“Tokenmaxxing” is making developers less productive than they think

image via TechCrunch

April 17, 2026, 6:42 PM

• •AI coding tools are generating more code, but a significant portion isn't 'sticking'.
• •Code churn (lines of code deleted versus added) has increased significantly with high AI adoption.
• •Engineers with larger token budgets produce more pull requests, but productivity gains don't scale proportionally.
• •Senior engineers often deal with less AI-generated code, resulting in less rewriting than junior engineers.

The use of AI coding tools is increasing code generation, but not necessarily improving efficiency, as evidenced by higher code churn rates. Companies are struggling to measure the effectiveness of AI in software development, with token usage becoming a misguided metric of productivity. Engineering managers are seeing high code acceptance rates initially, but the need for subsequent revisions significantly reduces the value of AI-generated code. The industry data indicates that while AI tools are accelerating code production, the overall quality and efficiency gains remain questionable, urging a shift towards more insightful analytics.

Read original article