# AI Activity Metrics Measure Tool Throughput Not Engineering Value ## Claim Once agents generate routine engineering artifacts, activity metrics measure tool throughput more directly than engineering value. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - This applies to artifact-count metrics such as commits, pull requests, lines of code, cycle time, and deployment frequency when AI materially produces the artifacts. - It doesn't make every activity metric useless; it removes the inference that more activity necessarily represents more human contribution or customer value. - The source's causal examples rely heavily on practitioner reports rather than controlled data, so the mechanism is stronger than its prevalence estimate. ## Evidence [[3. Resources/Readwise/Articles/The 8 software engineering metrics AI broke]] notes that one engineer with an agent can produce more commits in a day than a team once produced in a sprint, while the metrics that remain useful track outcomes such as escaped defects, recovery time, and customer results. ## Links - [[AI & the Future of Knowledge Work MOC]] — connects activity-measurement limits to evaluation of AI-mediated knowledge work. - [[A Metric Is Vanity Only Against Your Own Outcome Data]] — extends: AI makes the need for an outcome test urgent because ordinary tool use inflates the activity proxy automatically. - [[Prevented Work Is Invisible to Velocity Metrics]] — because the same dashboard overcounts generated artifacts while missing the review labor that stops defective ones from shipping. - [[The Bottleneck Moved From Code Generation to Review and Judgment]] — applies-in: activity metrics keep counting the cheap step after value moves to review and judgment.