# Measuring AI’s Unintended Consequences in Software Development Assessing the hidden economic impact of AI tools on maintainability, burnout, and long-term productivity. - **Source:** [O'Reilly Radar](https://www.oreilly.com/radar) - **Published:** 2026-11-11 - **Topic:** [careers](https://brifing.si/c/careers) - **Original article:** - **On Brifing:** https://brifing.si/s/58dfb815-2628-482b-b71c-310f742036d5 ## The Hidden Cost of AI While AI accelerates software development, it introduces unintended consequences like maintainability issues, developer burnout, and vendor lock-in. These factors create real economic impacts that can negate speed gains. Leaders must look beyond simple delivery metrics to understand if AI-generated code is sustainable or if it creates long-term technical debt. ## Defining the Cost of Change The cost of change acts as a tax on software evolution. When this cost is low, complex changes are achievable; when high, even simple tasks become difficult. This metric serves as a proxy for capturing the majority of unintended consequences, providing a lens through which engineering leaders can evaluate their development processes. ## Value-Add vs. Muda Value-add percentage tracks effort spent on business-critical features versus non-value-add work, known as muda. Muda includes bug fixes, incident resolution, and manual testing. Distinguishing between discretionary and nondiscretionary muda helps leaders identify if their teams are spending too much time on maintenance, which directly reduces the capacity for new initiatives. ## Implementing Lightweight Tracking Tracking value-add percentage does not require complex time sheets. By categorizing work increments—lasting several days to weeks—as either value-add or muda, managers gain visibility into team efficiency. Reviewing these metrics monthly allows leaders to spot trends, though it often takes three to six months to see clear, actionable patterns emerge. ## Leveraging DORA and Tokens Organizations should integrate DORA metrics like deployment frequency and change fail rate alongside AI-specific data. While token usage is a literal cost, it also serves as an advisory metric. High variance in individual token usage can signal a need for further investigation into how specific developers are utilizing AI tools. ## Modeling Maintainability Risks Maintainability should be modeled as a mandatory tax on development. By estimating future maintenance costs for every month of value-add work, leaders can project available engineering capacity. This model helps determine if AI-driven speed improvements are actually breaking even when accounting for the increased long-term burden of maintaining AI-generated code. ## Tracking Code Churn Lines of code are a poor productivity measure but a useful cost indicator. By categorizing commits as value-add or maintenance, teams can measure gross code churn. Modeling how value-add churn correlates with future maintenance churn provides a data-driven way to forecast the long-term impact of AI on the codebase's health. ## Addressing Vendor Lock-in Lock-in occurs when rising token costs force difficult trade-offs. If AI costs increase, companies may be forced to cut headcount or reduce AI usage. Modeling these scenarios involves updating capacity plans to reflect potential budget constraints, ensuring that the organization remains resilient against shifts in AI pricing or model capabilities. ## Monitoring Developer Burnout AI-induced mental exhaustion can lead to decreased productivity and higher turnover. While difficult to model accurately, leaders should monitor turnover rates relative to historical data and conduct periodic one-question surveys on mental energy. Exit interviews can also provide critical qualitative feedback regarding developer frustration with AI-integrated workflows and toolsets. ## The Risk of Gaming Metrics Value-add percentage is highly susceptible to manipulation. Managers under pressure may defer maintenance to artificially inflate value-add figures, potentially leaving successors with significant technical debt. Leaders must use these metrics to evaluate the overall development approach rather than individual performance to prevent fraudulent reporting and ensure long-term organizational health. ## Judgment Over Automation Metrics and models provide noisy, lagging indicators that cannot predict the future with certainty. They are tools to inform human judgment, not replace it. Decisions regarding AI adoption should be based on a holistic view of these models, balanced by the leadership's strategic priorities and the team's actual operational reality. ## Focusing on Business Outcomes Delivering more software is insufficient if the output does not align with business needs. By measuring unintended consequences alongside development speed, organizations can ensure they are building the right software sustainably. The ultimate goal is to balance immediate delivery speed with the long-term maintainability required for lasting business success. ## Sources covering this story - https://www.oreilly.com/radar/measuring-ais-unintended-consequences/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._