Ulukaya Labs is an independent software studio. We build tools that make AI agents reliable.

Tools for developers who let AI agents change their code. First release planned for early 2027.

• Run contracts

  In development

  Let an agent work unattended against a written finish line: the checks that must pass, an isolated workspace, and a log of evidence for every step.

• Adversarial review

  In development

  Several models try to break a change before it merges. A lead model sorts what they find into act on, consider and dismiss.

• Mistake-to-check

  Research

  When an agent repeats a mistake, turn it into a check that runs as code, and prove the check catches the failures it came from.

• Behavior gate

  Research

  A pre-commit gate that pins baseline tests by hash and rejects any commit that breaks one, whatever the agent reports.