Measuring Autonomy: A Maturity Model

Everyone building in this space claims some version of autonomy. But the word covers an enormous range — from a company that uses ChatGPT for drafting emails to a fully self-governing entity with no human in the loop. Without a shared framework for measuring autonomy, the term loses meaning.

This is a five-level maturity model for company autonomy. It is designed to be practical: each level has observable criteria, and the boundaries between levels are defined by what the system can do without human involvement.

Level 1: Assisted

At Level 1, humans do the work. Agents assist. The company uses AI tools to augment human productivity — writing assistance, code generation, data analysis, scheduling — but every consequential decision is made by a person. Every workflow has a human in the critical path.

Criteria:

• All business decisions require human approval.
• Agents operate only within single tasks, not across workflows.
• No autonomous spending or resource allocation.
• Human management structure is fully intact.

This is where the vast majority of companies are today. The tools are better, but the organizational structure is unchanged.

Level 2: Partial Automation

At Level 2, agents handle defined workflows end-to-end, but only within tightly scoped domains. A company might automate its entire customer support pipeline, or its content production process, or its financial reporting. But these automated domains are islands — they do not coordinate with each other autonomously, and a human decides which domains to automate and how.

Criteria:

• One or more complete workflows operate without human intervention.
• Automated workflows have defined boundaries and escalation paths to humans.
• Humans still set strategy, allocate resources, and coordinate across domains.
• The company could not operate at all if the human leadership disappeared.

Most companies that describe themselves as "AI-native" are at Level 2. They have automated pieces of the operation but not the connective tissue between them.

Level 3: Conditional Autonomy

Level 3 is where the shift becomes structural. The company operates independently across multiple domains, with agents coordinating across workflows. Humans are still present, but their role has changed from operators to overseers. They set policy, monitor performance, and intervene when the system exceeds its confidence boundaries.

Criteria:

• Agents coordinate across multiple business functions without human mediation.
• The system can allocate resources within policy-defined constraints.
• Humans intervene on exceptions, not on routine operations.
• The company could sustain basic operations for days or weeks without human input.
• Governance is partially encoded in system policy rather than human management.

This is the level where the firm starts to feel genuinely different from a traditional company. The humans are still essential, but they are no longer in the execution path for most operations.

Level 4: High Autonomy

At Level 4, the company is self-directing. Agents handle not just execution but planning, prioritization, and adaptation. Human involvement is limited to high-level governance: setting the mission, defining ethical boundaries, and making structural decisions about the system itself. Day-to-day and week-to-week operations proceed without human input.

Criteria:

• The system sets its own operational priorities based on objectives and environmental conditions.
• Treasury management and capital allocation are autonomous within governance constraints.
• The company adapts to market changes, competitive pressure, and operational failures without human direction.
• Humans retain veto power and can modify governance, but do not participate in regular operations.
• The company could operate indefinitely without human input, assuming no governance changes are needed.

Very few organizations have reached Level 4. Those that approach it are typically narrow in scope — single-product, single-market — because the coordination complexity of broad autonomy remains unsolved.

Level 5: Full Self-Governance

Level 5 is the theoretical endpoint: a company that governs itself entirely, including setting its own strategy, modifying its own governance structures, and operating without any human in the loop. The entity is fully autonomous in the same sense that a biological organism is autonomous — it perceives its environment, makes decisions, acts, and adapts, all on its own.

Criteria:

• No human involvement in any operational or strategic decision.
• The system can modify its own objectives and governance within foundational constraints.
• Self-sustaining economically — generates and manages its own revenue and resources.
• Can engage with legal, regulatory, and commercial systems independently.
• Reproduces or scales itself without human initiation.

No organization operates at Level 5 today. Whether any should is an open question with significant ethical and safety implications. The framework includes it because the trajectory points here, and understanding the destination helps navigate the path.

Why most organizations are at Level 1-2

The gap between Level 2 and Level 3 is the hardest to cross. It requires not just better tools but a fundamental redesign of how the organization works. Moving from automated workflows to autonomous cross-functional coordination demands:

• Robust inter-agent communication protocols.
• Policy systems that can handle ambiguity and edge cases.
• Treasury and resource management that agents can operate directly.
• Monitoring and governance infrastructure that replaces human management.

Most organizations stall at Level 2 because they try to automate the existing structure rather than redesigning for autonomy. The companies that reach Level 3 and beyond are the ones that start from the system design, not from the org chart.

This model is not prescriptive about where any company should be. Higher is not always better. The right level of autonomy depends on the domain, the risk tolerance, and the maturity of the available infrastructure. But knowing where you are — honestly — is the prerequisite for knowing where you are going.