# Delegated vs. agent-owned identity: whose account should an AI agent use?

> When an AI agent needs to log in, it can use your account or its own. How to choose, and how to keep a record of who did what either way.

- URL: https://www.smith.seattlecompute.com/blog/delegated-vs-agent-owned-identity-for-ai-agents
- Published: 2026-10-13 (updated 2026-10-29)
- Type: article
- Topics: AI agent login, delegated identity AI agent, AI agent credentials, AI agent accountability, AI agent own account
- Publisher: The Seattle Compute Company (Smith, https://www.smith.seattlecompute.com/)

> Use delegated identity when the account itself is the asset, like your own profile. Use agent-owned identity when it is interchangeable and you need to know who did what.

The moment an AI agent has to sign in to something, you face a choice most tools skip: whose account is it using?

## Two models

**Delegated identity** — the agent works from *your* account. Choose this when the account itself is the asset: your professional profile, with its network and credibility, or a membership only you hold.

**Agent-owned identity** — the agent gets an account of *its own*. Choose this when accounts are interchangeable and you need to know who did what: a GitHub login, a research-portal seat, a SaaS tool.

| | Delegated | Agent-owned |
| :-- | :-- | :-- |
| Whose account | Yours | The agent's |
| Good when | The account is the asset | Any seat will do |
| Attribution | Needs an agent-level record | Built into the account |
| Revoking | Disconnect the agent from your account | Delete the agent's account |

## The failure mode to avoid

The common pattern is pasting a password into an agent and hoping. Then something happens and nobody can say whether it was you, the agent, or a second agent. That is the real problem, not the login itself.

## What accountability needs in both cases

1. **Credentials stay server-side.** Keep them in an encrypted vault; the agent proves who it is and the broker acts for it.
2. **Every action is tied to an agent and an account.** "Agent X did this, using that account" should be a lookup, not an investigation.
3. **Limits you set.** Caps on what an agent may do from your account, such as how many outreach messages it sends.
4. **A way to stop it and step in.** Halt running jobs, and take over the live screen when a step needs a person.

## Where Smith fits

[Smith](/) supports both models for work that sits behind a login, with the same record of what was done and by whom. [Get started](/signup).

## FAQ

**What is delegated identity for an AI agent?**
The agent signs in with your account. Your password stays in an encrypted vault, and the vault fills in the sign-in on the right site, so the agent never sees it.

**What is agent-owned identity?**
The agent has an account of its own on the site. You can revoke it without touching your own login, and the record shows the agent, not you, took each action.

**Which should I choose?**
If the account carries something personal, like a network or a reputation, use delegated. If any account would do and you want to know who did what, use agent-owned.

**Is an action still recorded if the agent uses my account?**
Yes. Every action is traced to the agent that took it and the account it used, in both modes.

---
Site index for LLMs: https://www.smith.seattlecompute.com/llms.txt