上下文工程实战指南:用四大策略构建生产级 AI 智能体
掌握上下文工程四大核心策略 Write、Select、Compress、Isolate,用 LangGraph + Python 代码实例构建生产级 AI 智能体。含三种上下文失败模式分析、Token 预算管理实践,以及 Anthropic 和 LangChain 的生产经验总结。
Marcus has been gluing systems together for twelve years - first as an integrations engineer at Tray.io, then four years at MuleSoft (post-Salesforce acquisition) leading a team that built connectors for regulated-industry customers. He moved full-time into LLM orchestration in 2023 after a side project - an n8n workflow that triaged his consulting firm's intake email - replaced an actual headcount. He focuses on the boring middle layer: idempotent webhook receivers, dead-letter queues for tool-call failures, and getting Temporal to play nicely with OpenAI's Assistants API. He's published two open-source n8n community nodes (one for Pinecone hybrid search, one for Anthropic prompt caching) and contributed retry-backoff improvements to the LangChain JS repo. Lives in Atlanta. Writes about what actually breaks in production agents, not what looks good in a demo.
掌握上下文工程四大核心策略 Write、Select、Compress、Isolate,用 LangGraph + Python 代码实例构建生产级 AI 智能体。含三种上下文失败模式分析、Token 预算管理实践,以及 Anthropic 和 LangChain 的生产经验总结。