Healthy AI First · Episode 1 · 17 min
It Feels Faster
16 experienced developers expect AI to make them almost a quarter faster. Measured, they are slower, and afterward they still believe they were faster. The opening episode of Healthy AI First tells why our gut feeling is a poor measuring instrument when we work with AI, what other studies say against it, and where the series goes over the next 7 episodes.
Listen on
Companion document
All studies, sources and claims from this episode in one PDF, to read along or look things up later.
Download the companion document (PDF)Chapters
- 0:00Opening
- 0:57Welcome
- 1:44Your Guess
- 2:19The Experiment
- 3:32The First Twist
- 4:55The Second Twist
- 7:24A word from us
- 8:14The Experiment Breaks Apart
- 9:59The Other Side
- 11:27Three Questions for Every Headline
- 12:25What It Costs
- 14:21The Journey of This Series
- 15:53Your Number
Show notes
Sources
- METR (2025): Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. 16 developers, 246 tasks, +19% completion time with AI; expected time reduction beforehand 24%, estimated time reduction afterward 20%. https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/ · Paper: https://arxiv.org/abs/2507.09089
- METR (February 2026): Update on the follow-up study (57 developers, 143 repositories, 800+ tasks). According to METR the estimate is unreliable; 30-50% of developers said they were selectively not submitting some tasks. https://metr.org/blog/2026-02-24-uplift-update/
- Cui et al.: The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers. 4,867 developers; +26.08% completed tasks (pull requests) per week among users. https://papers.ssrn.com/abstract=4945566
- METR (May 2026): Survey of 349 technical professionals; median self-reported value of working with AI 1.4x to 2x. https://metr.org/blog/2026-05-11-ai-usage-survey/
- DORA (2024 and 2025): Accelerate State of DevOps Report / State of AI-assisted Software Development. https://services.google.com/fh/files/misc/2024_final_dora_report.pdf · https://research.google/pubs/dora-2025-state-of-ai-assisted-software-development-report/
- Vella & Blincoe (preprint 2026): two survey waves (October 2024 / April 2025); share of repeat respondents whose developer experience got worse in at least one area rose from 14% to 27%; shift toward verification work. https://arxiv.org/html/2605.23135
- Fan et al. (CHI 2026): 60 participants, three Python tasks; about 22% less completion time with AI, lower reported workload; verification load was statistically associated with rising stress and fatigue (mediation, not proof of cause). DOI 10.1145/3772318.3791176
- Shen & Tamkin (Anthropic, preprint 2026): 52 mostly junior developers learn the Trio library; comprehension quiz 50% with AI vs. 67% without. https://www.anthropic.com/research/AI-assistance-coding-skills · https://arxiv.org/html/2601.20245v1
- Foundation of the series: research report "Healthy AI First Software Development", version 2 (October 1, 2026).
Daniel and the project manager are fictional characters used for illustration.
Healthy AI First is AI-researched and AI-generated, narrated in Björn's voice, recreated with AI, and reviewed by him.
Bringing AI into your team?
Björn is one of the founders of CLN Work. CLN Work helps companies adopt AI in small, verifiable steps: finding where real capacity is, which processes can be automated, and how to use AI interactively. The decisions stay with you.
Book a free 30-minute conversation