Cognition has added GPT‑6 Astra to its autonomous software engineer, Devin, to automate testing and produce verifiable evidence of behavior. Devin is used by companies worldwide—from large banks to tech‑native startups—and as Cognition’s engineering teams produce more code, manual code review has become increasingly burdensome. The company views Astra’s testing and verification capabilities as a way to streamline the review process.
Where Astra is applied
Cognition is deploying GPT‑6 Astra across multiple parts of its product lineup. According to Yan, the improvements touch the core cloud agent that powers Devin as well as the CLI and desktop products. The intent is for the model to not only suggest changes but also execute tests and document the results.
Example: testing the Otter Run iPhone game
In one showcased case, Devin used Astra to test Otter Run, an iPhone game. The system produced a recording of the game running in a simulator and returned a report that listed checks that passed and areas that remained untested.
The recording shows the application’s runtime behavior, while the report documents the scope of testing. Engineers can use those outputs to inspect how the software behaves and identify which parts still need attention.
Faster customer response and fixes
Yan says Astra also helps Cognition respond to customers more quickly. When a customer submits a screenshot of a bug, the team can hand it to Devin via Astra; Devin fixes the issue and returns a screenshot showing the result. This shortens the time from problem report to demonstrable resolution.
Why this matters
Cognition believes stronger testing through GPT‑6 Astra can lead to a more efficient code review process. If Devin can test its own changes and provide evidence of the outcomes, engineers may be able to evaluate modifications with less manual code inspection, potentially enabling faster shipping and quicker responses to customer issues.
