AI Coding Tools Linked to Longer Work Hours and More Post-Release Bugs, Studies Find
Key Takeaways
- ▸90% of surveyed developers use AI at work, but Google's DORA report found AI usage correlates with increased software delivery instability and more frequent code rollbacks
- ▸Multiple studies show developers working longer hours with AI tools, including a 19.6% increase in out-of-hours code submissions despite productivity gains
- ▸UC Berkeley research found employees using AI began working during breaks and lunches, raising concerns about burnout and unsustainable work practices
Summary
Despite widespread adoption and claims of productivity gains, artificial intelligence coding assistants may be creating as many problems as they solve for software developers. According to multiple recent studies, including Google's DevOps Research and Assessment (DORA) team survey of nearly 5,000 technology professionals, while 90% of developers now use AI tools at work and over 80% report productivity improvements, these gains come with significant drawbacks. The DORA report found that increased AI usage correlates with higher "software delivery instability"—meaning code is more frequently rolled back or patched after release due to unexpected issues.
Research from UC Berkeley's Haas School of Business, published in the Harvard Business Review, revealed that employees at one U.S. tech company worked longer hours, took on more tasks, and worked at a faster pace after adopting AI coding tools. Developers reported using AI during lunch breaks and meetings, with former downtimes becoming less refreshing. A separate study by Multitudes analyzing over 500 developers found a 27.2% increase in merged code contributions, but also a troubling 19.6% rise in out-of-hours work submissions.
The findings suggest that while AI can automate tedious coding tasks and enable "vibe coding"—where even inexperienced developers create prototypes by describing intentions to AI—it doesn't eliminate the need for human oversight. Developers still must verify AI-generated code, handle edge cases not reflected in training data, and fix bugs that slip through. Industry experts warn that without proper management, AI tools risk exacerbating burnout and stress among software engineers, particularly as companies pressure workers to "do more with less" following widespread tech layoffs.
- AI-generated code still requires extensive human verification and debugging, as tools struggle with edge cases and specific business requirements not in training data
Editorial Opinion
The AI coding assistant paradox reveals a fundamental truth about automation: tools that promise to save time often just raise performance expectations. What's particularly concerning is the industry-wide pattern of deploying AI alongside layoffs and efficiency mandates, effectively using these tools to squeeze more output from fewer workers rather than genuinely improving work-life balance. If the tech industry can't manage AI's impact on its own highly-skilled workforce, it bodes poorly for AI's deployment across other sectors.
