Locus Is Building Stripe for AI Agents, and the Timing Could Not Be Better

The Macro: Agents Are About to Have Wallets

Here is a problem that almost nobody is talking about yet. AI agents are getting autonomous enough to do real work. They can book travel, order supplies, hire contractors, and negotiate deals. But they cannot pay for anything. Every time an agent needs to complete a transaction, a human has to step in, enter a credit card, approve a purchase order, or manually transfer funds. That bottleneck defeats the entire purpose of autonomy.

The traditional payments infrastructure was not built for non-human actors. Stripe, Adyen, and Square all assume there is a person on one or both sides of every transaction. Their fraud detection, their identity verification, their chargeback systems are all designed around human behavior patterns. An agent that makes 500 purchasing decisions per day does not look like a normal customer. It looks like fraud.

The crypto world saw this coming before the traditional payments world did. Smart contracts on Ethereum and Solana enable programmatic payments without human intervention. But crypto payments have their own problems. Volatility, regulatory uncertainty, limited merchant acceptance, and a user experience that makes enterprise finance teams break out in hives. Stablecoins like USDC helped with the volatility issue, but the rest of the friction remains.

A few companies are circling this opportunity. Skyfire is doing agent-to-agent payments. Payman is building payment tools for AI. Nevermined focuses on AI commerce infrastructure. But none of them have established dominance, and the market is early enough that the winning approach has not been determined yet. The question is whether agent payments look more like traditional fintech, more like crypto infrastructure, or something entirely new.

What I find compelling about this moment is the convergence. Agents are getting capable enough to need payments. Stablecoins are mature enough to enable them. And enterprise companies are starting to deploy agents that actually need to transact. Six months ago this was theoretical. Today it is an engineering problem with real demand behind it.

The Micro: Coinbase Meets Scale AI

Locus gives developers an API to connect their AI agents to funds. You define spending limits, set permissions per agent, create approval rules, and get a complete audit trail of every transaction. The agent gets an identity with budget constraints. It can spend up to its limit on approved categories, and every payment is logged with the reason the agent made the decision. Currently it runs on USDC on Base, with ACH and wire transfer support on the roadmap.

Cole Dermott and Eliot Lee are the founders. Cole was at Coinbase building B2B payment products, which means he understands both the crypto infrastructure and the compliance requirements that enterprise buyers care about. Eliot was at Scale AI maintaining data pipelines for major AI labs, which means he understands how agents actually work in production. That combination is almost suspiciously perfect for this specific problem. You need someone who knows payments and someone who knows agents. They have one of each.

They came through Y Combinator’s Fall 2025 batch. The product is live and they are onboarding beta users with $10 in free USDC to experiment. The demo they have built is illustrative: a micro-task payment system where agents evaluate work submissions, score quality, and distribute payments automatically while respecting spending rules. That is a real use case, not a contrived demo.

The policy enforcement layer is the interesting part. You can tell an agent “maximum $50 per payout, $500 per day, must include justification.” The agent operates within those constraints autonomously. That is exactly what an enterprise finance team needs to see before they will let an autonomous system anywhere near company funds. Without those guardrails, agent payments are a nonstarter for any serious business. With them, the conversation changes from “absolutely not” to “let us run a pilot.”

I keep thinking about the Stripe analogy. Stripe won payments not because it processed transactions better than anyone else, but because it made the developer experience so good that every startup defaulted to it. Locus has the same opportunity in agent payments. If the SDK is clean enough and the documentation is good enough, it becomes the default choice for anyone building an agent that needs to spend money. Defaults are extremely powerful in infrastructure.

The Verdict

I think Locus is early to a market that is about to get very big very fast. The number of autonomous agents deployed in production is growing exponentially, and every one of them will eventually need to transact. The team has the right backgrounds, the product is technically sound, and the timing is better than it has any right to be.

At 30 days, I want to see how many developers have integrated the SDK and what their use cases look like. The diversity of use cases will tell you whether this is narrow infrastructure or a platform. At 60 days, the question is transaction volume. Not revenue, not signups. How much money is actually flowing through the system? That is the only metric that matters for a payments company. At 90 days, I want to know whether enterprise buyers are showing up. A startup can build on USDC and stablecoins, but the real money in agent payments is going to come from Fortune 500 companies deploying procurement agents and operations agents that need to transact in dollars through traditional banking rails. The ACH and wire transfer roadmap is not optional. It is existential.