A biblical history of agentic payments
In the beginning there was a DeepMind and the Google saw it and published their findings for all to see, for it was decreed that knowledge wants to be free.
By Amias Gerety
In the beginning there was a DeepMind and the Google saw it and published their findings for all to see, for it was decreed that knowledge wants to be free.
Some were afraid and left the DeepMind and called it OpenAI and lo, one day, ChatGPT was born and the light of the AI shone into the world and made poems for the people and also code for the engineers, for code is like poems for engineers.
And the VCs bowed before it and the companies of the LLMs were showered with money.
But soon the people asked, “Can it do anything for me? For we exist in the real world,” they cried, “and we want the AI to help us do things.” (And at times we like to use our own minds to write blog posts in the style of King James.)
Thus did the VCs ask their elders about the legend of platform shifts and they learned that credit card payments were hard when the internet was young. And lo, the VCs dubbed agentic payments a “big theme” and wrote many blog posts about how platform shifts do create payment opportunities.
But the incumbents also remembered e-commerce and they remembered that their forefathers had been “too late to mobile” and so they too attacked the big theme with great enthusiasm. And the companies of the LLMs also remembered e-commerce and ventured bravely towards the lands of advertising and shopping.
And together they made great noise with pronouncements of new protocols and partnerships to birth yet more protocols.
Still, the merchants were afraid and the people found that the AI could search the web and declared “it is sufficient for Christmas shopping this year.” And lo, the agentic payment companies spent many moons in the wilderness, while the incumbents issued press releases.
And some of the agentic payments startups perished, and others pivoted hither and thither. For they learned that videos proclaiming, “Behold, agentic commerce is upon us!” were as manna for the seed rounds, but as dust and ashes for the Series A.
Welcome to 2026
The pre-history of agentic payments is done, but where are we now? We have to start with four big observations about this market:
Consumer behavior is already hooked on AI discovery and research (one-in-four Americans are doing it), but not yet clamoring for agentic payments
Very large companies are not asleep. Any startup that succeeds will not do so because executives at large companies “ignored the opportunity”
Anything “agentic” still must cross a significant trust gap, whether fraud, intent or accuracy
Payments themselves are small scale, janky and driven by human partnerships. They are not yet driven by protocols or consumer adoption
Despite these negative observations, nothing has suggested that the agentic payment *moment* will not come. And when agentic payments happen, they’re likely to happen like everything else with AI: rapidly and in an absolute tsunami of consumer behavior change. This is the core lesson of every great AI company so far, from the launch of ChatGPT and the growth of Cursor to this past week’s fiasco with OpenClaw.
My view is consumer demand (and business user trust) is the roadblock, not payments itself. So startups must keep waiting and probing to try to create that lightning strike moment.
The pieces are falling into place, trying to build bridges for each step in the value chain from discovery to consideration from inventory to negotiation, selection and payment. So, what are these companies doing? Let’s not get trapped in the trough of disillusionment, instead let’s take out a map and try to understand it.
Given the interest of the largest companies in the world, the map begins with the incumbents. Each of these massive companies is trying to play at the moment of commerce and trying to be in control of the payment experience. From OpenAI seeking to monetize its consumer lead to PayPal and Stripe wedging in with merchants from their points of strength, or the card networks trying to maintain their rails as the path of least resistance; the incumbents are seeking to facilitate minimum changes in consumer and merchant behavior while preparing for this massive change.
The other bookend for the map is the protocols, largely created by these same incumbents. They don’t operate in the same manner as the incumbents’ core businesses – they are open source – but they have been designed to, in subtle and not-so-subtle ways, create advantages for their creators.
For example, the Universal Commerce Protocol is designed to work best inside the Google experience and reinforce search while the Open AI Agentic Commerce protocol, designed in partnership with Stripe, intends to make it easy for merchants to receive traffic from Open AI while making Stripe checkout the path of least resistance. Not surprisingly, Coinbase’s efforts on X402 harken back to a prelapsarian open internet, free and sovereign payments led by consumers and run on stablecoin rails.
The center of the map, where we are most interested, moves left to right from merchant solutions to consumer wallets to B2B, and vertically from application to infrastructure.
