ACMF Maturity Model
Agent Capability Maturity Framework - Progressive autonomy with automatic progression based on demonstrated performance and automatic regression on failures.
Four Maturity Levels
OBSERVE
New agents, learning baseline patterns
Observe only, no execution. Agent watches human decisions and learns patterns.
LEARNING
Building confidence, validation phase
Can execute but all actions require human review and approval.
ASSISTED
Proven agents handling routine tasks
Full execution capability, but critical operations still require human approval.
AUTONOMOUS
Mature agents with proven track record
Fully autonomous execution with audit trail only. No human approval needed.
Progression Requirements
An agent graduates from one mode to the next only on demonstrated performance in its current mode. Five criteria are tracked continuously, and every one of them must be satisfied at the same time. The bar rises at each transition: reaching AUTONOMOUS requires more tasks, a longer success streak, and a lower override rate than the earlier transitions.
Sustained success rate
A high proportion of tasks in the current mode completed successfully, measured over the whole mode rather than a recent streak.
Minimum task volume
Enough tasks completed in the current mode for the record to be meaningful. The bar rises with each transition.
Consecutive successes
An unbroken run of successful tasks immediately before graduation.
Low human override rate
Reviewers rarely needed to change or reject the agent's output while it was under review.
Consistent confidence
The agent's own confidence in its results stays high, so graduation is not earned on lucky outcomes.
All Criteria Must Be Met
A strong success rate alone is not enough; every criterion must hold at once. Metrics reset when an agent advances so the new mode is judged on fresh performance.
Automatic Progression & Regression
Automatic Progression
Agents automatically progress to higher maturity levels when they meet all progression criteria.
Performance Tracking
Continuous monitoring of success rate, confidence, and human override rate
Metrics Reset
Performance counters reset upon progression to track new mode behavior
State Persistence
Maturity state survives restarts and redeploys; an agent never silently resets to a higher or lower mode
Automatic Regression
Agents automatically regress to lower maturity levels when they experience too many consecutive failures.
Failure Threshold
A run of consecutive failures triggers automatic regression; a single failure does not
Cascade Regression
AUTONOMOUS → ASSISTED → LEARNING → OBSERVE
Detailed Logging
Regressions logged with full context for post-incident analysis
RIGOR Integration
ACMF mode determines how RIGOR executions proceed, particularly at the point where a generated change is about to be applied.
Human Approval Decision Tree
Execution Status by Mode
- OBSERVE: Nothing is executed; the agent only records what it would have done
- LEARNING: Execution pauses for approval before any change is applied
- ASSISTED: Routine changes proceed; critical operations pause for approval
- AUTONOMOUS: Changes are applied and recorded in the audit trail
State Management
- • Mode is restored automatically on startup
- • Per-agent isolation
- • Metrics tracking: success/failure counts, confidence, override rate
Continue Exploring
Learn more about related architectural components