Why CS Education Must Burn Down and Let AI Change the Game

The refactoring feat. Google's 250 million line C++ monorepo was cleaned by a small team using compiler-assisted AST transformations, not AI. This remains the largest refactoring in history. The lesson: deterministic tooling (like OpenRewrite) is cheaper and more reliable than AI for large-scale code changes.

The education program. Google's readability program, which Titus oversaw, was a massive mentoring effort that taught idiomatic code through review. Tens of thousands participated. This shows that code consistency and education can scale through process, not just tooling.

Where AI adds value. Titus argues that productivity metrics (code output, features) miss AI's real impact: enabling teams to tackle harder, previously impossible problems. Example: writing a JPEG parser in weeks. The value isn't doing the same things faster but changing the game.

CS education crisis. Most CS curricula haven't changed since 2000. With AI, students can produce slop without understanding. Titus advocates grading prompts and tests instead of code, and shifting from implementation to specification, design, and validation. The middle of the bell curve is collapsing.

Team dynamics endure. Psychological safety and learning culture remain the strongest predictors of team success. AI doesn't change that. In fact, the shift to managing agents makes communication and clarity even more critical.

The path forward. Titus suggests CS enrollment may drop, replaced by software engineering programs focused on practice, teamwork, and domain expertise. The advantage is no longer raw coding speed but deep problem understanding and the ability to evaluate AI outputs.