Levels and autonomy tiers

Both scales are proposals for this framework, not industry standards. Most organizations should aim for level 2 or 3, not level 4.

Readiness levels

Level	Name	What it looks like
0	Ad hoc	Individuals use AI; no shared process; personal gains do not reach delivery
1	Assisted	Tools are licensed and widely used; process is unchanged; the queue forms at review
2	Gated	Review and test gates adapted; AI review on every change; some outcome metrics
3	Governed	Risk-tiered merge policy; independent verification; named owners; outcome metrics drive decisions
4	Adaptive	Platform built for agent volume; most low-risk changes merge without humans; people focus on intent, risk and audit; reassessed continuously

Anthropic's published setup is an example of level 4 on the dimensions it has described. In its Claude Code repositories, changes merge with at most one human in the loop, who may be the author, and the company reports no increase in reverts or incidents. It is an extreme case, and its figures are the company's own.

Autonomy tiers

Tiers set how much human involvement a change needs. The examples are illustrative; each organization sets its own.

Tier	Example changes	Human involvement	Evidence required
T0 Low risk	Docs, test-only changes, internal tooling	None beyond AI review	Automated checks, AI review, automatic rollback
T1 Standard	Routine features in well-tested services	One human, who may be the author	Passing tests, AI review, canary release
T2 Elevated	Shared libraries, schema migrations, authentication, payments	Named owner plus a second reviewer	T1 plus staged rollout and a written rollback plan
T3 Critical	Safety-critical, regulated, financial ledgers, security-sensitive	Explicit sign-off or change board	T2 plus audit trail and independent verification

Scoring rules

1. A dimension score is the average of its three statements, from 0 to 4.
2. The readiness level is the lower of two numbers: the whole-number part of the overall average, and the whole-number part of the weakest dimension plus one. It never exceeds 4.
3. The autonomy floor is the lowest score among D2, D3, D5 and D6.
4. Each tier is Ready, Conditional (extra evidence or monitoring first) or Not yet:

Tier	Ready at a floor of	Conditional at a floor of
T0 Low risk	2.0	1.5
T1 Standard	2.5	2.0
T2 Elevated	3.0	2.5
T3 Critical	3.5	3.0

T3 changes always keep explicit sign-off. Ready at T3 means the evidence supports agent-authored critical changes reaching sign-off, not that sign-off can be dropped.

Proposals

The thresholds and the level rule are starting values. They change only through the calibration log, with the evidence behind each change.