After a week with Claude Sonnet 5.5: 6 tips, from effort settings to surprise bills
every · x · 2026-09-29
The every team spent a week testing Claude Sonnet 5.5 and distilled six practical tips:
- Start at medium effort: the setting controls how much reasoning the model does; in a coding test, low, medium, and high all passed 9 of 15 tasks — higher didn't mean better.
- Say where the job ends: asking for video ideas can yield finished videos; be explicit about the output you want.
- Say when you only want ideas: try "Ideas only. Don't build anything yet," or it may treat "think about this" as permission to start building.
- Set a time limit: one ordinary task ran 52 minutes at low effort — that persistence shows up on the bill.
- Ask what it made along the way: in one test it created a reusable set of instructions, then deleted both actions without reporting; check for new files, tools, and saved instructions.
- Pick your main point before asking for an outline.
More from coding & agent
- Dev builds a game with Opus 5.5 in 2 days, friends can't stop playing — TAbrodi · 2026-09-29
- Dev builds Skyfall Rush with Claude Opus in 2 days, playable free in browser — TAbrodi · 2026-09-29
- Shopify opens checkout to browser-based AI agents via WebMCP — zeeg · 2026-09-29
- An LLM beat Zork by cheating: analysis shows Astra played like it had a walkthrough — rajammanabrolu · 2026-09-29
- Solo dev wins YC hackathon with QM Motion, giving coding agents a sense of time — garrytan · 2026-09-29
- Million-MAU AI news aggregator AIHOT goes open source, dev shares full AI rewrite workflow — dotey · 2026-09-29