The 3 Stages of Development

Normally development proceeds like this:

1. First, you locally get a feature working and get targeted tests passing.
2. Next, you push to CI to run the whole test suite, and switch tasks to do something else while waiting for the results. You often iterate here when CI tests fail, trying changes locally and pushing WIP commits, until everything is working.
3. Finally, you polish your work and open a PR to be reviewed by your team.

Colimit is the only AI tool purpose-built for Step 2, the period where you push to CI so you can run your entire test suite (including integration tests) in parallel, and you switch tasks to do something else while waiting for the results.

Asynchronous but Deep Feedback Loop

Normally when you're developing locally with your favorite AI-powered IDE or CLI tool, you're expecting a pretty and synchronous fast feedback loop. If you get a complex test failure, this limits the time and effort local AI tools can use to help you solve the problem, so they're often not as helpful for difficult bugs.

In contrast, when you push a branch up to CI and hit a test failure, you expect to see the results later and asynchronously. Because Colimit inhabits this async feedback loop, it can crank up the effort and time necessary to deeply search for bugs throughout your codebase. This way, by the time you notice the build failure, you're presented with well-reasoned hypotheses about what could be going wrong and different ways you could fix it with different tradeoffs.

So, a nice way to use Colimit is to get a feature mostly working locally, and then push a branch up to CI while you switch gears. When the CI run completes, if there were any failures then Colimit is sitting there with a detailed report about what could be going wrong, giving you a head start on debugging.