Stop hunting for one smartest model: right model per job plus a learning loop compounds
PrajwalTomar_ · x · 2026-09-16
Wrapping up his TokenRhythm thread, Prajwal Tomar argues most people are still shopping for a single "smartest" model — the wrong game. Running 5 businesses, he picks the right model per task and keeps a system that retains lessons from every run, turning daily AI use into a compounding advantage.
He also tested OpenSquilla on an AI-built MVP before client handoff: it found a production risk with the exact code, and when challenged, corrected itself, fixed the issue, and reran the test successfully — with the full recovery visible, not just the final answer.
More from coding & agent
- Transformers.js v4.3.0 ships structured output: force exact JSON Schema in the browser — nicodotdev · 2026-09-16
- Agents slow down as memory notes pile up, unlike humans who speed up with practice — sujingshen · 2026-09-16
- 5 Open-Source AI Agent Repos Exploding on GitHub Right Now — Shruti_0810 · 2026-09-16
- Dev puts Codex voice in CarPlay via iOS app, codes during road trips — athyuttamre · 2026-09-16
- Developer builds an LLM decision API: parallel questions, probabilities and scores for code logic — DevDminGod · 2026-09-16
- Neko Master: open-source dashboard for real-time local gateway traffic auditing hits 4k stars — tom_doerr · 2026-09-16