New Test for AI Code: Does It Work Like Real Software?

Cognition released FrontierCode on 8 June 2026, a test that measures AI code quality the way professional engineers do. Instead of checking if code passes simple tests, it evaluates whether code is readable, fits well with existing systems, and handles unexpected problems—the real standards software teams use. Each test takes 40+ hours to design properly.