# Creating an agent with AI

Describe the specialist, its inputs, expected outputs, and boundaries; generate the draft, inspect every field, and save only after correcting assumptions.

![Creating an agent with AI in Foxora v5.0.3](/docs/images/v5.0.3/studio/new-agent.webp)

## Where this lives

Open **studio → Agents**.

An agent is a reusable specialist with its own name, role, instructions, model, skills, tools, and optional provider key. Agents can work directly in tasks or as members of a crew.

> **Safety:** Do not place secrets in agent instructions. Add provider keys through the agent model settings and keep tool access limited to the agent’s actual job.

## Steps

1. Open **studio → Agents**.
2. Describe the specialist, its inputs, expected outputs, and boundaries; generate the draft.
3. Inspect every field.
4. Save only after correcting assumptions.
5. **Confirm the result.** Run the same test in a fresh task and confirm the agent follows its role, uses the intended tools, requests the expected approvals, and produces a verifiable result.

## Choose the right path

- Create with AI when you can describe the outcome but not every field.
- Create manually when exact instructions and permissions matter.
- Use a crew when the objective needs several distinct specialists.


## Confirm it worked

- Run the same test in a fresh task and confirm the agent follows its role, uses the intended tools, requests the expected approvals, and produces a verifiable result.
- The screen, command, file, run, or destination named in this guide reflects the expected state.