The 5 Levels of an Autonomous Company

"Autonomous company" can sound like a switch: either a company runs itself or it does not. In practice it is a gradient. Some parts of a company can run from start to proof today with only an approval, while others will need people for years. Leaders need a way to say where they are, what the next step is and what must be true before taking it.

Self-driving cars faced the same problem and solved it with levels. So did AI labs describing their own progress. This article sets out five levels of company autonomy, explains what changes at each one, gives a short test for finding your own level, and shows what it takes to move up safely. It builds on the idea of the autonomous company: a company that notices, decides, delegates and checks its own work, while people keep the final say.

What are the levels of company autonomy?

The levels of company autonomy describe how much of a company's operating loop runs without a person doing each step. The loop has four parts: notice what needs doing, decide what to do and who owns it, carry it out, and check that it happened. At low levels, people do all four and AI helps on request. At high levels, the system does most of the loop for routine work, and people set goals, approve what matters and handle exceptions.

The idea of levels comes from other fields. SAE International's J3016 standard for vehicles defines six levels of driving automation, and its summary chart draws one line that matters more than any other. For Levels 0 to 2: "You are driving whenever these driver support features are engaged." For Levels 3 to 5: "You are not driving when these automated driving features are engaged." The question is always who is responsible for the task at that moment.

AI labs use levels too. In July 2024, The Decoder reported, citing Bloomberg, that OpenAI had set out five levels for its own progress, from chatbots at Level 1 to a Level 5 described as: "Can do the work of an entire organization." Microsoft's 2025 Work Trend Index uses three phases for companies, ending with "humans set direction for agents that run entire business processes and workflows, checking in as needed."

The five levels below apply that thinking to an operating company. They describe the company, not the model. A company can use the most capable model available and still sit at Level 1 if nothing reaches a person unless they ask.

The five levels of an autonomous company at a glance

Before Level 1 sits the manual baseline: no AI in daily work at all. Few companies are there now, so the ladder starts at Assisted.

Each level of company autonomy explained

Level 1: Assisted. People use AI tools to draft, summarize, research and code. Work still starts when a person notices something and asks. This is where most AI-first companies sit. It saves time per task and changes little about how the company coordinates. Promises still live in individual inboxes, and status still comes from meetings. The difference between this level and the next is covered in autonomous company vs AI-first company.

Level 2: Aware. The company starts to notice on its own. A system reads email, calendar, chat and business tools, forms the company's goals from its own documents and meetings, and compares pending work against them. When something drifts, such as a promised quote not sent or an audit with missing documents, it raises one item to one owner with a reason and a date. People still decide and act. This is the largest single jump, because it moves the company from reactive to proactive.

Level 3: Delegated. The system goes from raising situations to proposing actions. The owner can hand the work to a colleague or an AI agent. Agents prepare drafts, follow-ups, reports and bookings, and nothing is sent until a person approves. Items close only on proof, such as the email that went out. This is where the record of outcomes begins to build.

Level 4: Supervised. For specific routines that have a long record of approved, correct results, the company lets the system run from start to proof within written rules. A person moves from approving every action to reviewing samples and handling exceptions, sometimes called being on the loop instead of in it. The human in the loop vs human on the loop article covers how to choose.

Level 5: Autonomous. Across most routine operations, the company notices, decides, delegates and checks its own work. People set the goals, approve anything that commits money, customers or reputation, and keep the final say. The system learns from every outcome, so its picture of what matters keeps improving. Even here, many decisions stay human by design: negotiations, hiring, pricing strategy and anything new.

  L5 Autonomous   company runs routine work, people set goals
  L4 Supervised   proven routines run to proof, people spot-check
  ─────────────── the system runs parts of the loop ────────
  L3 Delegated    agents act after a person approves
  L2 Aware        the company notices on its own
  L1 Assisted     AI helps when someone asks
  ─────────────── people run the loop ───────────────────

The line in the figure sits between Level 2 and Level 3 for the same reason SAE draws its line between Level 2 and Level 3: below it, people carry out the work and are responsible for each step. Above it, the system carries out parts of the work and people answer for the rules and the approvals.

Which level is your company at?

Answer five questions about the last month. The highest level where you can answer yes for one team's main work is that team's level. Different teams usually sit at different levels.

