Cognition Enhances Devin's Self-Testing and Code Review Efficiency with GPT-6 Astra

Cognition has introduced GPT-6 Astra to improve the testing and verification capabilities of its autonomous software engineer, Devin. The goal is to reduce the code review burden on engineers and increase deployment speed.

Self-Testing and Result Verification

Cognition is applying GPT-6 Astra across its entire product lineup, including Devin, CLI, and desktop products. Devin uses Astra to run the iPhone game 'Otter Run' in a simulator and generates reports that include passed checks and unverified areas. This process provides recorded videos demonstrating the application's behavior alongside reports documenting the test scope, enabling engineers to clearly understand how the software works and what tasks remain.

Customer Support and Code Review Efficiency

When customers send bug screenshots, Devin uses Astra to fix the issues and returns result screenshots, thereby speeding up customer support. Walden Yan, co-founder of Cognition, stated that by testing its own work and providing evidence, Devin is expected to reduce the time engineers spend manually verifying code, allowing them to deploy more features.