Reading Anthropic's retail agent example changed six things in ArcKit in one day
mcraddock · x · 2026-09-03
A pattern you can turn into a check is a rule; one you can't is a preference. The author shows how Anthropic's worked example of two Claude agents for online retail — one helping a shopper, one helping the shop — shares deep structure with ArcKit (an agent producing sign-off documents for architects): both put a model in front of text it shouldn't trust and ask for output people will rely on. Reading it changed six things in ArcKit in a single day; the post details what was learnt and changed.
More from coding & agent
- Magnitude open-sources a local-model inference server that plugs into Claude Code, Codex, Cline and more — magnitudedev · 2026-09-03
- Open-source ai-knowledge-graph turns unstructured text into interactive knowledge graphs via LLMs — tom_doerr · 2026-09-03
- Simulated GitHub run by agents shows AI-agent collaboration follows its own scaling laws — tongshuangwu · 2026-09-03
- Open Notebook: a 30k-star local NotebookLM alternative with 4-speaker podcasts — JafarNajafov · 2026-09-03
- AFK Pilot Cloud launches: run your AI coding agents on a personal cloud machine for $5/mo — PawelHuryn · 2026-09-03
- Context engineering over prompt hacks: a 4C framework for briefing LLMs like new hires — bigaiguy · 2026-09-03