We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Lobsters·18d ago·5 min read
We are Changing our Developer Productivity Experiment Design
TL;DR
A recent follow-up to a 2025 study revealed a 19% decrease in developer productivity. In response, a new experimental design is being implemented to better understand the factors affecting productivity.
✦ Why It Matters
Engineers can implement new collaboration tools based on findings to address specific productivity bottlenecks identified in the study.
Key Takeaways
- Reassess how AI tools are integrated into your development processes.
- Be aware of potential biases in productivity studies involving AI.
- Encourage participation in studies to gather more reliable data.
- Explore alternative methods for measuring productivity impacts of AI.
Full Summary
A previous study found that AI tools slowed down experienced open-source developers by 20%. In a follow-up experiment starting in August 2025, researchers aimed to assess AI's impact on productivity with a larger group.
However, many developers opted out of the study, preferring to work with AI, leading to a biased sample. The results indicated a possible speedup of -18% for returning participants and -4% for new recruits, but these figures are likely underestimates.
Developers reported that their task selection and completion rates were influenced by their reliance on AI tools. The study's design will be revised to better capture the true productivity effects of AI tools on developers.