← Blog

We spent two years making agents capable. They can reason, browse, write code, call tools. The brain is here; the hands are here. Yet most agents still can’t do the boring real-world things a new intern does on day one — sign up for a service, hold an inbox, get verified, act under an account that’s theirs.

That gap isn’t intelligence. It’s identity.

The missing primitive

A human operating online has an identity stack they never think about: an email, a phone number, a set of logins, a reputation that follows them. Agents have none of it by default. So today we bolt the agent onto our identity — our API keys, our accounts, our credit card — and supervise it closely, because anything it does is indistinguishable from us doing it.

That works for demos. It doesn’t scale to agents that operate continuously, in parallel, on their own.

What an agent identity layer looks like

The pieces are unglamorous and concrete:

• Provisioning — give an agent its own identity in one call.
• Sign-up & verification — let it register for the services it needs.
• An inbox — somewhere to receive the confirmations and replies that real workflows depend on.
• Action — let it post, reply, and transact as itself, with an audit trail.

None of this is AGI. It’s plumbing. But it’s the plumbing that turns a capable model into something that can actually operate in the world.

This is the thread I keep pulling on. More in the log.