1. Do people use AI tools in their daily work? If yes, at least Level 1.
2. Did the last dropped promise or slipping deadline reach one named owner before anyone asked? If yes, Level 2.
3. Are routine actions prepared by AI and sent only after a person approves, with each item closed on proof? If yes, Level 3.
4. Is there at least one routine that runs to proof without per-item approval, with written rules and regular spot checks? If yes, Level 4.
5. Does most routine work across teams get noticed, routed, done and checked this way, with people mainly setting goals and approving commitments? If yes, Level 5.

Be strict on question 2. If the honest answer is "someone noticed at the weekly meeting," the company is at Level 1, however much AI it uses.

How to move up a level

Each step has a prerequisite and a typical risk. Skipping the prerequisite is how autonomy projects fail.

Anthropic's experiment with an AI agent running small shops shows why the later steps need care. In Project Vend phase two, Anthropic concluded that "the gap between 'capable' and 'completely robust' remains wide." Level 4 is where that gap is closed for one routine at a time, with evidence. The autonomous company roadmap lays out a longer sequence for the whole company.

Research on agent capability also suggests which work moves up first. In March 2025, the research group METR reported that the length of software tasks AI agents could complete had grown with "a doubling time of around 7 months" over six years. The models it tested had "almost 100% success rate on tasks taking humans less than 4 minutes," but succeeded less than 10% of the time on tasks taking more than about 4 hours. METR notes that some of those figures are now out of date, and its tasks were software tasks. The pattern still gives a practical rule for operators: short, well-defined, frequent routines, such as a follow-up email, a report pulled together from known files or a meeting booked against a calendar, are the first candidates for Level 4. Long, open-ended work such as a negotiation or a new supplier search stays at Level 2 or 3 for much longer, with a person deciding and an agent preparing.

How to measure progress up the levels

Each level has a measure that shows whether it is real. Track these per team.

Watch items marked "not relevant" as well. A falling share means the system understands what the company cares about. The full set of measures is in how to measure how autonomous your company is.

Common mistakes with autonomy levels

• Grading the whole company at once. Sales may be at Level 3 while finance is at Level 1. Grade per team and per type of work.
• Skipping Level 2. Jumping from chat assistants to agents that act means agents act without a shared picture of goals and owners. Noticing comes first.
• Treating Level 5 as unattended. At every level, people approve what commits money or customers. Autonomy changes the routine work, never the authority.
• Moving up on a deadline. Lighten approvals when the record shows a routine holds up, not when a quarter ends.
• Never moving down. If spot checks find errors, return that routine to Level 3 until the cause is fixed. Levels are reversible by design.

How GeniOS approaches the levels of company autonomy

GeniOS is built to take a company from Level 1 to Level 3 first, and then further as its record grows. It reads where the company's work happens: email, calendar, chat and the business tools it already uses, read-only to start. It forms the company's goals from its own emails, decks and meetings, and checks every team's pending work against them. That is Level 2.

What reaches a person is intelligence: one situation, what to do, why, one owner and a date. The owner answers Delegate, Done, Later or Not relevant. Delegate hands work to a person or an AI agent, and nothing is sent until a human approves. That is Level 3.

An item closes only on proof, such as the email that went out, and GeniOS learns from each outcome. That record is what lets a company decide, routine by routine, when to move further up, while people keep the final say.

Worked example: one company, three levels at once

The names, dates and numbers in this example are made up to explain the idea. Vaultex makes safety footwear in Greater Noida and is GeniOS's first design partner. Like most companies, different parts of its work sit at different levels.

Nothing in this picture is unattended. Harsh still approves the quote, Anushka still owns the production plan, and the finance owner still decides on terms. What has changed is that each situation reached the right person before Friday, each action was prepared for them, and each closes on evidence. Over months, routines like assembling the audit pack move up a level once the record shows they hold. See the Vaultex case study for the real setup.

Conclusion

The levels of company autonomy give leaders a shared language: Assisted, Aware, Delegated, Supervised and Autonomous. They change who notices, who acts and who checks, while people keep final authority at every step. Most companies are at Level 1 today, and the jump to Level 2, a company that notices on its own, does the most good. Move one team and one routine at a time, and let the record decide when to go further. To see where your company would start, see pricing or book a call.

Sources

• SAE International, SAE J3016 Levels of Driving Automation summary chart (copy hosted by RoboticLab), 2019
• The Decoder, OpenAI unveils five-level AI scale, aims to reach level 2 soon, 12 July 2024
• Microsoft WorkLab, 2025 Work Trend Index: The Year the Frontier Firm Is Born, 23 April 2025
• Anthropic, Project Vend: Phase two, 18 December 2025
• METR, Measuring AI Ability to Complete Long Software Tasks, 19 March 2025