Tokenmaxxing: A Hidden Productivity Barrier for Developers

There’s an old saw in management: What you measure matters. And, typically, you get more of whatever you’re measuring.

For years, the conversation around software engineering productivity has been contentious, pivoting on metrics such as lines of code. Yet with the advent of AI coding tools, managers face new challenges in determining relevant measures of productivity.

The emerging trend of significant token budgets—essentially, the allotted AI processing capabilities for developers—has become a point of pride within Silicon Valley. However, evaluating productivity through this lens is unconventional and may not align with the desired outcomes. Such measures might encourage AI utilization or token sales but do little for genuine efficiency improvements.

New research from firms in the “developer productivity insight” sector reveals that while tools like Claude Code, Cursor, and Codex lead to an increase in accepted code, they simultaneously result in a higher frequency of necessary revisions. This trend suggests that claims of enhanced productivity may be overstated.

Alex Circei, the CEO and founder of Waydev, is in the process of creating an intelligence framework to better analyze these trends. His company collaborates with 50 organizations employing over 10,000 software developers. (This reporter had not previously met Circei, despite his past contributions to ToolsMixAi.)

Circei states that engineering managers are reporting code acceptance rates between 80% and 90%. However, they fail to account for the necessary revisions occurring within weeks of acceptance, which can reduce the true acceptance rate to between 10% and 30% for AI-generated code.

In response to the rapid evolution of coding tools, Waydev has overhauled its platform after its 2017 inception, now offering solutions that analyze metadata from AI systems. These tools provide engineering managers with insights into the quality and cost-effectiveness of the generated code, ultimately enhancing their understanding of AI implementation and effectiveness.

While it’s in the interest of analytics firms to spotlight issues, a growing body of evidence suggests that large organizations are still grappling with the efficient use of AI tools. Major players like Atlassian have recognized this, exemplified by their $1 billion acquisition of DX, another firm focused on engineering insights, aimed at helping clients understand the ROI of coding agents.

Industry data consistently indicates that while code volume has surged, a significant portion fails to become enduring contributions.

GitClear revealed through a January report that while AI tools boost productivity, “regular AI users averaged 9.4x higher code churn than their non-AI counterparts,” thereby negating some productivity advantages.

Faros AI analyzed two years of customer records for its March 2026 report, discovering an 861% increase in code churn associated with high AI adoption.

Jellyfish, which positions itself as a leading intelligence platform within AI-driven engineering, gathered data from 7,548 engineers in early 2026. Their findings showed that those with the highest token budgets produced more pull requests, though this did not translate to proportional productivity increases. Essentially, tools are generating volume but lacking real value.

Developers themselves voice concerns about accumulating code review burdens and increasing technical debt, even as they enjoy the capabilities of new tools. A notable observation is that junior engineers tend to accept more AI-generated code, which subsequently leads to greater rewriting demands.

Despite these challenges, developers do not see a retreat from AI programming as viable.

“This is a new era of software development, and you have to adapt, and you are forced to adapt as a company,” Circei told ToolsMixAi. “It’s not like it will be a cycle that will pass.”

Analysis: This evolution in engineering practices underscores a critical pivot in how organizations measure productivity. As AI tools proliferate, businesses must reassess their metrics and focus on the long-term quality of code rather than mere volume. Understanding these dynamics could significantly impact resource allocation and development efficiency in the future.