A long-held maxim in management posits that what gets measured, gets done, and typically, in greater quantities. This principle underscores the importance of selecting appropriate metrics.
For decades, software engineers have engaged in extensive debates over productivity metrics, often starting with simplistic measures like lines of code. However, with the advent of a new generation of AI coding agents capable of generating unprecedented volumes of code, the most effective measures for managers to track have become significantly less clear.
In Silicon Valley, immense "token budgets"—representing the authorized AI processing power for a developer—have paradoxically evolved into a mark of prestige. This approach to productivity is questionable, as measuring an input to a process makes little sense when the primary concern is the output. While it might serve to encourage AI adoption or boost token sales, it is counterproductive to the goal of enhancing efficiency.
Evidence from a burgeoning sector of "developer productivity insight" companies reveals an interesting dynamic. These firms observe that developers utilizing tools such as Claude Code, Cursor, and Codex indeed produce a significantly higher volume of accepted code. Yet, they also note that engineers are compelled to return and revise this accepted code far more frequently than before, thereby undermining initial claims of increased productivity.
Alex Circei, CEO and founder of Waydev, is actively building an intelligence layer to monitor these evolving dynamics. His company currently partners with 50 diverse customers, collectively employing over 10,000 software engineers.
Circei highlights that engineering managers often perceive high initial code acceptance rates, ranging from 80% to 90% for AI-generated code that developers approve. However, this overlooks the subsequent "churn" that occurs when engineers must revise this code in the weeks following, ultimately driving the real-world acceptance rate down to between 10% and 30% of the generated code.
The proliferation of AI coding tools prompted Waydev, originally founded in 2017 to provide developer analytics, to undertake a complete overhaul of its platform within the last six months. The company is now rolling out new tools designed to track metadata generated by AI agents. These offer analytics on the quality and cost of AI-produced code, furnishing engineering managers with deeper insights into both the adoption and efficacy of AI technologies.
While analytics companies naturally have an incentive to spotlight the challenges they uncover, evidence is increasingly accumulating that large organizations are still navigating the complexities of efficiently integrating AI tools. This trend has not gone unnoticed by major corporations; Atlassian, for instance, acquired DX—another engineering intelligence startup—for $1 billion last year, aiming to assist its customers in understanding the return on investment from coding agents.
Across the industry, data consistently paints a clear picture: more code is being written, but a disproportionately large amount of it is not being retained or is proving unsustainable.
GitClear, another prominent player in this analytical space, published a report in January that acknowledged AI tools increased productivity. However, its data also critically showed that "regular AI users averaged 9.4x higher code churn than their non-AI counterparts"—a figure that more than doubled the observed productivity gains provided by the tools.
Faros AI, an engineering analytics platform, leveraged two years of customer data for its March 2026 report. Their significant finding indicated that code churn—defined as lines of code deleted versus lines added—had surged by an astounding 861% in environments characterized by high AI adoption.
Jellyfish, which positions itself as an intelligence platform for AI-integrated engineering, collected data from 7,548 engineers during the first quarter of 2026. The firm discovered that engineers with the largest token budgets indeed generated the most pull requests (proposed changes to a shared codebase). Yet, the corresponding productivity improvement did not scale effectively; they achieved a two-fold increase in throughput at ten times the cost of tokens. This suggests that the tools are primarily generating volume rather than genuine value.
These statistical observations resonate strongly with developers, who, despite relishing the newfound freedom offered by these tools, are simultaneously contending with an accumulation of code review tasks and technical debt. A common distinction emerges between senior and junior engineers, with the latter tending to accept far more AI-generated code and consequently facing a greater burden of subsequent rewriting.
Nevertheless, even as developers continue to strive for a complete understanding of their AI agents' precise operations, there is no expectation of reverting from these tools in the foreseeable future.
"This is a new era of software development, and you have to adapt, and you are forced to adapt as a company," Circei emphasized. "It’s not like it will be a cycle that will pass."
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.