The age-old adage that what you measure matters is being tested like never before. In the realm of software development, token budgets have become a badge of honour, yet they're failing to accurately reflect real-world productivity.

According to Alex Circei, CEO of Waydev, developers are accepting 80% to 90% of AI-generated code but frequently having to revise it. This churn drives down the actual acceptance rate between 10% and 30%. Tools like Claude Code, Cursor, and Codex generate more accepted code, yet engineers must often revisit that code, negating productivity gains.

Waydev's reworked platform now offers detailed analytics on AI-generated code, helping managers understand both the benefits and drawbacks of these tools. The data is clear: while more code is being written, a significant portion gets rewritten due to technical debt and review processes.

The rise in token usage has led major companies like Atlassian to acquire engineering intelligence startups for billions, aiming to harness coding agents' potential more efficiently. However, the evidence suggests that large organizations are still grappling with how to use these tools effectively. Companies like GitClear have found that AI users experience 9.4 times higher code churn than non-users.

Despite this, developers are embracing the new era of software development, understanding they must adapt. As Circei notes, 'It’s not like it will be a cycle that will pass.' The challenge lies in reconciling these powerful tools with practical realities and ensuring value is added rather than merely volume increased.