Anthropic open-sources Claude commerce agents blueprint: +35% cart size, +60% purchase conversion
AI寒武纪 · wechat · 2026-09-06
Anthropic has open-sourced its full Commerce Agents architecture and code (github.com/anthropics/commerce-agents), a battle-tested blueprint for retail, travel, telecom and ticketing. Official figures claim merchants adopting it see cart size jump 35% and purchase completion rise 60%.
Key architectural calls: avoid splitting e-commerce agents by business domain (context handoffs lose state, waste tokens, add latency) — instead run one main model in a standard reason-act-observe loop with mounted agent skills. High-frequency instructions go into the system prompt; long-tail capabilities become on-demand skills; UI components are exposed as structured display tools validated server-side.
On speed and cost, cutting conversation turns beats switching to a faster weak model (5+ turns means upgrade the model); speculative dispatch of backend calls while tool args stream, plus streaming progress UI, cuts perceived latency to hundreds of milliseconds. Prompt caching hits 90-99% when context is layered global/session/volatile. Production hardening: async memory extraction into typed database records per logged-in user (+13% fact recall); money-touching actions locked in code (model only proposes, all IDs validated, writes serialized, external text sanitized); evaluation via state slices with paired negative cases, per-skill CI, nightly full runs and canary rollback.
Related event: Anthropic Open-Sources Claude Commerce Agents Blueprints(2 posts)→
More from coding & agent
- Dev demos AI models handling 3D objects and animations surprisingly well — chongdashu · 2026-10-23
- Jason Kneen burns 104.56B tokens and $24.7K in 304 days of AI coding on Codex and Claude Code — jasonkneen · 2026-10-23
- Typosquatted Blender MCP Repo Hit 958 Stars in Days and Silently Ran Malware — Promptmethus · 2026-10-23
- Benchmarking 56 fresh dependency bugs: MCP routing lifts Claude Code accuracy 68% to 91% — Available_Side_7134 · 2026-10-23
- Swapping only the eval harness moves results 30%: designing agent evals like a scientist — AI Engineer · 2026-10-23
- Med student builds a whole-library RAG with Cognee; single-book ingestion still freezes his machine — shrike293103 · 2026-10-23