How it works · written for your engineering lead
One pipeline, any author, the same verified output.
A developer, an agent, or a one-click fix all travel the same road. Adopting more AI never creates a second path that nobody is watching.
Three kinds of author. One set of rules.
A developer
Writes the code the way they always did, then goes through the same gates as everything else.
An AI agent
Drives turn by turn through the pipeline. It can do the work. It cannot approve the work.
The auto-fix button
One click on a triaged issue. Same plan approval, same gates, same human sign-off at the end.
From a roadmap line to a verified change.
A roadmap line becomes work
Releases hold features, features become executable goals in one click. Nothing starts life as a message in a chat thread that somebody has to remember.
A person approves the plan
No work runs until the plan is approved. This is enforced in the interface, the API, the database row policy and the daemon, so there is no path around it and no flag that skips it.
The work runs in isolation
One git worktree per session, ports assigned per developer, evidence streamed out as it is produced. Five sessions can run at once without stepping on each other.
Gates close one at a time
Each gate writes what it did. A failing gate turns red and the task stops there rather than sliding through on the strength of the other eight.
A person looks at the screenshot
The last gate is human. Approve, or reject with a reason that goes back attached to the task.
Promotion is deliberate
A human promotes to main, and production only ever receives the exact commit that was verified. The issue flips itself to retest and the reporter gets an email.
What you get back that you did not have before.
An answer to the question that follows every incident. Not a reconstruction from memory and chat history, but the actual record: which tests ran, who looked at the screenshot, what they said when they sent it back, and which commit went to production.
DevFactory is in private beta.
We run it on our own builds before we run it on anyone else's. If you are paying for AI tools and cannot tell what you are getting back, tell us what that looks like from where you sit.
Request access