AI instruction sets should mirror a company org chart, essay argues

The essay is framed explicitly as an idea, not a tested result, so the org-chart mapping is a way of thinking rather than a finding. The author traces the view back roughly a year and a half, describing a chain in which the smartest AI directs several lesser AIs, which in turn direct more below them, until the ones at the end use tools, or algorithms. The argument is that companies have been optimized over a long time, so AI at scale may settle into the same shape.

Mapping assets onto that shape, the author places copilot-instructions.md at the top as the president, instructions just below it as fellow top-down thinkers, and skills as rank-and-file staff who reason from the bottom up. Skills are said to hold the criteria for choosing among tools, execute using those criteria, and sit above the algorithms at the very end of the instruction chain.

The unresolved piece is what sits between the top-down files and the bottom-up skills. The author calls this a middle layer that users must define, with agents fanning out through it in pyramid form, and admits the naming for it is still unsettled. Prompts are the part the author cannot place at all, given that any method written into a prompt is effectively a skill.