# What is Autonomy Expansion?

Also called graduated autonomy, autonomy ladder.

Autonomy expansion is the practice of increasing an agent's independence in defined stages as evidence of reliability accumulates, rather than granting full independence at deployment. A typical sequence moves from observation only, to drafts requiring approval, to independent action within limits, to independent action with sampled review. Each stage has entry criteria and a route back.

Staging works because reliability is process specific and cannot be assumed from performance elsewhere. An agent that handles scheduling flawlessly tells you nothing about how it will handle refunds. Each new process starts at a low stage regardless of the agent's history, and the evidence to advance has to be gathered for that process specifically.

Entry criteria should be written before the stage begins, expressed as observed numbers over a stated period rather than as a feeling of confidence. A criterion such as two hundred consecutive items with a correction rate under five percent is checkable. Advancing because things seem to be going well reintroduces exactly the judgment the staging was designed to replace.

The route back matters as much as the route forward. Conditions that return a process to a more supervised stage should be defined at the same time, whether a spike in corrections, a single serious error, or a material change to the underlying procedure. Without a defined reversal, autonomy only ratchets upward and quietly outruns the evidence supporting it.

Expansion should also be scoped by value or consequence rather than applied wholesale. A common arrangement grants independence for low value or reversible actions while retaining approval above a threshold, so the agent gains real reach on routine volume without ever making the small number of decisions that would be expensive to unwind.

## Key points

- Independence increases in stages as evidence accumulates
- Reliability is process specific and does not transfer
- Entry criteria stated as numbers before the stage begins
- A defined route back to supervision is required
- Scope autonomy by value or reversibility, not wholesale

## In practice

An agent handling expense claims starts by flagging claims it would approve while a person approves everything. After three hundred claims with agreement above ninety eight percent, it approves claims under a set amount independently and routes the rest. Anything above the threshold, any claim from a new supplier, and any duplicate still requires a person. Two consecutive weekly reviews with errors return it to full approval.

## Related terms

- [Shadow Mode](/en/glossary/shadow-mode)
- [Scope Of Work](/en/glossary/scope-of-work)
- [Quality Assurance Review](/en/glossary/quality-assurance-review)
- [AI Maturity](/en/glossary/ai-maturity)

[Back to the AI Glossary](/en/glossary)