AI & Code: 39% Boost in Productivity, But At What Cost?

Every week, Cleo Insight offers 5 minutes to turn the latest research papers into actionable insights.

The numbers are clear: a 39% increase in code integrations thanks to AI agents. A remarkable leap in productivity is silently changing how we create software. But under this high performance lies a fascinating paradox: experienced developers are adopting these tools more effectively, while others find themselves 19% slower despite believing they're 20% faster. 

This week, we dive into how AI isn’t just replacing human work, but profoundly reshaping the nature of expertise and redistributing the value in a rapidly evolving field.

AI Agents, Productivity, and Higher-Order Thinking: Early Evidence From Software Development (University of Chicago - Booth School of Business Nov 2025) 

The University of Chicago study reveals an impressive 39% rise in weekly code integrations after adopting AI programming agents, with no drop in quality. The analysis examines an AI-assisted coding platform that generates code from natural language instructions. Remarkably, experienced professionals are 6% more likely to accept the code generated by agents compared to juniors. AI is transforming the production process, shifting from syntax-related activities (typing code) to semantic activities (instructing and evaluating).

Takeaway: AI agents significantly boost code production, with an amplified effect among senior developers who can better direct them.

How AI Agents Are Transforming Software Engineering and the Future of Product Development (METR, Nov 2025)

This study shatters our certainties about AI: 16 expert developers, convinced they were 20% more productive with AI, were actually 19% slower. This troubling phenomenon, captured over 143 hours of filmed work, reveals our blindness to technology. The microscopic analysis of developers’ behavior unveils why: they spend considerable time understanding, adapting, and debugging AI suggestions, underestimating this cognitive cost. Even more revealing, the 25% of developers who truly gained efficiency shared one common trait: prior extensive experience with specific tools. This massive cognitive dissonance between perception and reality fundamentally questions the hurried technological adoption policies in companies.

Takeaway: AI creates a dangerous illusion of efficiency, demanding rigorous and objective evaluation instead of adoption based on subjective impressions.

💡 For you – a summary of both studies

• Transforming skills is inevitable: professional value is shifting from technical production to the ability to conceptualize problems and evaluate AI-generated solutions, radically transforming the most sought-after profiles.

• AI widens expertise gaps: contrary to the notion of democratizing skills, AI agents amplify the advantage of experts capable of formulating high-level strategic instructions.

• The trap of subjective evaluation: the striking gap between perceived and actual productivity demands a fundamental rethink of our technological evaluation metrics, promoting objective measures over significant periods.

At Cleo, we help companies turn information monitoring into a collective reflex. Our goal isn’t to provide yet another tool, but to help you integrate a true decision-making partner into your teams.