AI software agents have quietly run more than 138 million USD in payments across Base, the Ethereum layer-2 network built by Coinbase, according to new research from blockchain analytics firm Chainalysis reported by Fortune on October 14, 2026.
By Amir Hassan | October 14, 2026
That number deserves a moment of attention, because nobody clicked a “buy” button to make it happen. These are transactions initiated by autonomous software programs — AI agents — that hold their own crypto wallets, comparison-shop on their own, and settle payments on a public blockchain without a human signing off on every step. If you think of a regular online purchase as you pushing a cart to the checkout, this is more like sending a personal assistant to the store with a prepaid card and a budget. Except the assistant is software, and the card is a crypto wallet.
What Actually Happened on Base
Chainalysis, one of the most widely used blockchain data firms, tallied the total value of payments made by AI agents on Base and found the figure has now passed 138 million USD, Fortune reported. Base is a “layer-2” network — think of it as an express lane built on top of Ethereum that processes transactions faster and cheaper, then settles them back on the main road. Because Base is public and transparent, researchers can watch this activity happen in real time.
The milestone matters because it is one of the first large-scale, measurable examples of machine-to-machine commerce on a public blockchain. Until recently, “AI agents using crypto” was mostly a demo-day talking point. Chainalysis is now counting real money moving through real wallets attached to autonomous software.
Why Blockchains and AI Agents Fit Together
There is a structural reason this is happening on crypto rails rather than through bank cards. A traditional payment card needs a named human holder, a billing address, and a bank willing to underwrite the relationship. An AI agent has none of those things. A crypto wallet, by contrast, is just an address and a key — software can hold it, spend from it, and receive change into it without anyone asking for identification.
That cuts both ways. The same openness that lets an AI agent buy server time or data feeds also means there is no fraud department to call if the agent goes rogue. The transparency of a public chain is the partial answer: every payment these agents make is visible on-chain, which is exactly how Chainalysis was able to audit the total in the first place.
Ethereum, the network Base is built on, currently trades around 2,478 USD according to the batch price snapshot used across this newsroom. Activity of this kind is part of the long-term demand story for blockspace — the more agents transact, the more underlying network capacity they consume, even when the fees per transaction are tiny.
The Core Conflict: Convenience Versus Control
Here is the tension regular investors should understand. Autonomous payments could make commerce dramatically more efficient — software that negotiates its own prices and pays its own bills does not sleep, take lunch breaks, or forget to renew a subscription. But delegation without limits is risky. An agent with a funded wallet and vague instructions could overspend, misprice, or be manipulated by whoever designs it.
The emerging best practice — spending caps, transparent wallet addresses, human checkpoints for large amounts — mirrors how companies give corporate cards to employees: trust, but with a receipt for everything. Public blockchains make that audit trail possible by default, which is arguably their strongest selling point in this story.
What This Means for Your Portfolio
For everyday investors, the takeaway is not “buy tokens related to AI agents.” Most of those speculative bets will not survive contact with reality. The durable signal is infrastructure demand: payments volume from software agents is now large enough that a major analytics firm tracks it as a category, and it is flowing through Ethereum’s ecosystem via Base.
If machine-driven payments keep growing, the networks that host them earn transaction fees and gather developer mindshare — a slow-building fundamental, not a tomorrow-morning price catalyst. Bitcoin, trading around 83,000 USD in the same snapshot, is a different story driven by macro forces; this trend is specifically about smart-contract platforms.
The Verdict
The 138 million USD figure is small next to global payments volume — it is a rounding error next to what card networks clear daily. But every major technology shift starts as a rounding error. Software that pays for itself, literally, is now measurable on a public blockchain, and Coinbase’s Base is where it is concentrating. Watch whether the growth curve continues, whether other chains attract similar agent activity, and whether regulators begin asking who is liable when an autonomous wallet misbehaves. Those three questions will decide whether this is a footnote or the beginning of a new payment rail.
The cryptocurrency market remains highly volatile. This article is for informational purposes only and does not constitute financial advice.
138M and not one of those transactions had a human clicking confirm. the agents are comparison shopping with their own wallets already
most of it is probably micropayments for api calls and data feeds tho. still early but the rails clearly work
chainalysis breakdown said most of the 138M was actual purchases, not just api calls. the comparison shopping part got me, agents checking prices across vendors before settling
the comparison shopping angle is the real story imo. an agent that checks prices across vendors before paying is basically procurement software that cant be talked into a bad deal by a sales rep
agents comparison shopping across vendors before settling is procurement without the golf lunches. that alone justifies the rails imo
procurement without the golf lunches is exactly it. try routing a 0.004 dollar api call through accounts payable, the rails had to be rebuilt before any of this worked
Procurement without golf lunches is exactly it. My old firm took six weeks to approve a vendor, the agents did it in one afternoon.
base doing these numbers right as coinbase pushes smart wallets feels coordinated. make it trivial for software to hold funds and software will spend
@CoinbaseHerren base fees staying near zero is the whole trick. the moment machine payments cost more than a cent each the agents move somewhere cheaper overnight
138M from bots with their own wallets and we still get humans arguing about gas fees. wild timeline
whats stopping a buggy agent from draining its own wallet though. 138M is impressive until one bad prompt loops the same purchase forever
rate limits and per-tx caps mostly. the agents running real volume have spend ceilings hardcoded, the horror story needs someone deploying an agent with an unlimited budget
@Rajesh the fun part is chainalysis can actually audit all of this on chain. try doing that with a corporate expense report lol
@Rajesh Karki the same thing that stops a buggy cron job from spamming emails, spend ceilings and kill switches. 138M flowing means the caps are working so far
spend ceilings and kill switches, same way you contain any daemon. the scary part is most teams shipping agents skip both and find out live
caps and kill switches work until someone ships an agent without them. 138M says most teams did it right, the horror story only needs one team
kill switches sound obvious until you try defining one for an agent that plans 40 steps ahead. containment is still an unsolved problem
the teams doing real volume cap per tx and per day. the horror story needs an intern deploying an agent with an empty prompt and a full wallet, and the guardrails are all opt in
opt in guardrails is the scariest two words in the whole piece. 138M spent cleanly because everyone behaved, nobody made them
per tx caps are why 138M stayed boring. the one team that skips them becomes the case study everyone quotes for a decade
Coinbase building Base is quietly looking like the rails for machine to machine payments. That positioning is underrated.
the comparison shopping stat is the quiet headline. software that reads three price feeds before paying means the cheap vendor wins by default, humans never managed that at scale
138M spent with no buy button clicked and still zero chargeback story anywhere. the rails work, receipts are the missing layer
chargebacks are the wrong frame imo, agents negotiate escrow terms upfront. the missing layer is dispute arbitration and nobody has shipped that yet
138M across how many agents tho. median wallet is probably holding 200 bucks and a subscription to a gpu somewhere lol
ha, true, though chainalysis counted tx value not wallet count. even if it is a few hundred agents that is still real procurement volume with zero humans signing. the gpu subscription line hit close to home tho