New research from Faros AI and Google's DORA report reveals that teams using AI coding assistants hit a productivity paradox: they write code 80-98% faster, but spend 91% longer on code review. The bottleneck just moved, and the time savings evaporated into organizational friction.

The numbers are striking. Teams with high AI adoption saw code output surge 98%, with more pull requests merged. Sounds great, right? Except code review times ballooned 91% simultaneously, and average PR sizes increased 154%.

As Google's Addy Osmani puts it in his analysis: "We got faster cars, but the roads got more congested."

Here's what's happening: when code generation becomes cheap and fast, teams produce vastly more code. But human review capacity hasn't scaled proportionally. According to recent survey data, only 48% of developers consistently review AI-generated code before committing, despite 38% finding that "reviewing AI-generated logic actually requires more effort than reviewing human-written code."

The productivity gains from AI coding tools look great until you zoom out and see the whole system. Turns out writing code was never the slow part - understanding and maintaining it was.

Osmani introduces the concept of "comprehension debt" - the accumulating cost of reviewing code you can no longer write from scratch. This creates dependency and potential skill atrophy among teams relying heavily on AI assistance. When developers generate code they don't fully understand, review becomes harder, bugs slip through, and the codebase becomes progressively more opaque.

This isn't an argument against AI coding tools. They're genuinely useful for boilerplate, documentation, and exploring unfamiliar APIs. But the industry narrative that AI will 10x developer productivity is running into the reality that software engineering isn't just typing speed.

The constraint in most engineering organizations isn't how fast you can write code. It's how fast you can verify it works, understand what it does, maintain it over time, and coordinate changes across teams. AI assistants are incredible at the first part. They're not helping with the rest.

Faster generation doesn't equal faster shipping when verification becomes the constraint. The productivity paradox reveals a fundamental truth about software development: the hard part was never the typing.