Use case
Migrations across many repos.
Agents apply codemods, run tests, and open PRs with context that survives across long-running sessions.
Why this is hard
The work is boring and the risk is not
Four hundred call sites to update, and three of them have a subtlety that breaks production. Nobody wants the ticket and everybody needs it done.
The change spans more repos than one person can hold
Patterns repeat across codebases, but the exceptions do not. You need something that can apply the same fix everywhere and still stop when the tests fail.
A migration outlives any single session
Across many repositories over many hours, an agent that loses context at the connection drop starts over and contradicts itself.
How a large change gets done
The agent surveys before it edits
It maps every call site and groups them by pattern, so you can see the shape of the work before anything changes.
It does the mechanical bulk
The repetitive cases are applied consistently across repositories, which is the part humans do badly on attempt three hundred.
It flags what needs judgment
Ambiguous cases are separated out with the reasoning attached, rather than guessed at and buried in a large diff.
It verifies as it goes and splits the work
The suite runs against each batch, and changes arrive as reviewable pull requests instead of one unmergeable monster.
What you get back
- Call-site inventory
- Batched pull requests
- List of judgment calls
- Test results per batch
What holds up over hours
Sessions that survive hours and disconnects
A four-hundred-call-site migration stays one continuous job instead of twelve cold starts.
Scoped access across many repositories
One job applies the same pattern everywhere the gateway can reach.
Pull requests sized for human review
Work lands in batches with tests run per batch, not one unmergeable diff.
Who runs this
So we can basically double or triple our development capacity. We processed 300+ client compliance docs, and work we used to estimate at a month by hand now takes about 10 minutes.
Where teams go next
Common questions
Can one agent work across multiple repos?
Yes. Gateway policies can scope access per repo while sharing session context on one computer.
How long can migration runs take?
Computers persist for hours. Reconnect anytime without losing state.
Do agents run tests after each change?
Configure verification steps in your task or job definition.
Can migrations run from a ticket list?
Yes. Connect Linear or Jira and process migration tasks in batch.
What harness works best?
Claude Code, Codex, and Cursor all work. Choose based on your team's preference.