A lawyer wrote a markdown file. It became an AI skill.

Every team needed AI tools. Only one team could build them.

Anthropic's marketing team wanted an AI that drafted ad variations in their voice. Legal wanted an AI that flagged risky clauses in vendor contracts. Finance wanted automated month-end reporting. Customer support wanted instant ticket triage.

All of these were possible. None of them required real engineering work. But all of them sat in a queue waiting for the small AI tools team to build them — because that's how internal tooling worked everywhere else. The bottleneck wasn't capability. It was authorship.

A skill is just a markdown file. Anyone can write one.

Anthropic created a format called SKILL.md. A skill is one file — frontmatter at the top describing when to use it, then plain English instructions for how the AI should approach the task. No code. No engineering review. Just write what you'd tell a junior teammate, and the AI follows it.

The skill is read by every Anthropic employee's AI. The marketing team didn't need to deploy anything, ship code, or coordinate with engineering. They wrote what they knew, and the AI started following it. The AI got a new capability the moment the file was saved.

The format scaled. Within months, every team was authoring:

Marketing went from weeks of variations to half a second.

The ad-variations skill alone collapsed something that took the marketing team a couple of weeks of iteration into a 0.5-second AI call. The legal skill flagged 4x more risky clauses than the human-only review process. Finance month-end reporting dropped to minutes.

But the bigger shift was who was shipping AI. Anthropic didn't grow its AI team. It grew the number of AI authors — by a factor of ~20x within a year. Lawyers, marketers, and ops people became "the AI team" for their domain.