Jason Gorman

jasongorman.bsky.social

19 September 2026 05:44

The frontier labs would have us believe AI is dangerous because it's becoming so intelligent, and not because they're letting critical systems be controlled by autocomplete. A toaster would be dangerous if you gave it access to the nuclear launch system.

P(aul) Frazee

pfrazee.com

18 September 2026 17:40

Being able to agent-query the world corpus of specs would, in fact, rule so fucking hard

Why

why.bsky.world

18 September 2026 13:36

I wouldn’t let someone work for me if they refused to use coding agents, shows poor taste

Ethan Mollick

emollick.bsky.social

18 September 2026 13:35

Big paper on AI coding agents using Github data. The early auto-complete tools (like Copilot) led to 2.2x as much code, local agents like original Claude Code led to 7.4x, & current remote coding agents 17.3x(!) But human bottlenecks in the coding process means actual releases "only" went up 30%.

Simon Willison

simonwillison.net

18 September 2026 13:35

This thread of four tweets offers a neat mental model for working with coding agents