Faster, Token-Efficient Models Make Us Build More and Think Less
every · x · 2026-10-02
New models like Sol and Sonnet 5.5 are faster and more token-efficient, but that isn't unambiguously good. Developer Jack Cheng used to exhaust his weekly allocation in a day; now it lasts three or four, letting him act on ideas immediately — while spending less time deciding whether they belong. He even began questioning his personal AI tool project itself.
His fix: let new features sit overnight, and before bringing a hard question to AI, write down why it matters and what he expects the answer to be.
More from AGI Musings
- AI safety researcher David Krueger asks: do AI systems have real long-term preferences? — DavidSKrueger · 2026-10-02
- Buck's LessWrong essay on tiers of safety buy-in inside AI labs — anpaure · 2026-10-02
- Andy Matuschak: As software prices drop to zero, is 'everyone learns to code' still the dream? — andy_matuschak · 2026-10-02
- repligate on AI doom writing: 'I don't think they believe a word they're saying' — repligate · 2026-10-02
- App Store's Utility category is now a graveyard of apps AI can generate on demand — signulll · 2026-10-02
- Should you be kind to AI? Author argues cruelty to agents corrodes us — MatthewBerman · 2026-10-02