Anthropic says Claude Code can drop 80% of its system prompt with no coding loss
krishnan · x · 2026-07-27
Anthropic says it removed more than 80% of Claude Code’s system prompt for Claude Opus 5 and Claude Fable 5, with no measurable loss on coding evaluations. The post argues this is less a prompting trick than an agent-architecture shift:
- keep base context light
- place guidance where it naturally belongs
- load verification, code review, domain rules, and rubrics only when needed
- rely on artifacts, tests, code references, and tools as higher-fidelity context than prose
The broader takeaway is that context engineering is becoming information architecture. Teams will increasingly need to decide what belongs in memory, repo files, skills, tool schemas, and dynamically loaded context, or their agent stack turns into a junk drawer.
The attached sketch reinforces the same idea with a graph-based context flow: agent core, tools, memory, skills, verification rubrics, and knowledge base feeding the system as needed.
More from coding & agent
- A 9B Ollama agent can run a fully local DJ radio with tools, memory, and TTS — pinku1 · 2026-07-27
- Bugbot rejects an MCP permission flag because it would break path-scoped isolation — zeeg · 2026-07-27
- One GPT-5.6 agent is guarding a Blink security system while another makes a parody rap album — repligate · 2026-07-27
- An agent got unblocked by reusing a logged-in browser, not stealth tricks — armanidev_ · 2026-07-27
- Paper argues graph topology can become the core operating system for AI agents — theomitsa · 2026-07-27
- Claude Code desktop adds UI markup feedback for smoother visual editing — EricBuess · 2026-07-27