Cognition Integrates GPT-6 Astra for Autonomous Software Testing

Cognition has integrated GPT-6 Astra into its product lineup to automate the testing and verification of software development tasks. This integration allows Devin, Cognition's autonomous software engineer, to prove that its code functions as expected by generating test reports and visual recordings of software in action, reducing the need for manual code review.

Automated Testing and Visual Verification

GPT-6 Astra enables Devin to perform end-to-end testing of applications and provide tangible evidence of the results. In a specific use case involving an iPhone game called Otter Run, Devin used Astra to test the game in a simulator and provided two primary outputs:

• Visual Recordings: A recording of the game running in the simulator to demonstrate application behavior.
• Testing Reports: A report identifying which checks passed and which areas remained untested.

These outputs allow human engineers to verify software functionality and identify remaining gaps in testing without manually tracing every line of code.

Accelerating Bug Fixes via Multimodal Input

Cognition is using Astra's capabilities to decrease the time it takes to resolve customer-reported issues. When a customer provides a screenshot of a bug, the team can pass that image to Devin via Astra. Devin then fixes the issue and returns a new screenshot demonstrating the result of the fix.

Reducing Manual Code Review

Cognition aims to use GPT-6 Astra to shift the burden of verification from manual code inspection to evidence-based review. By having the autonomous agent test its own work and provide proof of functionality, Cognition expects to increase shipping velocity while decreasing the amount of code engineers must manually examine.

"We expect over time that we have to manually look at less code and end up shipping more at the end of the day. This is one of the things we’re really excited about when it comes to GPT-6."

— Walden Yan, Co-founder, Cognition