AI Agents Need Their Own Email, Wallet, and Identity

01 /The Concept

Locus is building a platform that enables AI agents to make payments autonomously.

The underlying insight is that AI agents are more like people than like software. Traditional software is a tool that people operate. AI agents execute tasks independently. And like people operating independently, AI agents will regularly need to pay for things – access to information, third-party services, or other agents' work – in order to complete the tasks they've been given.

The major players see this coming. Google and OpenAI are both developing payment protocols for AI agents. But both are focused on B2C scenarios: an AI agent buying products from an online retailer or booking travel on behalf of a user.

B2B payments – the transactions between businesses that represent the bulk of money moving through the global economy – have received far less attention. Yet AI agents will be just as involved in B2B procurement and vendor payments as they will in consumer purchases. And B2B payments come with substantially stricter requirements around security, compliance, and audit trails.

Locus is building specifically for that B2B use case.

Every AI agent on the platform has a defined corporate identity. Each agent is bound to a policy that governs its payment behavior: what it can spend, on what categories, and what approval chains are required depending on transaction size and purpose. B2B payments become fully controllable and auditable.

The platform supports any AI agent – internally built or third-party – connecting via API to the corporate payment and authorization infrastructure.

Locus's explainer video uses a concrete example:

- An AI agent is assigned to produce a post or ad campaign and reaches out to freelancers – human or AI – for proposals. - Human approvers select from the proposals according to a defined approval workflow. - The AI agent assigns the task, receives the deliverable, and initiates payment. - The payment is checked against the agent's policy, an invoice is generated, the payment is executed, the transaction is recorded in the accounting system, and the invoice is marked paid.

The platform is early – Locus recently entered Y Combinator and published its launch post on the YC site only a few days ago.

02 /Why It Matters

A week ago, Locus's founder asked Sam Altman directly: "Should AI agents be able to pay for things on their own?" Altman's answer: "Yes, of course."

As AI agents become genuinely autonomous actors, they will inevitably need to transact – paying for information, services, and the work of other agents. A self-contained "machine economy" will emerge in which AI systems transfer money to each other and to humans. That economy is embryonic right now. Which means now is exactly the right time to build the infrastructure.

Locus's claim about the B2B market's scale turned out to be well-supported. The projected global B2B payments market for 2025 is $97.88 trillion, versus $7.5 trillion for B2C (roughly equivalent to the entire B2C e-commerce market). Even narrowing to B2B e-commerce payments specifically, the market reaches $25.65 trillion by 2028.

B2B payments for AI agents might genuinely be the larger opportunity.

The space is already attracting competition. Skyfire ([related review](/review/a-teper-nauchi-ih-platit)) raised $9 million (including $500K from a16z after the original review) on a similar AI agent payment platform. Payman ([related review](/review/kogda-ne-my-im-platim-a-oni-nam)) raised its first $3 million in the week after founding – a notable pace – and has since closed two additional rounds, one for $800K and one undisclosed.

03 /Opportunities

Agent payment infrastructure is one component of a broader wave: building the full infrastructure stack for autonomous AI agents.

AI agents need what people need. People have email addresses – AI agents need them too (AgentMail, [related review](/review/vygodnee-postroit-infrastrukturu), is building exactly that – a Gmail-equivalent for AI agents, including the authentication flows required to sign up for online services). People host services on servers – AI agents need dedicated hosting and testing environments (Blaxel, [related review](/review/vo-vremja-zolotoj-lihoradki-lopaty-luchshe-dazhe-ne-prodavat-a-proizvodit), raised $7.3 million for that). AI agents need access to information – Dappier ([related review](/review/kak-zabit-sebe-mesto-v-novom-internete)) built a marketplace where agents (and others) can pay per-transaction for content from media sites. Agents need to act on behalf of their users on sites that lack APIs – Anon ([related review](/review/bez-jetogo-oni-nichego-poleznogo-dlja-tebja-ne-sdelajut)) has raised $14.9 million on that problem.

This is clearly not yet a complete map of the infrastructure agents will need. But the moment when agents need all of it is approaching quickly. Building now means being positioned when demand arrives.

The clearest entry angle: pick one infrastructure layer that agents will need – identity, payments, compute, browser access, data access – and build the B2B-grade version. The constraint to name is control and auditability: consumer-grade infrastructure won't pass enterprise procurement. That gap is where the actual market is.