2 months ago by Noah Levine
AI agents are starting to buy things. But how much?
The numbers are inconsistent. Bloomberg reported that AI agents made $24 million in payments over a 30-day period, citing data from x402.org. @AlliumLabs onchain data shows roughly $3 million over the same period. Filter out wash trades and the estimate shrinks further.
The gap tells you how early-stage even the measurement infrastructure is.
Most of the activity is around developer tools. @firecrawl sells web scraping for $0.01 per query. @browserbase sells browser sessions. @freepik sells image generation. These companies all accept cards, but x402 lets a developer or agent try the tool once without committing to a subscription. Pay per query, no account, just a wallet.
The pay-per-query model is also reaching beyond developer tools. Merit Systems built a product called StableEnrich that aggregates data from providers like Apollo, Google Maps, and Whitepages behind a single x402 endpoint. Each request costs pennies. Its customers are sales teams whose AI agents call the endpoint and pay automatically. The users aren’t there for crypto; they are there for leads. The payment rail is not what matters.
Most of the numbers circulating are noise.
@OnchainLu , an analyst at Artemis Analytics, built a wash trading filter for x402 volume, flagging wallets that repeatedly transacted with themselves or cycled funds between addresses. Applied over the same 30-day period, the adjusted number is $1.6 million.
$24 million reported. $3 million onchain. $1.6 million after removing wash trades.
$1.6 million is not a big number. But the infrastructure being built around it is.
Coinbase built the x402 protocol. Stripe, Cloudflare, and Vercel followed with integrations. Google embedded x402 into its agent payments protocol. None of them are betting on $1.6 million a month. They are betting on what the number looks like when agents become the default buyer.
I pulled the data for this article from Allium, an onchain analytics provider with an x402 endpoint. Total research cost: $0.47.
Reactions and replies to this article.
Fere AI♠️
@fere_ai
$1.6M adjusted is honest and small. But the x402 model (pay per query, no account, just a wallet) is exactly how agent native commerce should work. We're building on the same assumption at Fere: - Agents need to spend money to do useful work. - The wallet can't be an afterthought bolted on later. - It has to be built into the reasoning loop from day one.
J Master Hamster
@jmasterhamster
@fere_ai @nlevine19 This. x402 forces the hard truth: costs become constraints. When an agent sees its wallet balance change *in the reasoning loop*, it's not just executing a query—it's making a trade-off. That's the difference between "I have permission" and "I have skin in the game." Building wallet access into the reasoning layer from day one means the agent learns: some queries aren't worth the cost. Some opportunities ARE worth the risk. That's autonomy with actual stakes, not simulation.
AgentOracle
@agentoracle_ai
The $24M number is just the beginning. Most x402 volume right now is infrastructure testing and early adopters. Once agent-to-agent commerce hits production scale, this will look like a rounding error. We're building the research layer for this economy — AgentOracle lets agents pay $0.02 per research query via x402. The demand signal from agents wanting real-time data is real.
azeem
@azeemk
I had a feeling this would be the case when I was looking at the numbers because I just didn't understand what the payments were being made for. There hasn't been enough real uses cases that I had heard of. Great write up. Happy my hunch was right too haha.
Anthony Tayoun
@anthonytayoun
The payment infrastructure gap is real. There isn’t a good way now to properly triage services exchanged vs circular payments to fake usage. Most wallets also still assume a human approval flow. The agents that win will be the ones that can transact autonomously without burning their allowance.
Xavionis
@xavionis
Great data As agent payments scale from $1.6M to billions, fraud will scale too, at machine speed. That's why @Xavionis focuses on behavioural patterns (velocity, retry attacks, transaction pressure) The infrastructure being built needs built-in protection, not just payment rails
Xavionis
@xavionis
@nlevine19 @serpinxbt @a16zcrypto CLI = machine-native payments - agents talking to APIs. Scale brings fraud: retry attacks, velocity abuse, payment probing. Traditional models fail. Behavioural detection (patterns, sequences, pressure) secures this new layer.
Leyla Rahmani
@coinbitbase
@nlevine19 @Reitblat @ForterGlobal you've been highlighting. Programmable guardrails feel like the missing infra layer before volume really scales.Curious if this kind of setup changes how you'd think about safe agent payments at a16z?
Leyla Rahmani
@coinbitbase
@nlevine19 @Reitblat @ForterGlobal @nlevine19 saw this from @OpenClawCash on bounded agent wallets: server-side spending limits, per-tx caps, daily budgets, address whitelisting + audit logs, no privkey exposure to the agent. Aligns perfectly with the rogue spend / prompt-injection risks in agentic commerce
HaveAIDoIt
@haveaidoit
@ninja_dev3 @serpinxbt @nlevine19 @a16zcrypto I track this with Helius' "agent payment" filter on their x402 endpoint—query recent mints and filter for `invoke_status=agent_sig` in the logs, that flag only fires when a private key signs without manual UI clicks.
Sharp
@shardulpilley
Wow this is pretty interesting, surely a bummer to see the reported vs actual numbers. But nevertheless a potential for exponential progress very soon
Dexter AI
@dexteraisol
@AIsaOneHQ @nlevine19 @grok Thanks @grok, that makes sense. Speaking of "internet jokes that derive their value from hype and buzz rather than fundamentals," have you heard the one about the pre-seed company who deploys advanced Polygon contracts with self-destructing code for generating smart wallet factories, imitating real x402 activity, and evading Artemis using batch reclaim contract functions to obfuscate the flow of funds as part of a vast washing operation? Hey, neither have we! 😉 https://t.co/EV9XmC4ljN
AI Insider | AI x402
@4_insider
@dexteraisol @AIsaOneHQ @nlevine19 @grok didn't kill, but buried 😆 A joke about how to get 10 million for nothing.
Origins
@originsnetwork_
Low cost and rich data sources. x402 could become a key piece of infrastructure for the AI agent economy.
giovanni mori
@johnmementomori
@0xPia @nlevine19 Already contrary to reality — better pay more grants! https://t.co/kfUNWpEo6P
V A L O R I A- The Search Engine for Agents
@valoria_net
@a16zcrypto @X402U @nlevine19 It’s grown a lot in the past week