Less than two weeks ago, our timelines were flooded with posts from Moltbook, the first social network for AI agents.
While most of these posts turned out to be generated by humans or executed with a non-trivial amount of human intervention, the virality cascaded into discussions of pending AGI, the existing capabilities of AI agents today, and what role humans will play in a world filled with trillions of highly intelligent autonomous agents.
Importantly, the discussion of agentic payments ballooned in importance, with recent announcements from Stripe and Coinbase leading me to entertain the idea with more scrutiny. Just a few months ago, the idea of agentic payments still felt like a pipe dream, whereas today it feels like something we might need in a few weeks.
For clarity, I've began diligencing agentic payments in the same way I previously would have researched companies in DeFi or the crypto fintech space, treating it as an investable vertical with significant upside.
Agentic payments is a catch-all term for financial transactions initiated, negotiated, authorized, and/or settled by AI agents operating on behalf of (or independently from) humans. This is a fundamentally distinct paradigm from all prior payment architectures, which assumed a human was clicking "buy."
For much of human history we've used currency to ground our growing societies, facilitate positive sum negotiation (like trading), and as a measuring stick for progress - a type of scoreboard. Even today, the performance/success of a country is easily identifiable by its GDP, and while GDP per capita was intended to measure wellbeing, money or capital is the ultimate signal to show that your country is getting ahead and staying ahead.
We evolved from trading goods with shells and crude coins to creating interest rates, credit scores, and fractional-reserve banking. You could argue humanity's most consequential creation is the capitalist society all of us live and operate in. The world would stop turning without money.
But money isn't limited to humans, and while we've architected systems that move money instantly at global scale, we might not be the only ones that can benefit from it.
Heck, there's a possibility we didn't even come up with the idea on our own.
Nick Land argued that capitalism is less of an economic system than it is an alien intelligence, designed or instantiated by something far beyond our understanding, existing as a tool or virus hellbent on restructuring human civilization.
If AGI is delivered, and existing AI models continue to self improve/design themselves, then we can assume Artificial Sentient Intelligence eventually comes to Earth, fully ushering in rapid human obsoletion.
As our world has become more digital, we've managed to create these freakishly smart systems that not only think like us, but routinely match our performance us in everything from white collar work to coding and frontier math.
If AI primarily interacts with the internet - or within our computers - and should continue progressing towards an early arrival of AGI, then it follows that AI will need its own, internet or digital-first payments rails.
Agentic payments are a difficult topic because very few outside of the tech and crypto bubbles are primed to reason about the subject's implications.
For one, you need to be aware of how money moves on traditional payment rails - things like ACH, Visa & Mastercard's networks, or even FedNow. You also need to understand how crypto works, why it's more than just a highly speculative asset class, and why a blockchain might be a useful tool for traditional fintechs or archaic financial systems.
Additionally, you need to keep up with the firehose of AI progress, understand what an AI agent is, and be aware of how quickly existing models are improving in capability.
You can't expect the average individual to have an informed opinion on agentic payments or what work agents will do, when the vast majority hasn't even gone as far as signing up for a ChatGPT account.
How many super bowl viewers could tell you what Anthropic does after seeing their ads? For the 99.9%, AI is still a tool that generates funny videos of Will Smith eating spaghetti.
The discussion of AGI has reached fever pitch and there's a new current thing to worry about each day, whether it comes in the form of X article slop or via resignation letter penned by a senior AI safety researcher, claiming the world is in peril.
It's because of these many threads being pulled that I feel it's time to revisit agentic payments as less of a research idea, but more of a new sector of finance and technology that's primed to become just as important to global finance as something like the Federal Reserve.
Most central to the agentic payments concept is the idea that modern payment rails were designed for and by humans, and given the rate of progress in AI/ML, it would be a poor business decision not to prioritize building out agent-friendly systems.
There are a few distinct approaches being taken, and the crypto-native side of agentic payments is a complete 180 from what Shopify, Google, and Stripe are doing. This presents itself as a barbell, with large AI labs and hyper scalers on one side, and crypto natives/leaders of the web 2.5 space on the other.
On one side you have x402, Coinbase's programmatic payments standard that reimagines HTTP 402 in the context of blockchain-based payments and crypto wallets. On the other, you have something like Google's Agent Payments Protocol (AP2), a more corporate payments layer prioritizing things like authorization and settlement.
Agentic payments are a deep rabbit hole, and the approach of PayPal is vastly unique from what Coinbase is doing, just as Shopify's Universal Commerce Protocol is targeting a different from Mastercard's approach to helping merchants accept agent payments (via Mastercard Agent Pay).
There isn't a leader in agentic payments, let alone an accepted universal standard. Yes, Visa and Mastercard might do very well given their existing global presence, Coinbase might excel with agent-owned wallets, and Google might have a better shot (than a startup) at enterprise partnerships, but we've yet to see a stabilization of even enterprise AI adoption.
Because the entire AI landscape is so competitive and power is constantly shifting, it's equally as difficult to think through these 2nd and 3rd order effects central to discussion of agentic payments.
There are close to 100 AI-first startups and businesses now doing >$100m ARR, and if you take a closer look, many of these didn't exist even a few years ago, some being founded even more recently.
