/AI18h ago

NMSLIB creator Leo Boytsov argues AI agent productivity gains are highly uneven, warning they can introduce chaos in high-stakes projects

Andreas Kirsch notes Amdahl's law governs potential speedups.

120062
Original post

@srchvrs @ziv_ravid Yeah agreed. amdahl's law applies to this as well. However the bigger the chunk of work agents can do and self-verify the bigger the speed up ofc too

Leo Boytsov@srchvrs

Imagine a typical SWE at a typical co. At least half of the time is spent on alignment, meetings, and fighting chaos. No magic will speed them even 2x. Moreover, hastily created and reviewed code does create more tech debt. Google's last year report was a much smaller speed up. Of course, it could be more with modern agents, but I doubt it's 8x. https://newsletter.getdx.com/p/how-much-does-ai-impact-development-speed

3:35 PM · Jun 5, 2026 · 39 Views
Sentiment
Sentiment building, check back later.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS23LIKES1
Leo Boytsov@srchvrs

My impression that gains are very uneven due to reliability issues. Prototyping & research: a huge gain. High or medium-stake codebases, results will vary. If agents increase chaos the chunk of parallelizable work also decreases. It also probably isn't so magical if you don't code in Python or Typescript.

@srchvrs @ziv_ravid Yeah agreed. amdahl's law applies to this as well. However the bigger the chunk of work agents can do and self-verify the bigger the speed up ofc too

16hViews 23Likes 1Bookmarks 0