Anthropic Case Study · Context Engineers

The problem

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

Marketing wanted AI for ad workflows. Legal wanted contract redlining. Finance wanted month-end help. Every request needed engineering time. Engineering was the bottleneck.

What they built

A format. A shared library. Everyone authors.

Instead of building tools for every team, Anthropic gave every team the format to build their own — and a shared place to publish them.

Anthropic's skills stack

Layer 01 · Format

SKILL.md

A single markdown file with YAML frontmatter (name, description, when_to_use) and the rest is plain English instructions. No code. No training. Anyone on any team can author one.

↓

Layer 02 · Distribution

Skills repo

An open library of skills shared across the company (github.com/anthropics/skills). New ones get added like blog posts. The marketing team's "ad variations" skill, the legal team's "contract redline" skill, the docs team's "PDF report" skill — all in one place.

↓

Layer 03 · Use

Claude → Skill → Output

When a team member needs to do that task, Claude picks the right skill and runs it. The author is whoever wrote the markdown — not the AI team.

The result

Each team authors for themselves.

Austin Lau on the growth marketing team wrote a skill called "ad variations." It turns 30-minute manual creative iterations into 30-second AI batches, generating 100 image variations per round. The legal team uses Claude for contract redlining and email triage. The pattern repeats across teams: the team that needs the workflow authors it — and the skill lands in the shared repo for the rest of the company.

What used to take 30 minutes per ad now takes about 30 seconds. — Austin Lau, Growth Marketing, Anthropic

What you can steal

You don't need an AI team. You need a format that lets every team teach the AI.

If you're hiring AI engineers to build internal tools, you're bottlenecking yourself. Anthropic gave every team a markdown file and a shared library. That's the unlock. Your lawyers know more about legal review than your AI engineers ever will. Your finance lead knows more about your books. Give them a format and a place to publish — and the whole org becomes the AI team.

Sources

"How Anthropic's growth marketing team uses Claude" (claude.com)
"How Anthropic's legal team uses Claude" (claude.com)
Skills repo (github.com/anthropics/skills)

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

A format. A shared library. Everyone authors.

Each team authors for themselves.

You don't need an AI team. You need a format that lets every team teach the AI.

More case studies