Original data
What 68 AI loop patterns reveal
The AI Loop Library patterns suggest practical loops are less about open-ended autonomy and more about bounded, verifiable delegation.
Answer-first finding
In the first 68 AI Loop Library patterns, Engineering is the largest category with 21 loops, and 58 of 68 patterns include approval, human review, or approval-adjacent controls.
This is a small internal field dataset, not a market-wide benchmark. The useful signal is directional: practical loops cluster around verification, handoff, and bounded autonomy.
Category distribution
| Category | Loops |
|---|---|
| Engineering | 21 |
| Operations | 10 |
| Growth | 7 |
| Content | 6 |
| Evaluation | 6 |
| Personal Ops | 5 |
| Knowledge | 4 |
| Security | 4 |
| Design | 4 |
| Strategy | 1 |
Difficulty distribution
| Difficulty | Loops |
|---|---|
| Intermediate | 40 |
| Beginner | 15 |
| Advanced | 13 |
Verification signals
| Signal | Loops mentioning it |
|---|---|
| approval | 52 |
| source | 41 |
| test | 19 |
| log | 14 |
| human | 14 |
| screenshot | 8 |
| diff | 6 |
| metric | 5 |
| lint | 3 |
| browser | 3 |
| coverage | 2 |
Interpretation
The library’s strongest pattern is evidence. Good loops are designed around source checks, tests, diffs, logs, screenshots, browser checks, metrics, and explicit approval boundaries.
That matches the citation strategy: pages should expose compact, extractable facts instead of hiding useful detail inside interactive UI alone.