Autonomous business

An autonomous business can operate and improve its core activities within a defined mandate, without requiring people to perform or approve routine work.

It can attract and serve customers, deliver its product or service, manage operations and respond to what it learns. People retain ownership, governance and direction. They can contribute without being needed to keep every routine activity moving.

AI-native, AI-first and autonomous

Three different questions: how was the company designed, what does it prioritize, and how independently can it operate?

AI-native

Built from the beginning with AI central to how it creates value and operates. This describes the original design of the business, not its founding date or whether it sells an AI product.

AI-first

A strategy that asks what AI and automation can do before assuming a person must perform each step. An established company can adopt it, while choosing human work where it creates more value.

Autonomous

Describes how independently work can operate. Core activities run and improve within a defined mandate, without people having to perform or approve routine work. Ownership, governance and direction remain.

An incumbent can adopt an AI-first strategy and move toward autonomy. A founding date before or after 2022 does not establish whether a company is AI-native.

The Autonomous Business Scale

Five levels describe AI’s role in getting work done, from helping a person with a task to operating and improving the core business.

L1AI-assistedAI helps with tasks. People perform and coordinate the work.

A person uses AI to help with an individual task, such as investigating a customer question. The person decides what to do, takes action and replies. When that person stops, the wider process waits.

L2AI-draftedAI prepares work or proposed actions. People review and approve.

AI gathers information and prepares substantive work, such as a proposed account correction and customer reply. A person reviews, adjusts and approves before it takes effect. Throughput still depends on the attention available for review.

L3AI-operatedAI completes a bounded process. People set limits and handle exceptions.

The system completes a defined process within agreed limits, such as verifying a payment, restoring access and informing the customer. Routine cases do not need individual approval. Unclear or unusual cases go to a person.

L4AI-runAI connects processes across functions. People retain selected decisions and oversight.

Connected processes carry work across functions. A recurring support problem can trigger a product investigation, a fix and updated customer guidance. People retain selected decisions, relationships and oversight, without coordinating each routine handoff.

L5AutonomousThe core business operates and improves within its mandate. Routine human intervention is optional.

The business can detect changing needs, develop and test responses, deliver them and evaluate the results within its mandate. L5 applies across the core business. People retain ownership, governance and direction; routine work does not depend on their intervention.

How to use the scale

Start with a complete workflow: who initiates it, who does the work, who approves the result, and who improves it? Different parts of one business can sit at different levels. Support might be L3 while marketing is L2.

L3 applies to a bounded process. L4 connects work across functions. L5 is a claim about the core business as a whole, including its ability to adapt and improve within its mandate. Running existing routines for a period does not, on its own, demonstrate L5.

Improvement can mean testing a product change, updating instructions or learning from a recurring problem. It does not require training a new AI model. Human review can remain a deliberate choice for important decisions, and accountability remains.

Developed through the research and conversations behind Growth After Headcount, with contributions to and review of the chapter’s definitions by Henry Shi.

These conversations explore the work behind autonomy. They are not verified company-wide level assignments.

Revenue per employee

One lens on how much a business can accomplish with its people. It measures commercial scale relative to the team, not autonomy or profitability.

Annual revenue ÷ Average full-time-equivalent workforce

Use the same year for revenue and the average workforce. Include active founders, and make contractor contributions visible. If a source counts only payroll employees or uses year-end headcount, say so.

Six comparison bands

These are the book’s proposed comparison bands, not industry standards. A0 starts below $500K per person; A5 starts at $1B in annual revenue per person. A company’s valuation does not qualify it for a revenue band.

A0<$500K
A1$500K–<$1M
A2$1M–<$10M
A3$10M–<$100M
A4$100M–<$1B
A5$1B+

Each upper boundary is excluded: exactly $1M belongs to A2. A0 includes young companies and established labour-intensive businesses; it does not mean a company uses no AI or is poorly run.

L1–L5 describe how independently work operates. A0–A5 describe revenue per person. A company can become more autonomous before its revenue rises, or increase revenue without becoming more autonomous.

Read the figure alongside profitability, customer outcomes, contractor costs and AI spend. Revenue is earned over a period; annual recurring revenue (ARR) is an annualized recurring-revenue snapshot. Keep those measures separate.

Company snapshots

The book’s examples use different dates and measures. ARR figures and estimates illustrate the scale of a band; they do not establish completed-year revenue per person.

Selected snapshots, 2024–2026. Amounts in US dollars.
BandCompany / dateAmount / personBasis and sources
A0Target 2025~$252K$104.780B net sales ÷ 415,000 year-end employees, including part-time and seasonal staff.Financial statements · Headcount
A1AKOOL Nov 2024$800K ARR$40M invoiced ARR ÷ 50 employees in the leaderboard. An ARR snapshot, not completed-year revenue.Company release · Leaderboard
A2Super.com Jul 2025$1MLeaderboard-reported annual revenue of $220M ÷ 220 employees.Leaderboard
A2RB2B May 2026$3M ARR$9M ARR ÷ three people in the founder’s account and leaderboard.Founder account · Leaderboard
A3Midjourney Feb 2025$12.5M (estimate)Leaderboard snapshot: $500M annual figure ÷ 40 employees. Not confirmed annual accounts.Leaderboard
A4Medvi 2025~$200.5M$401M reported revenue ÷ founder and one employee. Outside contractors and clinical and pharmacy partners are excluded.Reporting
A5Open frontier$1B+No completed-year example established in these snapshots. A forecast or billion-dollar valuation does not qualify.

The private-company figures count the people included in each source and may exclude outside labour. Medvi’s small core team, for example, relies on external clinical and pharmacy operations. These figures do not measure the whole workforce supporting delivery.

The company comparisons draw on Henry Shi’s Lean AI Leaderboard , with the source basis shown for each row.

Where could your business go next?

Explore the conversations and decisions behind the framework.

Explore the podcast

Stay in the thinking.

Conversations with founders and business leaders. Notes from the book. What I’m learning as I build with AI.