The Velocity Mirage: Why AI Teams Ship More PRs but Deliver Value Slower

Your dashboard has never looked better. Pull requests merged are up almost double. Commits per engineer are climbing. Lines of code are pouring in. Every activity chart points up and to the right, and the AI coding tool you rolled out six months ago looks like the best money the org has spent all year.

Then you check the one number nobody puts on a slide: how long it takes for a customer to actually get something they asked for. It hasn't moved. In some quarters it's worse. The team is producing more of everything except the thing the business buys.

This is the velocity mirage. The metrics that are easy to count went up, and the outcome that matters quietly went the other way. It is one of the most expensive measurement failures in engineering right now, because it looks exactly like success.

The mirage is not a hypothetical. When Faros AI analyzed telemetry from more than 10,000 developers across 1,255 enterprise teams, the heaviest AI adopters merged 98% more pull requests — and watched their PR review times balloon by 91%. A Carnegie Mellon study of 807 repositories that adopted Cursor found a 281% surge in lines of code in the first month that decayed to roughly zero advantage by month six, while code complexity rose 41% and static-analysis warnings rose 30% and never came back down. The throughput is real. The value is not following it.

Why Activity and Outcome Have Come Unbolted

For most of software history, activity was a decent proxy for output. Writing code was the bottleneck, so a team that wrote more code generally delivered more. Counting commits and PRs was lazy, but it correlated with the thing you cared about, and the correlation was strong enough that nobody got badly burned.

AI broke that correlation. When generating code becomes nearly free, the volume of code stops telling you anything about delivered value. You can produce a 4,000-line pull request in an afternoon. That number says nothing about whether the feature works, whether anyone can review it, whether it will need to be rewritten next week, or whether the customer wanted it at all.

What actually happened is that the bottleneck moved. It used to sit at code authorship. Now it sits downstream — in review, integration, testing, and the human comprehension required to safely ship. Generating code faster doesn't relieve that bottleneck. It feeds it. Every extra line an AI produces is a line a human still has to understand, review, and own when it pages them at 3 a.m. The activity metrics measure the part that got cheap. The outcome metric measures the part that didn't.

This is why "PRs merged" has become a vanity metric. It counts the easy half of the pipeline and ignores the half that now governs your delivery speed.

The Metrics That Hide the Slowdown

The dangerous thing about the velocity mirage is that the misleading metrics aren't wrong — they're accurately measuring the wrong thing. Here's where each one leads you astray:

• Commits and PRs merged go up because authorship got cheap. They tell you how busy the keyboard is, not how much value crossed the finish line.
• Lines of code is worse than useless in the AI era. GitClear's analysis of over 200 million changed lines found copy-pasted code surpassed refactored code for the first time ever in 2024, and code churn — lines rewritten or reverted within two weeks — nearly doubled from 3.1% to 5.7%. More lines increasingly means more duplication and more rework, not more capability.
• Cycle time on individual PRs can even shrink, because each PR feels fast to write. But end-to-end lead time — the wall-clock from "customer asked" to "customer has it" — stretches as those PRs pile up in a review queue no human can keep pace with.

The 2024 DORA report quantified the trap precisely. As AI adoption rose, researchers estimated a 1.5% decrease in delivery throughput and a 7.2% reduction in delivery stability for every meaningful increase in reliance. Developers felt more productive and individually were producing more. The system delivered less and broke more often. The gap between the two is the mirage, measured.

Loading…