The AI productivity paradox: Why feeling faster isn't the same as going faster
summary
Claim: AI often feels faster but doesn’t deliver end-to-end throughput. Meta’s trial reports a 39% gap between perceived boost and measured outcome (developers feel ~20% faster while being 19% slower), GitHub shows over a billion contributions and 29% YOY merged-commit growth, and DORA finds throughput down ~1.5% per 25% AI adoption. Real gains come from treating engineers as customers and investing platform foundations: self-service deterministic gates, automated pipelines, nondeterministic checks like AI-driven exploratory testing, and a skills marketplace for consistent agent behavior.
AI-written summary · about
chapters (8)
- 0:00AI productivity paradox intro
- 1:13Why AI tools feel faster
- 3:41AI productivity data
- 6:28Where AI slows teams
- 9:58Platform engineering key
- 10:19Alex’s AI journey at Trading Technologies
- 14:52Steps to integrate AI in SDLC
- 21:11Key takeaways
related talks
44mSpeed Is Easy, Trust Is Hard: AI's Real Impact on Developer Productivity
46mMeasuring the Impact of AI on Developer Productivity
18mDoes AI Actually Boost Developer Productivity? (100k Devs Study)
18mHumans + AI: The Future of Software Development Teams
22mThe AI Productivity Paradox in Engineering
32mAI and Engineering Closing the Gap Between Expectations and Reality 1
45mAgentic AI Coding Practices for Speed with Quality
31mReflections of AI: A Trilogy in 4 Parts