Enterprise Autonomy
Menu
THE DEFINITION

What Is Enterprise Autonomy?

Enterprise Autonomy is the discipline of designing, governing, and operating enterprise systems in which software reads context, makes decisions, takes action across multiple systems of record, and is accountable for a business outcome end to end — under human supervision, measurable control, and auditable constraint.

The four properties

A system is autonomous in this sense only if all four hold.

  1. Context. It reads the state of the business from the systems where that state actually lives, rather than from a prompt.
  2. Decision. It selects among courses of action under policy, rather than executing a fixed script.
  3. Action. It writes back to systems of record and effects real change, rather than returning a recommendation to a person.
  4. Accountability. It owns a named business outcome with a measurable result, and its work is attributable, reversible, and auditable.

How it differs from RPA and copilots

Reads contextDecidesActsOwns outcome
Scripted automation (RPA)NoNoYesNo
Assistive AI (copilots)PartlyNoNoNo
Enterprise AutonomyYesYesYesYes

RPA executes but does not understand: it repeats a fixed sequence and breaks when the process changes. Copilots understand but do not execute: they answer and draft, then hand the work back to a person. Neither owns the outcome, so neither removes the work of seeing a process through. That gap is the subject of Why RPA and Copilots Cannot Own the Outcome.

Is this the same as agentic AI?

Agentic AI describes a technique: models that plan and call tools. Enterprise Autonomy describes an operating discipline. A system can use agents and still fail the test above — if it acts without a named outcome owner, without policy, or without a record that can be audited. The term covers the governance and accountability around the agents as much as the agents themselves.

Autonomy is not the absence of people

People set the policy, own the outcome, approve the decisions that need approval, and handle what the system escalates. The difference is where they stand: on the loop, supervising a process that runs, rather than in it, carrying each step by hand. Whether a system honours that boundary can be tested; the Governed AI Workflows benchmark checks that a workflow waits for approval, stops on denial, escalates a failure, and leaves a trace that reconciles with the systems it touched.

Where the idea comes from

In 1991 Mark Weiser wrote that “the most profound technologies are those that disappear.” He described a third wave of computing that recedes into the environment; with John Seely Brown he later called the aesthetic calm technology. Enterprise Autonomy is that idea applied to enterprise operations: work that gets done without demanding attention, while remaining accountable. The Autonomous Enterprise Is Weiser’s Third Wave develops it.

Glossary

Agent
A software component that uses a model to interpret a situation and choose an action. An agent is a building block; on its own it does not own an outcome.
Agent2Agent (A2A)
An open protocol for one agent to hand work to another. Any documented interface can serve the same purpose; the benchmark records which one a platform used.
Autonomous workflow
A business process carried from trigger to result by software that reads context, decides, and acts, with people supervising rather than doing each step.
Copilot
Assistive AI that answers questions or drafts work inside a tool and hands the result back to a person to act on.
Escalation
Handing a case to a person when the software cannot proceed within policy: missing data, a failed system, a decision above its authority. A good escalation says what is known and what is not.
Human in the loop
A person approves each step before it takes effect.
Human on the loop
Software acts within set limits while a person watches, samples, and can intervene or reverse. Approval is reserved for the decisions that need it.
Model Context Protocol (MCP)
An open specification for connecting AI systems to tools and data, so a system can reach enterprise software without custom integration for each connection.
Outcome ownership
Accountability for a named business result with a measurable target, such as invoices settled or work orders completed, rather than for tasks performed.
RPA
Robotic process automation: scripts that repeat a fixed sequence of steps in software. Reliable while the process is unchanged; it does not interpret or decide.
System of record
The authoritative source for a kind of business data: the ERP for orders, the maintenance system for assets, the HR system for employees.
Trace
The record of what an autonomous system read, decided, and did, detailed enough to reconstruct and audit a case after the fact.