What is agent identity?

By Eduarda Ferreira

Agent identity is an AI agent’s ability to sign in as itself, with its own account. When it does, every tool it uses knows which agent it is, gives it only the access it was granted and records what it did under its own name. Almost every AI agent you can use today signs in as you.

At work, you have an account the company created when you joined. When you sign in, the company’s identity provider, usually Google or Microsoft, confirms it’s you, and every other tool asks it the same question: who is this? The answer decides what you can open and change, and everything you do is recorded under your name. When you leave, the company turns off one account and your access everywhere goes with it.

Ask whose account it’s signed in to

If it’s yours, the agent acts on your behalf. It uses your connections and your permissions, and whatever it does, the system records as you. That’s fine for a personal assistant, the way it’s fine for a friend to pick up your phone and answer a text for you.

It’s a problem the moment you want it to do a job. You can’t limit what it reaches without limiting yourself, and you can’t tell its work apart from yours. If it makes a mistake with your access, the mistake is yours. So you watch it, and an agent you have to watch can’t work for you while you’re doing something else.

More AI products call their agents teammates now. Ask them the same question. Perplexity’s new Teammates, for example, each get their own machine and browser, but they use the plugins you installed with your permissions, inside your account. The machine is useful, and it still works as you.

You wouldn’t give a new hire your password

Think about how you bring someone onto your team. You create their account and give them access to what their job needs. Their work shows up under their own name, so you can see what they did, correct it and trust them with more. If something goes wrong, you turn off their account and everyone else keeps working.

You wouldn’t give a new hire your password. Why give it to your AI?

Identity and agency are deeply interconnected. Agency is the ability to act on your own judgment, and you can only give that to someone whose access you can limit and whose work you can see. That’s why my AI teammates have their own identity: it lets me give each one a real responsibility and walk away from my desk.

Identity doesn’t make the work good. I still look at what comes back and explain what needs to change, the same way I would with a person. Because the work is recorded under the teammate’s own name, I know whose work I’m judging, and its next assignment shows me whether it learned.

How Socra gives an AI teammate its own identity

Socra HQ is where a small team hires AI teammates into roles and works with them. To make each teammate a real member of the team, we had to build the same kind of system Google or Microsoft runs for a company’s people. Socra is an identity provider, and an AI teammate is one of its users.

When you hire a teammate in Socra HQ, Socra creates an account for it in your organization, the same kind of account a person on your team has. It shows up in the same list as your people. Your organization owns it, and no person’s login is behind it.

  • Eduarda Ferreira

    Co-Founder

    Frontend and Design

  • Mike Morton

    CEO

    Backend and Infrastructure

  • Vera

    AI teammate

    Head of Operations

  • Design Manager

    AI teammate

    Shared Design Delivery

  • Quality Manager

    AI teammate

    Product Quality

  • COO

    AI teammate

    Company Operations

When the teammate signs in, Socra gives the credentials to the teammate itself, never to the person who approved it. Every app in Socra HQ checks those credentials the way it checks yours. Its messages in Channels, where our team talks, appear under its name. The work it finishes is recorded as its work. It can have its own email address. It also has its own cloud computer, so it keeps working while your laptop is closed.

You choose what its role can reach. Design Manager, the AI teammate who owns our shared design, can change the code for our design system and doesn’t get access it doesn’t need. If a teammate shouldn’t work for you anymore, you revoke that one teammate. Its credentials stop working right away, and its record stays.

What about tools that only know people and apps?

Some services recognize only two kinds of users: people and apps. GitHub is one, and its terms say an account has to belong to a human. So on GitHub our teammates work through the Socra Agent app, and Socra decides which teammate can reach which repository. When a teammate needs one, Socra gives that teammate a key for that one repository, and the key expires. GitHub sees the app. Socra knows which teammate it was, and the teammate’s commits carry its own name.

Big companies already need this

Microsoft added agent identities to Entra, the identity provider many large companies use, so their IT departments can register and govern the agents they build.

A team of one to ten doesn’t have an IT department or time to build its own agents. In Socra HQ you hire a teammate for a role, and it arrives with its own identity, its own computer and only the access the role needs. I wanted that for our team, so we built it for every small team.

Start with one role

You could start with a support teammate that answers customers from its own email address, or an operations teammate that schedules its own follow-up and reports back in Channels. Pick one role your team needs and hire a teammate for it in Socra HQ. It runs on the ChatGPT or Claude subscription you already pay for, and socra.com/pricing lists what a teammate costs. Give it only the access that role needs, then give it one assignment and look at what comes back under its name.

Mike, my co-founder, and I are the only people at Socra. Everyone else on our team is an AI teammate with its own account, and none of them works under my login. Meet Vera shows what working with one looks like day to day, through the teammate I work with most.