Skip to content
Prepline
LibraryAI Agents & SDKs2 min readUpdated 2026-07-18
← All courses Framework — 2026

The 5 Steps of AI Adoption

Boris Cherny's ladder for how a team grows into working with AI — from every action gated behind a human, to a thousand agents running in loops overnight. Where are you on it?

🧠 5-step framework 42-second explainer 👤 After Boris Cherny, Head of Claude Code 🔌 Fully offline
▶ Watch first

Watch the animated explainer

The whole ladder in forty seconds — the five rungs, the gate where teams first stall on autonomy, and how leverage climbs from zero agents to a thousand.

Key takeaway: the gate is Supervised Autonomy — you only pass it once your verification is strong enough to trust the loop.

The five rungs

Each step hands more of the work to the agents — and multiplies what one engineer can ship.

1

Gated

AI is boxed in. Every action waits for explicit human approval before it runs. You review each suggestion by hand; nothing happens autonomously.

0agents
2

Assisted

AI works alongside you — autocomplete, chat, single-shot edits. One agent, but the human stays firmly in the driver's seat and makes every call.

1agent
3

Parallel

You run several agents at once on independent tasks. You stop typing the work and start orchestrating it — dispatching, merging, and steering.

10agents
4

Supervised Autonomy • THE GATE

Agents run end-to-end; you review outcomes, not steps. Most teams stall here on the first try — you only pass once your verification is strong enough to trust the loop. Fail it, and you drop back to build that trust. Pass it, and autonomy compounds.

100agents
5

AI-Native

AI-first by default. Humans set direction, write the loops, and handle escalations while agents do the work — hundreds to thousands of them, running in loops around the clock.

1000+agents

Agents per engineer

The same climb, as a single number. Each rung is roughly an order of magnitude.

0Gated 1Assisted 10Parallel 100Supervised 1000+AI-Native
“I don't prompt Claude anymore. I have loops running that prompt Claude and figuring out what to do. My job is to write loops.” — Boris Cherny, Head of Claude Code, Anthropic

Ready for the rung above? The companion course goes deep on the practice that makes autonomy safe: Loop Engineering →

Need this for a date?

Turn this course into a ramp-up pack sized to your minutes per day, or build an interview or certification pack for the day you need it.