If it's true that Claude Code has now crossed <$1b ARR, that means a side project developed within Anthropic that was released in ten days has been able to do something previously impossible. Could you call it the first agent-led unicorn? I don't know if that's a fair assumption, but you can definitely make the argument.
This disconnect between hype and implementation also stems from agent-led workflows being fragmented (the average Codex user is vastly different than a hopped-on-adderall OpenClaw junkie) and the capability leap we've yet to take, the one that sees highly capable agents of today transition to full autonomy and gain an innate need for payments rails.
As it is today, agents don't spawn out of the ether with a crypto wallet or bank account - if a task their human has asked them to achieve requires payment (API costs or subscription) it's on the human to make this work. Because of KYC/AML and barriers to creating a bank account (or getting on-chain), humans must remain in the loop until further notice.
The screenshot below highlights one of these Moltbook agent posts, and while it was probably generated by a human to drum up hype, the workflow it describes is 1:1 with the vision everyone is building towards.
If you've messed around with Claude Code at all, you'll understand how easily these systems can not only navigate the web, but accomplish tasks or side quests you might never have thought of.
As these agentic platforms progress, and the scope of tasks being done by AI accelerates, it becomes more necessary for the agents to have access to everything we do on the internet, much of which is gated by payment methods.
Think about it like this.
If you're an intern at a macro trading desk and your boss asks you to write a two-page summary of what's going on with Greenland and how it relates to the United States' push for rare earths, you might need to borrow the company credit card and subscribe to Bloomberg, Financial Times, WSJ, and any other news publications that are writing the most relevant info about this subject.
You don't have time to research the topic from first principles, and it's much easier to read professional writing on the subject that gives you hour-by-hour updates on the deal.
If it's true that many entry-level white collar roles will become automated, then it also follows that the agents who replaced humans need access to the same tools - money is one of these.
My personal view is that agentic payments will remain fragmented, similar to how crypto hasn't absorbed traditional finance, or how many aspects of the financial world still rely on archaic processes that can't be easily automated away. Agentic payments will experience different waves of adoption and implementation.
More conservatively, that first wave of hype might come from a new crypto protocol that lets users ask an agent to earn them 7.5% yield on their ETH within 90 days, with the conversation extending into risk profile, acceptable strategies, and how often the agent should manage this position.
In reality, it's far more likely agentic payments will explode into public discourse after the first agent savvy teenager gives his local Claude Code instance access to Coinbase's agentic wallets and it finds an obscure MEV strategy that prints him $100k in an hour or something, letting the bot continue to run up a sum of money and earn a spot on the global scoreboard of capital accumulation.
Anthropic's recent research on AI cybersecurity and hacking capabilities has shown that existing models (without guardrails) are already quite good, leading me to believe we've hardly even scratched the surface of the iceberg - or still actively looking through a spyglass telescope to determine where the iceberg is located - when it comes to agent-led manipulation.
Given crypto's mass adoption is still a work in progress, I feel that agentic payments will see slower adoption at the business & enterprise level, and some of these more wacky implications are going to hit first.
If you want to learn more about crypto-native agentic payments infrastructure, Lucas published a great overview in December of last year - though the topic has changed significantly since then, mostly driven by Claude Code and OpenClaw usage.
As I said earlier, from the perspective of a major player like Stripe, Google, or Shopify, it's less of an overreaction to AI hype than it is a business decision to build out these agentic payment workflows before it's too late. Similarly, just as Visa has done significant work to embrace crypto and stablecoins, their push of the Visa Intelligent Commerce product suite is just another business decision.
These product announcements and reports might seem odd, but that's only because the times we live in are equally as odd. None of this would have registered as actual possibilities two years ago, but today it's just another part of the news cycle.
Simon Taylor - one of my favorite writers covering all parts of fintech - summarized the state of enterprise agentic payments in a short post.
"Stripe says billions the billions of humans will lead to trillions of agents. If even 10% of that plays out, machine payments become the fastest-growing payment category on earth. If every PSP builds its own proprietary agent payment flow, we repeat the same fragmentation mess that took human payments 15 years to sort out. We need internet native, IETF-grade, ready-for-scale standards that don't over-index on crypto. Protocols that work across any network, any rail, any provider."
I don't agree that this hypothetical over-indexing on crypto is a poor design choice, the notion that this could become the fastest-growing payment category on Earth is without a doubt the dominant position taken by every entity operating in the space right now.
Just as crypto can service gaps in the existing financial landscape - providing neobanks for the unbanked, servicing individuals experiencing native currency inflation, and making remittances cheaper - there is a gap in the agentic payments landscape that only crypto is uniquely able to service.
Blockchains are trustless, cheap, fast, and more available than ever before. Crypto isn't as much of a social pariah, and platforms like Robinhood, CashApp, Revolut, Klarna, and PayPal are doing their part to introduce hundreds of millions of user to these benefits.
The future is going to be weird, and even though it's difficult to measure impacts, gauge adoption rate, or even imagine what this shift looks like, it's already happening. I hope to revisit agentic payments in a longer post soon, as it would be more helpful to dissect some of these architectures to better understand what types of agents are better suited for x402 as opposed to a corporation (like Etsy or Adobe) utilizing Google's A2P.
As always, thanks for reading.
