Markdown skill files may replace expensive fine-tuning

A skill is just a markdown file that tells the agent how to do something specific. Non-engineers can create skills (analysts, customers), no deployment is needed — change the file, it takes effect immediately, and everything is readable and auditable.

The key question: can skill files match hand-crafted fine-tuned models? If yes, the moat of “we trained on proprietary data” weakens dramatically. Anyone could write a SKILL.md file. Vertical AI startups whose differentiation was domain-specific fine-tuning face disruption from In agent-native architecture, features are prompts — not code — self-learning agents with skill files, not expensive training runs.

Bustamante’s “shadowing system” shows this in practice: if you don’t like how the agent does DCF valuations, write your own skill file and your version wins. This is why Context is the product, not the model — the skill layer IS the product, and it’s built on Files are the universal interface between humans and agents.

Tan extends the argument from cost to access with a portable readiness test — If a smart intern could follow it, an agent can run it — which makes clear English a programming language — and the claim behind it, that markdown is code and the compiler is a language model. His evidence is who is writing the files: media, events and finance staff at YC who never open a terminal, including a finance colleague who compiled roughly a hundred Excel workbooks into one app built with an internal agent.