What level is your AI system?

A human initiates, supplies context, evaluates the result, and performs the next action; AI does not own meaningful execution state.

A repeatable trigger follows a primarily predetermined path through AI and deterministic actions, with reproducible results and defined failure behavior.

An agent receives a bounded goal, inspects state, selects permitted tools and next steps, validates minimally, and stops under explicit conditions.

Persistent authoritative state, contracts, gates, retries, failure classification, escalation, audit, permissions, and resumability let ordinary execution survive session replacement.

A control plane dynamically selects deterministic code, models, agents, tools, or runtimes using shared contracts, quality, cost, policy, and observed state.

Sustained operation detects, diagnoses, recovers, verifies, and continues through ordinary failures under tested rollback, budgets, quality objectives, incident evidence, and protected approvals.