It’s extremely tempting to prompt an LLM to write an RFC, an ADR, or similar technical docs. In seconds, you get a clean, polished, multi-page document that seems to overflow knowledge and professionalism. While AI is brilliant at identifying edge cases, reviewing existing drafts, or suggesting missing considerations, it should never take the driver’s seat when writing core technical docs.

Drafting an RFC or ADR isn’t just transcription; it forces you to confront trade-offs, weigh organizational constraints, and internalize the consequences. Offloading the writing means offloading the actual reasoning. An AI doesn’t know your team’s historical tech debt, your org’s risk tolerance, or the subtle political trade-offs between two conflicting engineering teams.

AI-generated docs sound authoritative even when the underlying architecture is flawed. A poorly formatted human doc with deep insight beats a beautifully structured AI doc with shallow assumptions every single time. Use AI as an interactive reviewer, a devil’s advocate, or a proofreader. But never as the author.

Let AI find the holes in your logic, but keep human engineers firmly behind the wheel.