Cognition's Devin Uses GPT-6 Astra to Autonomously Test Its Own Code

OpenAI has published a case study showing Cognition's AI software engineer Devin now uses GPT-6 Astra to run end-to-end testing on its own generated code, closing the loop on autonomous software development. Astra handles verification tasks that previously required human review, evaluating outputs across functional correctness and edge cases. This represents a meaningful step toward fully self-auditing agentic systems, where a model generates, executes, and validates its own work without human intervention in the loop. For developers building agentic pipelines, this pattern — using a capable frontier model as an autonomous QA layer — is now validated at production scale. The integration is live and documented through OpenAI's official case study channel.