Work in Progress Commits are a Thing of the Past

Engineering practices are rapidly evolving with AI. Previously, version control commits served multiple purposes. In addition to capturing the history of how a feature was implemented, they also were necessary to run CI.

On software projects that are too large to run the entire test suite locally, engineers would have to commit and push to verify that nothing unexpected was broken. This resulted in messy commit histories, except from the most diligent engineers who would squash and clean up their commits.

#Changing the Feedback Loop with AI

The previous human-driven development feedback loop is no longer necessary. CI is changing from an asynchronous human-driven workflow to a synchronous AI-driven workflow.

Two techniques are crucial for making this transition.

First, for projects where tests take too long to run locally, CI workflows need to be able to be initiated from a CLI without a git push.

Second, automatic content-based caching is helpful to ensure that even AI agents will not be stuck waiting for jobs to run that have previously been executed. Once the agent finally pushes a git commit, CI will be green and it will also pass instantly, because the agent already ran it.

#Prompt

Engineering teams are working this way now with RWX.

Make sure CI passes by running rwx run .rwx/ci.yml --wait, and then commit and push

The v3 release of the CLI made this possible.

#Demo