# Goodhart's Law Becomes Automatic Under AI Tool Use ## Claim AI coding tools turn Goodhart's Law from something engineers do on purpose into something that happens by default. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Holds wherever an activity-based metric (PR count, deployment frequency, lines of code, cycle time) sits downstream of AI-assisted generation. - Doesn't claim every engineer using AI is gaming metrics intentionally — the point is the opposite: inflation now happens without intent, which is what makes it hard to catch with the usual "look for suspicious patterns" review. - Doesn't establish a magnitude — the source gives no quantified inflation rate, only the mechanism. ## Evidence The LeadDev piece reframes Goodhart's Law for the AI era: before AI, inflating a metric like deployment frequency or PR volume required deliberate manipulation. With AI coding tools, an engineer using the tool normally — no gaming intent at all — produces the same inflationary effect as a side effect of ordinary use. The distinction matters practically: a team investigating "why did this metric spike" can no longer treat intent as the diagnostic signal. > "In the pre-AI era, gaming a metric required deliberate effort. Now it's a side effect of normal tool use." ## Links - [[AI & the Future of Knowledge Work MOC]] — metric-gaming stops requiring intent once a tool optimises without being asked - [[3. Resources/Permanent Notes/AI Activity Metrics Measure Tool Throughput Not Engineering Value]] — extends; that note names what activity metrics measure once AI is involved, this note names the mechanism (automatic Goodharting) that gets you there - [[3. Resources/Permanent Notes/A Metric Is Vanity Only Against Your Own Outcome Data]] — applies-in; automatic Goodharting is exactly the failure mode that only an outcome-regression test can catch, since intent-based scrutiny no longer works - [[3. Resources/Permanent Notes/Rewarding AI Usage Manufactures Usage Not Value]] — contrasts; that note is about a metric being *deliberately* gamed once a reward is attached, this note is about inflation with no reward and no intent at all - [[3. Resources/Permanent Notes/A Completion Count Measures Compliance Until It Is Paired With a Build Count]] — extended by: a finished-course number says who completed, not who built; report it beside something the learner made and kept (Keegan). ## Tests / falsifiers - Track whether teams that switch from activity metrics to the three outcome metrics this article proposes (MTTR, business/customer outcomes, escaped defect rate) actually stop seeing the inflation pattern. - Look for a counterexample: an AI-assisted team whose activity metrics stayed flat or dropped despite heavy tool use — this would mean the "automatic" framing overstates how universal the effect is. ## Maintenance - Last tended: 2026-08-16 - Next review: 2026-11-16 ## Source - [[3. Resources/Readwise/Articles/The 8 software engineering metrics AI broke]] — LeadDev, Michael Hill; names the pre/post-AI shift from deliberate to automatic metric gaming.