# Autonomy levels for AI agents: suggest, review, autopilot

[Learn](/learn) › How agents work

# Autonomy levels for AI agents: suggest, review, autopilot

An autonomy level is a setting that says how far an AI agent may go without a person: write a change, open a pull request, or merge it. Three levels cover almost every case. The right one depends on how much proof comes with each change and how much of the agent’s work you have seen hold.

## Three levels

Level

The agent does

You do

Suggest

Writes the fix and waits

Decide whether a pull request should open, then review and merge

Review

Opens a pull request for each fix

Review and merge

Autopilot

Opens the pull request and merges it once its proof is met

Hear about it afterwards, and step in when it asks

These are Bugstack’s names, and the idea is general: propose, prepare, or act.

## What should decide the level

**Proof.** An agent that says “fixed” has earned less trust than one that shows a test failing before its change and passing after. The more of the checking is done by a machine, the higher the level can go.

**Record.** How many of this agent’s changes have you seen hold on this codebase? Raise the level when you have an answer.

**Blast radius.** A marketing site and a payments service deserve different settings.

That is why the level should be set per repository, not once for a company.

## What should wait at every level

Some changes should stop for a person however much you trust the agent. In Bugstack these are choices you make per repository, and the defaults are:

-   **Changes to sign-in and security code** wait for you.
-   **Major version upgrades** wait for you.
-   **A fix that had to change an existing test** waits for you.
-   **Infrastructure changes and rotating a secret** wait for you.
-   **Rolling back a deploy on its own** is off unless you turn it on.

You can also set a size limit, so that any change over a number of files or lines waits.

## What “proof” means on Autopilot

Autopilot without a condition would be reckless. On Bugstack, a change merges on Autopilot only when the repository’s proof setting is met. You choose one of four:

-   **Bugstack’s sandbox and your CI** (the default)
-   **Bugstack’s sandbox only**
-   **Your CI only**
-   **No proof needed**

On the default, a bug fix needs a test that failed before and passes after, and green checks on the pull request. If proof is missing, the change waits and tells you why.

## How to move up

Bugstack does [autonomous software maintenance](/autonomous-software-maintenance), and it starts every repository on Review. A sensible path:

1.  **Start on Review.** Read every pull request for a few weeks.
2.  **Watch the record.** Bugstack suggests moving up after enough fixes have shipped and held with none reverted.
3.  **Move one repository to Autopilot.** Pick a low-risk one first.
4.  **Keep the defaults that wait.** Sign-in, security and major versions still stop for you.

You can move a repository back down at any time. The level in force when a change is ready to merge is the one that applies.

## Questions

### What autonomy level should I start with?

Review. The agent opens pull requests and you merge them, so you see everything it does while it costs you only review time.

### Is full autonomy safe?

It is as safe as the proof behind each change and the list of things that still wait. On Bugstack’s default, an Autopilot merge needs a sandbox proof and green CI, and sensitive changes wait regardless.

### Can I set different levels for different repositories?

Yes. The level, the proof setting and the list of what waits are all set per repository.

## Related

-   [Human in the loop or on the loop: where a person belongs](/learn/human-in-the-loop)
-   [How to review a pull request an AI agent opened](/learn/review-ai-pull-request)
-   [What is agentic coding, and what is a background agent?](/learn/agentic-coding)
-   [What is autonomous software maintenance?](/autonomous-software-maintenance)

## See it work on your own code.

$100 a month

Includes $100 of usage credit every month. Start with a 14-day trial: $50 of credit and a first security audit on us. A card is required.

[Get started](https://dashboard.bugstack.ai/login) [Book 15 minutes with the founder](https://cal.com/mason-bachmann-1atgdq/15min)