Trigpoint
Capability · Judgments

Review code and architecture

4 concepts across 2 shelves.

This is a reading path: read in numbered order. The first three concepts cover the limits of a person reviewing AI output; the fourth covers where to draw a boundary.

№ 2026.10 · 01§ Human-Automation Interaction
01

Vigilance decrement

Subjects watching a clock hand for the occasional double jump soon began to miss it.

Use it to judge whether a review design that relies on someone watching will hold; if not, switch to rotation, spot checks and pushing anomalies to the person.

02

Automation complacency and bias

Missing things while monitoring, deferring while deciding: two effects, both rooted in attention.

Use it to judge whether you now check AI output less often because it has been stable.

03

Trust in automation

“The more they trust it, the better”: wrong.

Use it to judge whether you can tell in which situations a review tool is reliable and in which it is not.

04

Modularity and thin crossing points

“Cut along function and the boundary is a good one”: it is not that simple.

Use it to judge whether a module boundary or handoff sits in the right place: can you say what done looks like, is checking cheaper than doing it yourself, and how often does work go back and forth across it.