Reading code deserves as much help as writing it
CodePilot started as an internal tool at a payments company where four services in three languages shared one release train. The hard part was never typing. It was holding the whole system in your head long enough to change one part of it safely.
So the first thing we built was the index, not the completion: a picture of the repository good enough that a suggestion in one language could know what another language had declared. Everything since has been built on top of it.
The team is fourteen people across six time zones. We publish our evaluation method, we do not train on customer code, and we would rather ship one capability that is right than five that are nearly.
Who builds it
Fourteen people, six time zones, one release train.
Priya Raghavan
Co-founder, EngineeringSpent nine years on compiler toolchains before deciding that reading code should be as well served as writing it.
Tomas Lindqvist
Co-founder, ResearchWorks on ranking: why the second suggestion is so often the one you wanted, and how to make it the first.
Amara Bello
Head of Developer ExperienceWrites the docs, runs the beta cohort, and argues for removing a setting whenever one can be removed.
Kenji Watanabe
Staff Engineer, IndexingOwns the incremental index. Measures everything in milliseconds and is rarely satisfied.
Open roles
We hire slowly and write the job description with the person who will do the work. Send a note through the contact form and mention the role.
Get in touchSenior Engineer, Language Support
Remote: EuropeOwn two of the forty languages end to end: grammar, conventions, standard library, evaluation set.
Research Engineer, Ranking
San Francisco or remoteDecide which suggestion is shown first, and prove the decision with an evaluation set nobody argues with.
Developer Advocate
Remote: AmericasWrite the guides, run the workshops, and bring back the complaints that never reach an issue tracker.