Empirical progression of human-to-AI cognitive delegation from manual direct execution (D0) to full autonomous multi-turn systems (D5).
Aggregated telemetry audited under Zero-Incentive Protocol (ZIP-1.0) & Differential Privacy (ε ≤ 0.5)
| Stage Code & Name | Workload Share % | 90d Delta | Median Turns | Unreviewed Diffs % | Fragility Risk |
|---|---|---|---|---|---|
| D0Manual Baseline Human executes 100% of reasoning, synthesis, and execution. Zero generative AI delegation. | 7.2% | -18.4% | 1 | 0% | MINIMAL |
| D1Assisted Lookup Single-turn factual search, concept definition, and API syntax inquiry. Human reads and manually transcribes. | 14.8% | -12.6% | 2.1 | 8.4% | LOW |
| D2Augmented Co-Pilot Real-time inline code/text suggestions and tab autocomplete. Human verifies and accepts line-by-line. | 22.4% | -4.2% | 4.8 | 24.6% | MODERATE |
| D3Supervised Delegation Agent receives multi-file task prompt, inspects workspace, drafts changes, and generates unified git diff for human review. | 29.6% | +14.8% | 8.6 | 31.8% | ELEVATED |
| D4Autonomous Orchestration Terminal CLI agent loop runs commands, executes unit tests, fixes lint errors, and iterates autonomously until requirements pass. | 18.4% | +42.5% | 16.4 | 44.2% | ELEVATED |
| D5Ambient Autonomy Headless continuous agent daemons triggered by GitHub webhooks, Slack requests, or cron schedules with periodic human telemetry audit. | 7.6% | +88.2% | 34 | 58.6% | CRITICAL |