It's 2020. You're the senior engineer, fixing codebase mess after holiday. Fast forward to 2026: you have 13 PRs to review, one is +24506-3938 lines with an AI-generated description. AI removed the speed limit, making weak engineering cultures fail faster. Now engineers ask Claude to explain where data came from, but the project is so convoluted no one understands it.

Fixing bad decisions is hard. Reverting a 25,000-line PR requires migration plans, while more PRs keep coming. A person can generate 20,000 lines in an afternoon, but you still have to understand them. Bad engineers always existed, but now they can produce code at scale without judgment.

In the new AI economy, companies in London or San Francisco pay six figures for engineers who make good decisions. AI pushes salaries apart: good engineers become more valuable, bad ones become much more expensive. If you lack judgment to evaluate LLM recommendations, you contribute nothing beyond what anyone can prompt. The most valuable person on the team is the one who still knows what's going on.