AI Coding Agents Are Making Developers Work More, Not Less

The Productivity Paradox of AI Coding Agents

AI coding agents like Claude Code promised to make software development easier and faster. Instead, according to a new Bloomberg report citing a UC Berkeley and Yale study, they have kicked off what researchers call a productivity panic — a high-pressure race to build at any cost.

The eight-month study tracked employees at a 200-person tech company where AI tools were voluntarily adopted. The finding? Workers who used AI became more productive, but they did not use that productivity to work less. They worked more — longer hours, broader tasks, and fewer breaks.

How AI Intensified Work Instead of Reducing It

The study reveals a pattern that anyone who has used AI coding assistants might recognize. Product managers started writing code. Researchers took on engineering work. Role boundaries blurred as workers handled jobs that previously sat outside their remit. The AI made it easy to begin tasks that would have previously seemed too far outside one's expertise.

Breaks shrank too. Workers developed a habit of sending one last prompt before lunch or after hours. The always-available nature of AI assistants eroded the natural stopping points that previously structured the workday.

The Burnout Numbers

The consequences are showing up in burnout statistics. 62% of associates and 61% of entry-level workers reported burnout, compared to just 38% among C-suite leaders — the people most likely to be mandating AI adoption. Several participants noted that although they felt more productive, they did not feel less busy. In many cases, they felt busier than before.

This creates a troubling feedback loop: executives see productivity numbers go up, raise expectations accordingly, and workers absorb the increased pace as the new baseline. The treadmill gets faster, but nobody gets off.

The Bottom Line

AI coding agents are genuinely powerful tools. But without intentional company practices around AI use, the natural tendency is not work contraction — it is work intensification. The productivity gains are real, but they are being captured by increased output expectations rather than reduced work hours.

The question is not whether AI makes workers more productive. It clearly does. The question is: productive for whose benefit? When the tools that were supposed to free developers from drudgery instead create new forms of burnout, maybe it is time to rethink the AI productivity narrative.