Room to dream · Seven levels of autonomy

How far can you scale with AI? The ceiling.

From the first question to the company that runs itself. Dreaming is free — climbing happens one level at a time, and every level is earned with evidence from the one below.

▼ SCROLL DOWN TO CLIMB
L0
The on-demand assistant
You write to it when you remember to.
0
hrs/wk returned

What the AI does

Drafts, summarizes, analyzes, and explains whatever you paste in. Every conversation starts from zero: you are its only memory.

What you do

You ask the questions, copy the answers, sign everything. 100% of the judgment and 100% of the legwork are still yours.

The lock

The AI drafts, the human signs. Nothing with names, credentials, or personal data goes in without anonymizing it first.

L1
The connected colleague
You no longer explain who you are.
0
hrs/wk returned

What the AI does

Connected to your email, your documents, and your system (read-only): it answers with YOUR data. Every morning the inbox digest and draft replies are waiting for you.

What you do

You approve and send. You decide what it can read and what it can't — folder by folder, system by system.

The lock

Read-only connectors, company accounts, explicit permissions. Reading is not touching: at this level the AI still writes to nothing.

L2
Supervised automation
It runs on its own — but in a sandbox.
MOST CLIENTS START HERE
0
hrs/wk returned

What the AI does

Complete flows run on their own: register, classify, calculate, report. But they write to a TEST ledger mirroring the real one, and every run leaves a log.

What you do

You compare the test ledger against the real records and promote by hand. You accumulate weeks — months — of evidence before loosening the leash.

The lock

Mirrored test environment + manual promotion + a log for every run. Trust isn't declared: it's documented.

L3
Operating by exception
You only see what doesn't add up.
0
hrs/wk returned

What the AI does

Runs the process end to end in the real system: reconciles, cross-checks spreadsheets, posts, archives. And it surfaces ONLY the exceptions — whatever didn't add up.

What you do

You manage a queue of 10 odd cases, not 400 rows. Every exception you resolve refines the rules: the system learns your judgment, in writing.

The lock

Explicit green-light rules: what it may touch, up to what amount, within what scope. Everything reversible, everything audited, nothing silent.

L4
Software that writes itself
You describe the system; the AI builds it.
0
hrs/wk returned

What the AI does

Builds and maintains your tools: internal apps, integrations, reports, dashboards. It proposes experiments, runs them, documents them, and shows you the results.

What you do

You are the architect. You define what exists and why. The question is no longer "can we build it?" — you always can. It's "does it deserve to exist?"

The lock

Separate environments (dev / test / production), review before every deploy, and nothing reaches production without your explicit green light.

L5
The digital department
Agents with a role, a schedule, and a KPI.
0
hrs/wk returned

What the AI does

Teams of agents with fixed responsibilities, 24/7: one reconciles, one watches inventory, one preps purchasing, and one audits the others.

What you do

You manage a digital workforce: you hire (create an agent), evaluate (measure its KPI), and fire (switch it off). Your org chart has humans and agents.

The lock

A budget per agent, permissions per role, agents auditing each other — and a red button that shuts it all down, tested every month.

L6
The AI-native company — the ceiling
One architect, a hundred hands.
the time is yours

What the AI does

The entire operation runs itself: purchasing, logistics, finance, reporting, even its own tools. The operating org chart is made of agents, not job titles.

What you do

What no AI can be: the judgment, the relationships, the strategy, the accountability. Everything that goes out into the world carries your signature.

The lock

You. At this level human judgment isn't the system's brake — it's the rudder. Without it, everything else is just speed.

The ceiling isn't the AI doing your work.
It's you doing the work of a hundred — and still signing everything.
LADDER RULE: NEVER CLIMB TWO LEVELS AT ONCE. EVERY LEVEL IS EARNED WITH EVIDENCE FROM THE ONE BELOW.
Find out what level you're on →