Anthropic just published a 23-page document on how their own teams use Claude Code internally.

Not their marketing team. Not their sales pitch. Their actual employees, doing actual work.

The findings are worth your time.

What they found

A legal team member built an accessibility app for a family member. In one hour. No developer involved.

A finance team learned to describe their data workflows in plain text -- "query this dashboard, get these numbers, produce an Excel report" -- and Claude executed the entire thing. No Python. No SQL. No ticket to IT.

A data scientist with "very little JavaScript experience" shipped a 5,000-line TypeScript dashboard for model performance analysis. Not a prototype. A production tool the team uses daily.

A growth team of one person cut ad copy production from two hours to fifteen minutes, then built a Figma plugin that generates 100 ad variants on the spot.

Security engineering reduced Kubernetes incident debugging from 15 minutes to 5. Not by replacing the engineer -- by giving them a partner who could trace stack traces and suggest exact commands.

The pattern

Every single non-engineering team became more productive by building, not just prompting.

That's the part most people miss.

The conversation around AI is still stuck on "will it replace developers?" Wrong question. The real shift is that everyone else -- legal, finance, marketing, operations -- stopped waiting for developers and started building what they needed.

The legal person didn't become a developer. The finance team didn't learn to code. They just stopped being blocked.

What this means for your company

If you run a 20-person company with no dedicated IT team, this is your moment.

You don't need to hire three developers. You need to give your existing people the tools to build. A sales manager who can automate their own reporting. An operations lead who can build their own dashboard. A customer service agent who can create their own workflow tools.

The technology exists. It's not experimental. Anthropic's own teams are using it daily, and they're documenting exactly how.

The real question

This isn't about choosing between AI and people. It never was.

It's about whether you give your people the ability to solve their own problems, or keep them waiting in a ticket queue for someone else to do it.

Less about replacing humans. More about unleashing them.