There's an old management cliche that what gets measured gets managed, and software engineers have been debating how to measure themselves for decades, starting with the classic 'lines of code' metric. Now, as AI coding agents like Claude Code, Cursor, and Codex flood repositories with more code than ever, managers are left wondering what, exactly, they should be counting. In a bizarre new status game, enormous 'token budgets' - the amount of AI processing power a developer is authorized to burn - have become a badge of honor in Silicon Valley, which is a profoundly weird way to think about productivity. Measuring an input makes little sense when you presumably care about the output, unless your goal is simply to encourage more AI adoption or, conveniently, sell more tokens.
A new class of 'developer productivity insight' companies is providing the data to puncture this hype. They're finding that while developers using AI tools generate a lot more accepted code, they also have to return to revise that 'accepted' code far more often, which seriously undercuts any claims of a productivity boom. Alex Circei, CEO and founder of Waydev, is building an intelligence layer to track these dynamics; his firm works with 50 different customers employing over 10,000 software engineers. He notes that engineering managers see initial AI code acceptance rates of 80% to 90%, but they miss the subsequent churn, which drives the real-world, lasting acceptance rate down to between just 10% and 30% of the generated code.
The rise of these tools led Waydev, founded in 2017 to provide developer analytics, to totally rework its platform in the last six months. The company is now releasing new tools that track the metadata generated by AI agents, offering analytics on the quality and cost of their code to give managers insight into both AI adoption and its actual efficacy. While analytics firms have a vested interest in finding problems to solve, the evidence is mounting that large organizations are still fumbling their AI tool usage. Major players are taking notice - Atlassian acquired another engineering intelligence startup, DX, for $1 billion last year to help its customers understand the return on investment on coding agents.
The data from across the industry tells a consistent and slightly depressing story: more code is being written, but a disproportionate amount of it isn't sticking. GitClear published a report in January finding that while AI tools increased productivity, its data showed 'regular AI users averaged 9.4x higher code churn than their non-AI counterparts' - more than double the productivity gains the tools provided. Faros AI, drawing on two years of customer data for its March 2026 report, found that code churn - lines deleted versus lines added - had increased by a staggering 861% under high AI adoption.
Jellyfish, an intelligence platform for AI-integrated engineering, collected data on 7,548 engineers in Q1 2026. Its finding was particularly telling: engineers with the largest token budgets produced the most pull requests, but the productivity improvement didn't scale. They achieved two times the throughput at ten times the cost of tokens. In other words, the tools are generating volume, not value. These statistics ring true to developers who report that code review and technical debt are piling up, even as they enjoy the newfound freedom to generate code at a breakneck pace.