Jarad DeLorenzo proposed a genuinely useful metric called the Agent Autonomy Ratio (AAR): the time an agent works productively without intervention, divided by the human effort spent preparing the task. By his accounting it has climbed from roughly 0.1 in the chat era to 24+ with modern agentic tools. The trajectory is real, and the metric deserves the attention it's getting.
But it has a blind spot, and the blind spot rewards the most expensive habit in the agent workforce.
The missing denominator
AAR counts prep time as the human cost. It doesn't count review. Which means the highest possible AAR belongs to the person who briefs in thirty seconds and ships whatever comes back unread — exactly the failure mode that principal-agent theory predicts for fluent, confident, unverified output.
The fix is one term:
True Leverage Ratio (TLR) = unattended agent time ÷ (brief time + review time)
Same insight, honest denominator. TLR can't be gamed by skipping review, because skipped review isn't leverage — it's deferred cost, payable with interest when the confidently wrong report reaches your client.
The full dashboard
| Metric | Definition | What it measures |
|---|---|---|
| DQ | 12-scenario assessment, 0–10 | The person's delegation skill — predicts everything below |
| TLR | Unattended time ÷ (brief + review) | Real productivity amplification |
| First-Pass Yield | % of delegated tasks accepted without rework | Brief quality |
| Touch Count | Interventions per delegated task | Handoff discipline |
| Span of Supervision | Median concurrent delegated workstreams | Personal org-chart width |
| Delegation Capital Rate | % of tasks started from an existing brief | Compounding |
The relationships matter more than any single number. Better briefs raise First-Pass Yield, which raises TLR. Reuse cuts brief time, raising TLR again. Span of Supervision is the ceiling TLR eventually hits. And the assessment score — DQ — predicts the lot, because all five operational metrics are downstream of the same three moves: brief, handoff, review.
For teams: run the DQ assessment before and after any AI program, then track TLR and First-Pass Yield as the operating metrics. Those two numbers survive contact with a CFO.