AI can build and maintain software at a professional level

Can it debug unfamiliar repositories?

Tests whether the system can localize faults and patch them in codebases it has not seen before.

Repository debugging is improving quickly, but still struggles on architecture-heavy bugs and subtle multi-file regressions.

Can it refactor without breaking behavior?

Measures whether code changes preserve external behavior instead of only improving local style or structure.

Refactoring quality is often good when tests are strong, but behavior-preserving change remains unreliable on messier codebases.

Can it design reasonable architectures for real requirements?

Tests whether the system proposes structures that reflect real tradeoffs, not only textbook decompositions.

Architecture proposals are often credible for small and mid-sized systems, though still need human judgment for long-term fit.

Can it ship and maintain a working small application?

Measures whether the system can carry a real app through build, release, and follow-up maintenance loops.

Shipping small apps is feasible in demos, but dependable release hardening and maintenance still require heavy human ownership.