The SI Automation Maturity Model
Five levels, from manual work to self-improving systems. You may claim a level only when you can show the evidence listed for it.
| Level | Name | What it means | Evidence required |
|---|---|---|---|
| SI-0 | Manual | Humans run everything; software assists. | None. |
| SI-1 | Scripted | Fixed workflows with AI inside individual steps. | Documented workflow and error alerts. |
| SI-2 | Agentic | AI plans and acts toward goals; humans approve outcomes. | Defined goals, approval step, action log. |
| SI-3 | Verified-autonomous | AI acts alone inside verified boundaries; every action is logged and reversible. | Verifier with test set, autonomy ladder, audit log, rollback. |
| SI-4 | Self-improving | The system improves its own workflows against independent verifiers, under human-set objectives and limits. | Independent verifier, gated promotion of changes, containment limits, human-held kill switch. |
Most organizations today sit between SI-1 and SI-2. The work of this field is moving them up safely.
How to use it
Pick one automation. Find the highest level for which you can show every item of evidence. The next level's missing items are your roadmap.
Version history
v1.0, October 2026: first public release. Critiques and proposed changes are welcome through GitHub issues.