts opened at the central bank itself.
Law Still Comes Before Any Wallet On A Phone
None of the testing means households will hold a digital euro next year. Issuance depends on European Union legislation first, then a separate decision by the ECB’s Governing Council. Current planning talks about readiness for a possible first issuance around 2029 if the legal track holds. A parliamentary committee already approved its position on the core proposal, covering issuance, legal tender, distribution, privacy, data protection and technical features. Final law is not the same as a committee vote. Anyone treating 2029 as a booked holiday should sit down.
Distribution design is easy to forget and expensive to ignore. People would not open a basic account at the central bank the way they open a current account today. Supervised intermediaries would pass the unit along. That choice keeps the central bank out of daily customer service. It also means banks and processors have to connect. Earlier work on reusing existing European payment standards was meant to cut that integration bill. Standards sound dull. They are how you avoid a second rail that nobody wants to plug in.
Why Firms Should Care Even If They Hate The Phrase CBDC
Plenty of people in crypto will shrug. Another public coin study. Fine. I still think the application window is worth a look if you run checkout, wallets, receipts or IoT billing. Not because a launch is certain. Because the questions on the form are the same questions your product team will face if any large public unit ever appears: How do you show a receipt without leaking a basket? How do three people pay one restaurant bill without a group chat full of IOUs? How does a condition clear without turning money into a voucher?
- Map your current payment pain to one of the four prototype themes.
- Pair with a counterpart that sits on the other side of the till.
- Write the privacy story before the feature story.
- Budget for Frankfurt workshops, not only remote decks.
- Keep the pilot and the innovation call in separate folders so legal does not mix them.
Joint applications are encouraged for a reason. A pretty agent demo from a lab with no merchant is a toy. A merchant with no cryptography help will stall on the receipt. The bank is fishing for systems, not monologues.
What “Cash-Like” Has To Mean If Agents Get Involved
Cash is still the mental model. You hand over a note. The shop does not receive a biography. Online payments do not work that way, and the bank has not pretended they do. The live question is how close an offline or low-value digital payment can get to that feeling. Add an agent and the feeling changes again. You are no longer only handing over value. You are delegating judgment.
Delegation needs limits that a non-engineer can set. Daily caps. Merchant categories. A kill switch that does not hide in a settings maze. If those controls are weak, the privacy conversation is theater. If they are strong, micropayments and machine billing become less scary. I would rather watch a boring limit screen than a glossy agent that “just handles it.”
Design order that actually holds up: 1. User can see and stop the agent 2. Data collected is the minimum needed 3. Payment logic is explained in plain words 4. Fancy autonomy comes last
Multiparty And Conditional Flows Are The Quiet Revolution
AI will get the clicks. Shared bills and conditions may matter more on day one. Families already split rent. Clubs already pool gifts. Marketplaces already hold funds until a parcel arrives. Doing that on a public digital unit, with a receipt attached, is less glamorous than robot commerce and more likely to show up in a real app.
Conditional execution also maps to public services. Pay after a permit is issued. Release a deposit when an inspection clears. Those flows exist today with emails and waiting rooms. A cleaner trigger is not magic. It is operations. The prototype window is short, six months. Teams that pick one narrow condition will learn more than teams that promise a general-purpose contract engine.
How This Sits Next To Private Crypto Rails
Readers here live in a market of stablecoins, cards on chain and batch settlement experiments. A public digital euro is not a substitute for those tools and is not trying to be a speculative asset. It is meant to be a digital form of central bank money for everyday use. That is a different job. It can still change the competitive field. If micropayments become cheap on a public unit, some tokenized billing ideas lose their edge. If they stay clumsy, private rails keep the niche.
I do not see this as a death match. I see two stacks being stress-tested at once. One stack is supervised and political. The other is faster to ship and louder online. Users will not pick a white paper. They will pick the button that works at the till. That is why prototype quality in 2027 may matter more than slogans in 2026.
Risks That The Workshops Should Not Soft-Pedal
Fraud against agents is an obvious one. Social engineering gets easier when the payer is software. Model error is another. An agent that “optimizes” a household budget can still buy the wrong thing at scale. Concentration risk sits in the background too. If a handful of agent vendors sit between users and the unit, you have rebuilt a platform choke point with a friendlier name.
There is also a simple operational risk: workshop tourism. Firms apply, collect a logo, and never ship a build. The reporting requirement is the antidote. Proofs of concept and written results force a sharper standard than a panel photo. If the bank publishes even a fraction of those notes, outsiders can judge whether the year produced software or souvenirs.
A Realistic Calendar Without The Hype Fog
November 9 is the innovation deadline. Early 2027 starts the prototype semester. The first half of 2027 hosts the AI and public-service workshops in Frankfurt. The second half starts the twelve-month beta pilot with the selected providers. Legislation still has to finish. Only then does an issuance decision even come into view, with a readiness target spoken of for 2029. That is a long fuse. It is also why applying now is about shaping the test, not booking revenue.
In my experience, long public projects punish two kinds of people. The ones who ignore them until launch day. And the ones who treat every consultation as a finished product. The useful stance is narrower. Watch the prototype themes. Watch the privacy language. Watch whether merchants actually show up for the beta. Those three signals will tell you more than any keynote line about the future of money.
What I Would Build If I Had A Seat In The Room
If I were filling the form, I would not lead with a general AI wallet. I would pick one painful micropayment and one shared bill. A parking meter that charges by the minute without a card minimum. A household utility split that posts a single receipt. Then I would wrap both in a control panel a parent could use. No model names on the home screen. No claim that the agent “understands you.” Just limits, logs and a receipt that does not gossip.
That product would not trend. It might survive contact with a data-protection lawyer. For a public unit, that is the higher compliment. Fancy autonomy can wait until the boring path works. If the workshops forget that, they will produce demos that look like 2027 and rules that look like a problem for 2031.
So here is the sober read. The ECB is opening a lab year, not a launch year. AI agents, machine payments and tiny transfers are on the research list. Receipts, splits and conditions are on the build list. A separate beta will test ordinary checkouts with dozens of providers. Law and a later issuance vote still sit on the critical path. If you sell payments in Europe, the useful move is to treat the November deadline as a product question, not a press release. What would you let software spend in your name, and how would you take that permission back?
