A software developer warns that AI coding assistants create a dangerous fork in the road. One path uses AI tools to accelerate experimentation and deepen understanding. The other leads to building systems of AI slop—code you don’t understand. The author fears this shortcut traps engineers, preventing the experiential learning needed to grow and solve harder problems over time.

To escape this trap, the author advocates treating AI as a tool for understanding, not a replacement for thinking. Key principles include throwing away initial AI-generated code to build fresh, being highly opinionated about problem breakdowns, and writing final documentation manually. This approach prioritizes deep system knowledge over simply getting code shipped quickly.

A practical workflow emerges for medium-sized problems. Start by researching and prototyping messily to get oriented, then discard that work. Next, design the correct structure using human insight, focusing on APIs and libraries. Finally, implement the solution using a skeleton and iterative commits, ensuring you write your own commit messages and truly understand the final architecture.