AI agent Devin now tests its own code

Cognition, the maker of the AI coding agent Devin, has integrated the new GPT-6 Astra model so Devin can not only write code but also test it and show it works. The goal is simple: engineers review less code, and teams ship more.

Previously, the heavy lifting fell on humans — Devin produced code, and a developer had to spend time confirming it actually worked as intended. Now Devin runs its own tests, presents the results, and proves the solution works before handing it off for final review. That removes one of the biggest blockers to trusting AI-written code.

Ask your contractors or in-house team how they actually verify AI-generated code: by eye only, or with automated tests and proof of results. The same idea — an agent checking its own work instead of just producing output — sits behind 1Cifer's automatic checks, which compare documents and periods to catch errors before a person does.