"We use AI" can mean a chat window open in a browser tab or a billing process that nobody has touched in three months. Those are very different companies, and they need different next steps.
We use a five-level scale to tell them apart. It is borrowed from self-driving cars, where the same question had to be answered: who is doing the driving, and who is responsible when something goes wrong?
The scale
L0, Manual
People do everything. Software stores the data, but every step is started and finished by a person.
L1, Assisted
Assistants help people do their work faster. Someone asks for a draft, a summary, or an analysis, and then decides what to do with it. The person is still the one who acts.
L2, Automated
One process runs on its own, from start to finish. An agent reads the incoming invoice, matches it, books it, and follows up on what is missing. People see the exceptions and the results.
L3, Connected
Agents work across the company and hand work to each other. The sales agent's signed order becomes the operations agent's task and the finance agent's invoice, without a person carrying it from one system to the next.
L4, Autonomous
The company runs itself. People set the destination: strategy, limits, and the decisions that are theirs.
Where companies are
No company has reached L4. That is the reason for the scale: it is a map of work still to be done, not a set of labels to award.
In our experience, most companies that use AI are at L1, and some have one or two processes at L2. L3 is where the hard engineering is today, because connecting agents means shared context, clear hand-offs, and limits that hold across systems.
How to tell where you are
A company does not have one level. Each process has its own. For any process, three questions settle it:
- Who starts the work? If a person has to ask, it is L1 at most.
- Who finishes it? If a person copies the result into another system, it is not yet L2.
- What happens to the output? If another agent picks it up without a person in between, that link is at L3.
Asking these process by process gives you a map. It usually shows a few L1 areas, many at L0, and one or two obvious candidates for L2.
Moving up is earned
A level is not declared. It is earned on evidence.
A process moves to L2 when its logs show that the agent resolves the work inside agreed limits, at an agreed quality, for long enough to trust it. Until then, it runs with a person approving each action.
The same rule applies to every later step. Limits widen when the data justifies it, and they narrow again if quality drops. We describe how those limits are written in What an Agent Does Alone, What It Asks, and What It Never Does.
What does not change
At every level, people stay responsible. More autonomy means agents do more of the work. It does not mean nobody is in charge.
If you want to map your own processes on this scale, book a call. For the bigger picture, read What Is an Autonomous Company?.
What goes on autopilot first?
Tell us how you work today. We will tell you where to start.
Book a call