Two tasks on a spring break Saturday. I carved out about an hour and a half of computer time to get the content synthesis pipeline running and generate interactive labs for the free tier. Minimal supervisory effort: eight minutes of prompting produced a week and a half of human-equivalent output.
Task Log
| # | Task | Human Est. | Claude | Weeks | Factor | Sup. Factor |
|---|---|---|---|---|---|---|
| 1 | Generate 66 interactive labs across 35 educational domains | 40h | 35m | 1.0w | 68.6x | 800.0x |
| 2 | Set up content synthesis pipeline and synthesize 35 educational domains (environment fixes + batch processing + quality review + lesson generation) | 20h | 50m | 0.50w | 24.0x | 240.0x |
Aggregate Stats
| Metric | Value |
|---|---|
| Total tasks | 2 |
| Human-equivalent hours | 60h (7.5 working days) |
| Claude wall-clock time | 85m (1.4h) |
| Supervisory time | 8m |
| Tokens consumed | ~595,000 |
| Weighted avg leverage factor | 42.4x |
| Weighted avg supervisory factor | 450.0x |
Analysis
The lab generation task at 68.6x carried most of the leverage. Once the domain specifications and lab framework exist, generating 66 labs across 35 domains is pure content stamping. The AI reads each domain's goal hierarchy and produces labs that exercise the leaf nodes. Three minutes of prompting, 35 minutes of execution, and the result would take a curriculum developer a full work week.
The pipeline setup task at 24.0x was lower because it involved debugging: fixing a virtual environment, resolving dependency conflicts, patching a batch script, and then running the actual synthesis and quality review passes. Environment troubleshooting is where AI leverage drops. The machine is fast at generating code but still has to iterate through build-fix-retry cycles just like a human would, only faster.
The supervisory leverage of 450.0x is the standout number. Eight minutes of my time on a Saturday afternoon turned into seven and a half days of engineering output. Spring break barely noticed.