Tan walks through a real skill file to show how unmagical it is. When a meeting recording lands, transcribe it with speaker labels, pull out the commitments made and who made them and the deadline, cross-check every person named against the library and link their pages, file the summary here and the full transcript there — and if anything contradicts something already believed, flag it rather than overwrite it. That is the whole artifact: a page of English. The test he draws from it is the portable part: “If a smart intern could follow it, an agent can run it.”
The claim underneath is one he notes he has taken flak for and believes more than ever — markdown is actually code, and if you can write clear instructions in English you are a programmer, because the compiler is a language model. The evidence he offers is who is doing it: media people, event staff and finance team members at YC who never open a terminal are writing skill files and scheduling jobs, including a finance colleague who compiled roughly a hundred Excel workbooks into a single app built with an internal agent. Tan’s description of her is the point — “She is not a programmer. She is a manager of agents now.” This extends Markdown skill files may replace expensive fine-tuning from a cost argument to an access argument, and it is bounded by Intelligence location — code vs prompts — determines system fragility and flexibility: the intern test tells you the instructions are clear, not that the work belongs in prose rather than in code. The quality bar it implies is the same one behind If an action left no artifact, it didn't happen — legibility is the precondition, not a nice-to-have — writing the procedure down is what makes it executable at all.