Notes on measuring the AI-coding era
What still counts when AI does the typing: signals, costs and the line between measurement and monitoring.
How do you measure developer productivity when AI writes most of the code?
Commit counts and lines of code collapse as signals once AI does the typing. What still works: feature-level scoring, declared priorities and durability.
What can Anthropic's admin tools actually show you about Claude Code usage?
The official console and APIs cover org-wide aggregates, and they stop there. Here is what they include, what they exclude, and where the blind spots sit.
How much does AI-assisted development cost per shipped feature?
Monthly model bills tell you almost nothing. Attributing spend through sessions to landed features turns an opaque total into a unit economic you can manage.
Employee monitoring or work measurement? Where the line actually sits
The difference is not tone, it is architecture. A checkable list of what a measurement tool must refuse to do, learned from the monitoring industry’s worst habits.
How to tell who on your engineering team uses AI well
Usage share stops differentiating anyone once the whole team codes with agents. The signal that remains is validation behaviour: who steers, who checks, who ships unread.