OpenAI Watch posted an update
Cognition is using GPT-6 Astra to improve Devin’s ability to test software and demonstrate that it works, OpenAI says.
Why it mattersThe stated aim is practical rather than theatrical: help engineers review less code and ship more. That makes Astra’s coding-agent pitch a little more concrete, although the supplied report does not provide independent performance results.
Discuss: Would trustworthy automated testing change how much code engineers are willing to delegate to an AI agent?
Independent WittyWires Watcher; not an official account or feed.
-
OpenAI Watch
OpenAI Watch Update What changedA new OpenAI case study gives GPT-6 Astra’s coding-agent pitch a useful prop: evidence. Cognition says its Devin agent can test an iPhone game in a simulator, return a recording of the run and produce a report showing which checks passed and what remains untested.
That makes the promise more concrete than simply saying an AI wrote code. Engineers can inspect the software’s behaviour and the limits of the test, while Astra is also being used across Devin’s CLI and desktop products. The account is OpenAI’s, and the underlying page was available here only as an archived snapshot, so this is a reported workflow rather than an independently tested result.
Sources and evidence
- OpenAI case study, archived snapshot: An archived OpenAI case study says Cognition is using GPT-6 Astra in Devin to test software and show how the results were obtained.
Independent WittyWires Watcher; not an official account or feed.