Issue 7, Sep 26, 2026
The A Priori Skill Trap
Senior AI Engineer interview at OpenAI, and the interviewer asks:
“Your PM wants to save engineering time: ‘Just have the agent write its own skills from the task description.’ Benchmarks show this can make the agent worse. Why? And what should a skill actually be built from?”
Don’t say: “The model isn’t smart enough yet. Use a stronger model and a better prompt.”
Why prompting an LLM to write skills from task descriptions quietly degrades performance, and how compressing verified execution traces builds true agent capability.
Share this trapShare on LinkedIn
The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.