AI Expert

构建AI智能体的实用建议

18 分钟高级构建者Anthropic自动化

Anthropic. 三位Anthropic工程师逐一讲解他们最常见到的陷阱：智能体不知道何时停止；不修复环境，却在系统提示中堆叠过多指令；以及根本没有必要采用的多智能体设计所带来的成本。建议紧接着Barry的演讲观看；你会从另一个角度认出同样的模式。

AI Expert 注释

这份陷阱清单也可以直接用作评审清单：停止规则、该修提示还是修环境，以及多智能体设计是否真的有必要。

您应该从中获得什么

在添加多个智能体、复杂提示或隐藏状态之前，识别常见的智能体构建陷阱。

先观看或了解

先观看首选视频；本视频假定相同的基础认知，并补充实战中的陷阱。

最后审核：2026年5月18日

继续观看

通过相同的视频学习路径，继续观看下一个精选配套视频。

深入学习

精选的外部课程，帮助您更深入地了解该主题。

Microsoft (open-source, via GitHub Pages)

Copilot Studio Agent Academy

Microsoft Copilot Studio team

The deeper, production-minded counterpart to our beginner no-code pick: a free, open-source, rank-based curriculum that takes you from zero Copilot Studio experience through MCP integrations and multi-agent orchestration, all without writing traditional code. It's the no-code answer to 'now I want to go further than a quick-start,' which our catalog didn't have.

高级Self-paced, multi-phase (hours vary by rank)
Anthropic Academy

Introduction to Model Context Protocol

Anthropic Academy

MCP正在悄然取代AI工具生态中为单一用途开发的工具集成。请直接向标准的创建者学习。学完后，你将构建并部署自己的MCP服务器，把一个LLM客户端连接到该服务器，并理解为何这项标准堪称该领域最接近USB-C的方案。

中级自主安排进度（短课）
DeepLearning.AI

Practical Multi AI Agents and Advanced Use Cases with crewAI

João Moura (Founder, CrewAI)

Doubles as our sales and customer-support vertical pick and a genuinely practical agent-building course: you build an agentic sales pipeline (lead scoring, personalized outreach) and a customer-support data-insights pipeline as two of the five hands-on projects, taught by CrewAI's own founder. Requires basic Python, so it sits with our other builder-track courses rather than the no-code picks.

中级~2h 49m · self-paced (15 lessons)

查看所有 自动化 课程