Cognition is integrating GPT-6 Astra to enhance Devin, its software engineering agent, according to OpenAI. The stated improvement centers on Devin’s capacity to test its own software work and provide evidence that the code functions as intended.
The described objective is to lower the volume of code that human engineers need to review, allowing teams to ship more.
Why it matters
Code review is a common bottleneck in software development. If an agent can reliably test its own output and demonstrate that it works, that could shift how much manual verification engineers perform.
Who should care
Software engineering teams and developers using or evaluating autonomous coding agents may find this relevant, particularly those weighing tools that automate testing and review workflows.