“We use AI” can mean one person asking ChatGPT to draft an email. It can also mean a business where AI does real work every day and someone measures it. Those are very different places to be.
This ladder gives each step a name, so you can see where you are today and what it takes to reach the next rung.
- LEVEL 0Ask
Chat for answers and drafts
- LEVEL 1Assist
AI works from your files and tools
- LEVEL 2Repeat
Best workflows become shared skills
- LEVEL 3Run
Skills run on schedules and triggers
- LEVEL 4Operate
Governed agents, measured monthly
Level 0Ask. Chat for answers and drafts.
People open a chat window, ask a question, and copy the answer somewhere else. It helps, but nothing is connected to your business and nothing is shared.
Level 1Assist. AI works from your files and tools.
The AI can see the documents and systems your team already uses, so its answers are about your business, not the internet in general.
Level 2Repeat. Best workflows become shared skills.
When someone finds a good way to do a task with AI, it gets saved as a skill: written instructions anyone on the team can reuse. One person's good habit becomes the team's habit.
Level 3Run. Skills run on schedules and triggers.
The work no longer waits for someone to ask. It starts on its own, every morning or whenever a new email, form, or order comes in.
Level 4Operate. Governed agents, measured monthly.
AI agents carry out ongoing work under clear rules, and the business checks every month whether that work is paying off.
Where the payoff usually is.
You don't have to reach Level 4 to see a return. The jump that pays is usually from Level 1 to Level 3: two or three workflows redesigned, automated, and measured.
Adapted from Anthropic's five-level model. Source: Anthropic, Deploying Claude across your organization: How Anthropic uses Claude Cowork (Apr 2026).