AI Usage in Development Hits 93%, Yet Productivity Gains Stagnate at 10%

At this year's Pragmatic Summit, Laura Tacho, CTO at DX, presented compelling research on the integration and impact of AI coding assistants in software development. The findings are based on data collected from 121,000 developers across over 450 companies. A notable 92.6% of developers utilize an AI coding assistant monthly, and 75% engage with these tools weekly.

Despite this widespread adoption, productivity improvements don't exceed 10%. The stagnation is evident as the initial time-saving jump seems to have plateaued, with developers reporting an average of four hours saved weekly, similar to figures from Q2 2025.

Interestingly, AI now authors 26.9% of production code, a rise from the previous quarter's 22%. This highlights a shift where nearly a third of the code merged into production from frequent AI users is AI-authored. Moreover, the onboarding process for new developers has significantly improved, with the time to the 10th Pull Request being halved from Q1 2024 through Q4 2025.

Tacho emphasizes the contrasting impact of AI based on organizational structure. In well-managed setups, AI serves as a "force multiplier," enhancing speed and quality, while struggling organizations find that AI usage often highlights rather than remedies existing issues. A crucial takeaway is the need for strategic AI integrations at an organizational level, beyond isolated individual tasks.

The research also touches on popular AI tools such as Codex. With a recent update to GPT-5.3 Codex, the desktop app has surpassed one million downloads. At Cisco, 18,000 engineers use Codex, leading to a 50% reduction in code review time.