Write the Damn Code (antonz.org)

The piece argues against the current trend of treating prompt engineering as a substitute for actual programming. Instead of endlessly refining prompts to coax perfect code from an LLM, the author recommends actively writing code and using AI as an assistant: ask a model for an initial draft and refactor it, write the first pass yourself then have AI review and improve it, implement the critical pieces and let AI fill in the rest, or provide an outline for the model to complete. If the model produces acceptable output in one or two iterations, great—but if not, stop the back-and-forth and get your hands dirty. For the AI/ML community this is a practical workflow shift with important implications: it improves reliability, debuggability, testability, and maintainability versus “programming in English.” Human-authored scaffolding reduces ambiguity and lowers the risk of subtle LLM errors, while guiding models toward desired abstractions and interfaces. The advice preserves AI’s productivity benefits without ceding ownership of design or critical logic, encouraging developers to balance model use with code literacy to achieve faster, more robust outcomes.