When a high-flying developer arrives, they often rewrite core systems, introduce exotic tools and languages, and reject every pull request. Teams race to keep up but rarely grasp the new code’s structure. When the “rockstar” quits, their messy, opaque codebase lands on someone else’s desk. Getting it to run can take days or weeks, and fixing even small bugs feels like detective work through alien libraries and paradigms. Generative AI brings a new wave of these one-off experts. An LLM floods projects with thousands of lines in minutes, oblivious to existing architecture or team norms. It’ll suggest “best practices” that add complexity without clear benefits, review code with a laundry list of nitpicks, and raise expectations so high everyone else feels slow. Teams lean on AI more, creating tangled code that only another AI can untangle—often leading to unpayable technical debt. You can avoid this by treating AI as a tool, not a lone contributor. Break tasks into small, guided prompts. Insist on code you and your colleagues understand. Slow down when you’re lost, simplify interfaces until they match real needs, and remember that sometimes writing a few lines by hand is faster than wrestling with generated noise. Craftsmanship still matters—and it can’t be outsourced.