# What is an AI employee? (/docs/answers/what-is-an-ai-employee)

---
type: Document
title: What is an AI employee?
description: "An AI employee owns a defined job from start to finish inside a business's real systems, with a person at the approval gate: the definition, how it differs from AI tools and agents, and when one is worth building."
status: stable
order: 3
generated: { by: "process:claude-code", at: 2026-08-31T06:30:00Z }
verified:
  - { by: "human:rrizwan98", at: 2026-08-31T07:00:00Z }
ksor:
  audience: [public]
  owner: human:rrizwan98
  approval: { by: "human:rrizwan98", at: 2026-08-31T07:00:00Z }
trust_tier: human-reviewed
build_id: sha256:29a7cfca69186e9d9939db2ba7c8df2502e12724e0562328fda366616971799f
dirty: true
---

An AI employee is a system that owns a defined job from start to finish inside
a business's real tools, without a person typing instructions for each task.
Work arrives on its own, gets done the same way every run, is checked against
defined standards, and stops for a human decision exactly where risk lives. It
is a role in a workflow, not a chat window.

## AI tool, AI agent, AI employee: the difference

| | AI tool | AI agent | AI employee |
| --- | --- | --- | --- |
| Who starts the work | You, every time | You, per task | The workflow: work arrives on its own |
| What it owns | One suggestion | One task, end to end | A whole job, run after run |
| Where it runs | In your editor or chat | Wherever you point it | Inside the business's real systems |
| What you type | A prompt each time | A brief per task | Standards once; then approvals |
| Software example | Code autocomplete | "Build me this script", done in one sitting | A delivery run that researches, plans, codes, tests and reports, and waits at three approval gates |

The distinction is ownership. A tool helps a person do a step. An agent does a
task when asked. An employee holds a job: the same defined work, the same
checks, every time, with nobody re-explaining it. What that looks like in
practice is described in [what I build](../about/what-i-build.md).

## The approval gate

An AI employee is safe because of what it does not decide. Pricing, taste,
risk and the final yes stay with a person, at gates written into the job
itself. The system stops there on purpose and waits. The full list of what
always stays human is on [what automates](../method/what-automates.md).

## What an AI employee is not

- **Not a chatbot.** A chatbot answers questions; an employee finishes work.
- **Not unsupervised.** Every run passes its gates; output is checked, not trusted.
- **Not a person replacement claim.** It takes the repeatable share of a job;
  the judgement share grows for the people who remain.
- **Not general.** It does one defined job well. A vague job produces a vague
  system.

## When one is worth building

A job is a candidate when its steps are repeatable, checkable, reachable and
either reversible or gated. Count the job first with
[count your own workflow](../method/count-your-own-workflow.md): the **could
be** number tells you what the employee could own, and the gates tell you
where it must stop. One job first, then the next; the path from one AI
employee to an AI-native operation is the ladder on
[from one task to a company](../method/from-one-task-to-a-company.md).

## Common questions

**Is an AI employee the same as an AI agent?** An employee is an agent that
has been given a job: defined work that arrives on its own, standards to meet,
gates where a person decides. The technology overlaps; the ownership does not.

**How much of a job can an AI employee own?** Count it. In the record's
[worked counts](../method/worked-counts.md), runs from five businesses land
between 64% and 80%, and every one keeps its judgement calls with a person.

**Which job should get one first?** The one you can describe as a checklist
and undo cheaply if a run goes wrong. The [workflow audit](../audit/overview.md)
exists to find that job with you.