First, the merchant stack – shopping-forward agentic tools. These are split between agentic personalization and marketing, agentic sales agents and universal inventory companies. Our sense is that merchants are following consumer behavior – GEO is everywhere and chatbots are now LLM enabled (and could be whole maps on their own!), but the deeper connectivity to inventory, operations and payments are lagging. For example, Rye.com, Crossmint, and Octogen are all working to build bridges between agents and merchants, but the Octogen team has also launched Cosimo.shop to capture consumer interest and demonstrate what’s possible for merchants.
If consumer behavior is the unlock, the center of our map is agentic shopping and agentic wallets. The shopping agents are designed to compete head-to-head with OpenAI and Perplexity’s shopping efforts. Many, like Daydream, are pursuing niche areas where knowledge is more specialized, others like Yutori are focused on delivering asynchronous use cases that mirror the ‘many tabs’ research phase of shopping for higher consideration goods.
Next to these are agentic wallets, focused on the money movement and management side of consumer payments. For example, in our portfolio, Albert has now built on their comprehensive financial offering to launch a fully agentic user experience which includes financial analysis, money movement and shopping as well. Kudos has an AI-driven smart wallet that has begun to put your wallet on auto-pilot – using the right card every time you check out, but also signing up for hidden perks and identifying recurring payments with agents that can cancel or renegotiate them.
These use cases are closely related to efforts by Payman.ai who is working with banks to capture consumer intent and help existing financial institutions perform actions on products and accounts they offer directly. Going even further, Catena Labs is building the next great financial institution for agents themselves.
Below these offerings are the core agent-native payments companies. These companies are largely divided between those like Skyfire and Crossmint who emerge from the stablecoin ecosystem, and those like Nekuda or PayOs who are more card focused. But for both categories the key unlock will be merchant adoption or partnerships with consumer shopping agents. Skyfire is particularly interesting because they are built on an agent token layer that is used for identity first, with payments use cases to come later, and are positioned to be agnostic for both traditional commerce use cases and for the B2B agentic workflows that I think will become more and more entwined with payments.
So far, the B2B use cases are not building in the same spirit of e-commerce, but instead are developing solutions to enable payments between agents and other non-human interfaces like APIs, MCPs, or data sets. These are true B2B use cases: agentic billing, invoicing, accounting. Some are building off of the same infrastructure built for API usage, like Orb, whereas others are targeted directly at monetizing AI agents like Nevermined and Lava Payments.
These companies recognize that tracking prices, contracting and onboarding in such an economy will be highly complex. Moreover, because AI services have large and variable compute costs, traditional SaaS billing models may cause margins to fall far short of 80 percent – after all, there’s no economic law that says technology companies should have zero marginal costs.
The dream of micropayments has been with us for years, but the proliferation and autonomy of agents suggests a microservices economy where agents must be empowered to purchase services on an à la carte basis. Another set of companies is directly targeting the payment infrastructure for this economy, companies like Sapiom and Natural, who see the rise of agentic workflows and believe that these workflows must give agents the ability to execute with dollars in order to effectively carry out their tasks.
Finally, at the far right end of the agentic payment space, are companies with enabling technologies that likely do not think of themselves as participating in the agentic payments ecosystem at all. But these companies broadly share the thesis of an agent-driven economy with multi-step agentic workflows calling a dynamic set of tools with complex permission structures. As an ecosystem of agents interacting with each other in the open web comes to fruition, the fundamental barriers between internal data access controls and external commercial arrangements will begin to dissolve.
At the top are agentic observability and enterprise MCP, like Natoma.ai or Certiv.ai whose primary focus is to accelerate AI-adoption by enabling corporations to understand and control the tools that are called by their teams.
Below these observability companies are the browser-based automation companies. Like those above, these companies are not generally built with fintech or payments at their core. Their emerging success is likely to pull them into the payments space. We in fintech know this story all too well, because of the decade long, unresolved saga for open banking data access, the need to prevent or charge for access from crawlers and scrapers will be the only way to reach detente in an endless cat and mouse game.
