Updated
Updated · OpenAI · Sep 11
Cognition Deploys GPT-6 Astra to Test Devin Code, Aiming to Cut Manual Review
Updated
Updated · OpenAI · Sep 11

Cognition Deploys GPT-6 Astra to Test Devin Code, Aiming to Cut Manual Review

3 articles · Updated · OpenAI · Sep 11

Summary

  • Cognition is using GPT-6 Astra to let Devin test its own code and return proof that changes work, as the company tries to speed software review across its products.
  • Astra’s role is to run tests and show results: in one example, Devin tested the iPhone game Otter Run, then produced a simulator recording plus a report on passed checks and untested areas.
  • The same setup is helping customer support move faster—when users send a bug screenshot, Devin can use Astra to fix the issue and send back a screenshot of the result.
  • Cognition says stronger automated testing should let engineers inspect less code manually over time, helping the company ship more software with the same review process.

Insights

If an AI writes the code and tests it, who catches the blind spots the AI doesn't even know exist?
Could visual proof generated by an AI create a dangerous false sense of security for developers reviewing critical systems?
As AI takes over routine code verification, will human engineers lose the fundamental skills needed to spot catastrophic architectural flaws?