A Framework for Understanding AI at Work

PART 1 - The Anatomy of AI

AI isn't one thing. It's a stack of three layers, and what's possible depends on all three working together. Click any block to explore.

Most people think AI = the model. In practice, the app and harness determine what AI can actually do in your workflow.

Pause & reflect

Think about the AI tool you use most at work. What's the model, the app, and the harness? Which layer is doing the most work for you - and which might you be underusing?

PART 2 - The Practice of AI

AI use can look very different depending on three factors: how much context you give it, how repeatable your use is, and how much independent action you allow it. Explore the four modes below.

- not a hierarchy, but a map

How each mode scales

	Single use (one person, one task)	Persistent (one person, ongoing)	Shared (the whole team)
Conversation	Fresh each time	-	-
Context	Upload a file	Claude Project	Shared Projects
Packaged Expertise	Paste a prompt	Saved Skill	Team-wide Skill
Supervised Agency	One-off task	Refined workflow	Standard agentic workflow

Pause & reflect

Think of one work task you regularly use AI for. How much context do you give it, how repeatable is your approach, and how independently does the AI act? Could someone else on the team pick up your approach and start using it right away?

PART 3 - From Individual to Collective

Strong AI practices exist on this team, but they're siloed inside individuals and sub-teams. Here is one path to turn individual fluency into institutional capability - six steps that form a continuous improvement loop.

How do we turn individual fluency into institutional capability?

The human-oversight principle from the opening applies here with the most weight. When a practice is used by one person, the stakes of a quality failure are contained. When it's used by the whole team, the stakes multiply.

Pause & reflect

Above is one proposal to build shared AI practices across the team. What do you agree with, and would you do anything differently from what's proposed? Try to be as concrete as possible.