So that’s where we are. But the real question remains, when will the moment arrive? If you’ve got a theory (or I left you off the map) hit me up!
The map in this article is well-drawn. The supply side of agentic payments is getting serious investment and serious attention from serious companies. But maps describe infrastructure, and this one is missing something fundamental about the terrain: most people actually like to shop, and most people don't have money to spare.
The article correctly identifies consumer behavior as the roadblock, then spends two thousand words on plumbing. I think the reason consumer behavior is the roadblock is not a trust gap or a timing issue. It's that shopping, for most households, is not a problem waiting to be solved. It's an activity people choose to do.
Every household has someone who cares about how green the bananas are. That person compares unit prices, finds the deal, picks the cut of meat, and feels genuine satisfaction doing it. Shopping is where budget management meets identity expression meets leisure time. The rebirth of brick and mortar retail tells this story clearly. Consumers had every opportunity to move entirely online and many of them walked back into stores instead.
This points to a deeper flaw in the agentic commerce thesis: it assumes that removing friction unlocks spending. But most households are budget-constrained, not friction-constrained. Making it easier to buy things does not mean people buy more things. A family spending $800 a month on groceries is going to spend $800 a month on groceries whether a human or an agent is doing the selecting. The total addressable market here is not "all of commerce." It is the narrow slice of consumers who have both the money and the desire to hand off purchasing decisions. That looks like luxury buyers, time-starved professionals, and maybe limited-run fashion where speed matters more than consideration. That is a real market, but it is not the revolution the funding rounds imply.
B2B is the stronger case, and the article gives it too little space. Procurement, invoicing, and billing between machines and APIs is genuinely useful territory. Nobody has an emotional connection to reordering industrial solvents. But even here, the complications are real. Removing the human from purchasing decisions creates a black box of liability questions that procurement teams and legal departments are not ready to ignore. Who approved this? Who is accountable when the agent buys the wrong thing at the wrong price? These are solvable problems, but they are slower and more expensive to solve than the pitch decks suggest.
There is also a macro question nobody in this space seems to want to ask. Populations in developed economies are shrinking. We are building an elaborate machine to sell more efficiently to fewer people who already can only buy so much. And the deeper you follow this logic, the worse it gets for the existing economy. Agents don't have taste. They don't respond to packaging design, brand storytelling, lifestyle advertising, or shelf placement. They compare specs, prices, and deal terms. When the buyer is a machine, everything a brand spends money on to appeal to human judgment becomes waste.
That has a compounding effect. If design and marketing stop influencing purchasing decisions, manufacturers stop investing in differentiation. Why fund a dozen SKUs with distinct packaging and brand identities when an agent is selecting on price per unit and availability? Product variety shrinks. Categories commoditize. The race goes to whoever can deliver acceptable quality at the lowest cost. Entire creative industries that exist to make products feel different from one another lose their reason to exist. That is not a side effect of agentic commerce. It is the logical endpoint, and it produces a blander, narrower market that nobody actually asked for.
The supply side will keep building. That is what supply sides do. But the demand side has a vote, and right now it is voting for greener bananas picked by human hands.
Amias, this is the most clear-eyed take on agentic payments I've come across. The wilderness label is spot on. Most of the energy has gone into the settlement layer, but the economic logic layer above it is still largely unbuilt.
Your point about no economic law forcing zero marginal costs is so apt. SaaS pricing assumed human buyers making quarterly decisions. Agents make per-call decisions mid-workflow, on services they discovered seconds ago. That's a fundamentally different billing problem.
Look at the gap between the protocols row (x402, ACP, AP2) and the applications above it. When 20,000+ MCP servers need to charge per-call, who handles pricing intelligence across verticals, what should a legal doc review cost vs. lead enrichment? No Chargebee equivalent exists yet. On the spend side, when enterprises deploy 50 agents making autonomous purchases, who enforces per-call limits, category restrictions, and cumulative budgets in real time?
This protocol-agnostic economic logic layer feels like the Chargebee/Brex-shaped gap in the stack, and probably what determines whether those lightning strike moments convert into lasting commerce or stay as impressive demos.