Factory AI

Factory runs software tasks through Droids that plan changes, work across a repository, and return code for engineers to review.

Where Droids earn their keep

Factory is aimed at issue-sized engineering work rather than next-line autocomplete. It performs best on scoped maintenance, test coverage, refactors, and feature requests where expected behavior is clear. The result is more like delegated implementation than an extended chat session, although architectural judgment still belongs with the developer.

A practical working loop

1. Describe the outcome: provide acceptance criteria, relevant constraints, and any project-specific instructions.
2. Check the proposed approach: correct mistaken assumptions before the Droid starts making broad changes.
3. Inspect the result: review the diff, test evidence, and reasoning before merging anything.

Usage and team fit

The free tier offers a practical way to test Factory on a real repository. Paid subscriptions increase usage allowances, while enterprise arrangements cover larger teams, governance needs, and centralized administration. Actual consumption varies with task length, repository size, and model activity, so monthly cost is not always obvious in advance.

Factory makes the most sense for teams with established tests and review practices. Vague tickets can produce unnecessary edits, and large unfamiliar codebases may need careful instructions before Droids consistently choose the right patterns.