Blog post argues your AI prompt style reveals your psychology, not optimal technique; LLMs are mirrors, not oracles.
Key Takeaways
System prompts and interaction styles (iterative vs. full-spec, partner mode vs. command-line discipline) reflect user identity, not transferable best practices.
Author, a Google/Facebook release engineer, uses an iterative, conversational prompt she calls “partner mode” and acknowledges it proves nothing generalizable.
LLM productivity gains are real only when output validates the method; if code ships faster, you have a toehold in reality, otherwise it is solo roleplaying.
AI amplifies existing technical skill and will likely increase income inequality, not reduce it, regardless of equality pledges.
Skeptics and boosters alike use the shadow glass: both confirm pre-existing beliefs rather than testing the tool empirically.
