[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-78998":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":25,"hasPages":23,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":33,"readmeContent":34,"aiSummary":35,"trendingCount":16,"starSnapshotCount":16,"syncStatus":15,"lastSyncTime":36,"discoverSource":37},78998,"agentic-ai-system-course","bryanyzhu\u002Fagentic-ai-system-course","bryanyzhu","Use agent to learn agent - A skeleton course on how to design, build, and operate production AI agents ","",null,"JavaScript",365,52,4,2,0,3,31,176,17,76.67,"MIT License",false,"main",true,[27,28,29,30,31,32],"agentic-ai","agentic-workflow","ai-agents","course","system-design","tutorial","2026-06-12 04:01:24","# Agentic System Course - Use Agent to Learn Agent\n\n**Join the [discord channel](https:\u002F\u002Fdiscord.gg\u002FdWSnHAFdpb) if you want to learn and build together! **\n\n---\n\nThis is a 22-chapter skeleton course on how to design, build, and operate production AI agents — written to be read with your own AI partner at your side. **An agentic system** is an AI system that can autonomously pursue goals by planning, making decisions, using tools, adapting based on feedback, having memory, etc — instead of only responding to a single prompt. A few use cases are: \n\n- **Healthcare**: Multiple AI agents can coordinate patient intake, insurance verification, scheduling, lab monitoring, and discharge planning to automate hospital operations and reduce administrative burden.\n- **Customer support**: Specialized AI agents can collaboratively handle ticket triage, knowledge retrieval, sentiment analysis, escalation routing, and follow-up communication to improve support efficiency and response quality.\n- **Supply chain**: Multi-agent AI systems can continuously optimize procurement, inventory, production scheduling, and logistics by adapting in real time to disruptions and changing demand.\n\nSimilar to [Andrej Karpathy's idea file on LLM-wiki](https:\u002F\u002Fgist.github.com\u002Fkarpathy\u002F442a6bf555914893e9891c11519de94f), **this course is giving you the skeleton and your agent will help you put the muscles on it**.\n\n**This course is**:\n\n- A *skeleton* — load-bearing topics, patterns, and decisions, with trade-offs.\n- Written to age slowly. Framework specifics rot fast; architectural patterns do not.\n- A file pair (course + CLAUDE.md) designed for AI consumption as much as human reading.\n\n**This course is not**:\n\n- A step-by-step tutorial. There is no walked-through project.\n- Tied to one stack. The course never says \"use LangChain\" or \"use Pydantic AI.\" Your AI partner suggests the stack that fits your project.\n- A reference manual. When you need an exact API signature, ask your AI or read the docs.\n\n\n---\n\n## How to start\n\nHow you start depends on your background. The course works the same either way — only the prompts differ.\n\n### If you are technical\n\nClone the repo, open it in your usual IDE. At the same time, point your AI agent (Claude code\u002FCodex) at the project root, and try one of these prompts when you study a chapter:\n\n- *\"Give me three real-world examples of where this matters.\"*\n- *\"Suppose you are interviewing me, quiz me on this topic with five follow-up questions, easy to hard.\"*\n- *\"What's a question I should be asking that I haven't?\"*\n- *\"I just read about [pattern X]. I am building [your project]. Translate the pattern into the smallest version that works in whatever language and tools fit, and explain each piece as you write it.\"*\n- *\"Forget my project for a moment — show me how OpenCode (or Hermes Agent, or any leading coding agent) handles this, and what we should borrow from it.\"*\n\nYou can also just point your agent at Ch.22's design canvas and walk through it with your specific project in mind — that's the fastest path from \"I have an idea\" to \"I have a spec.\"\n\n### If you are not technical\n\nDon't worry about understanding any chapter before you do — your AI partner will explain everything as you go.\n\nFirst download this project to your computer. Either directly download the zip file, or if you have `git` installed, run `git clone \u003Cthis-repo-URL>` in a terminal. Then use either Claude code or Codex to open this folder, then start chatting. A few prompts you can try:\n\n- *\"I want to build [your project]. The goal is [the outcome you want for users]. Design an agentic system for me using this course as the guide, put it in a EXEC_PLAN doc. If uncertain, ask for clarification.\"*\n- *\"Can you build a MVP working prototype of [your project] using the patterns in this course? Explain what each piece does in plain language and record it in a detailed implementation plan doc.\"*\n- *\"Give me a PRD for [your project]. The goal is [the outcome you want for users]. If you are unsure about requirements, use Ch.22's design canvas as the spine and ask me.\"*\n\nThe agent writes the code and designs the system. Your job is to see their response, try the product, ask questions, raise features, and push back when something doesn't make sense.\n\n\n---\n\n## Course structure\n\n| Chapters | Theme |\n|---|---|\n| **Ch.00** | How to use this course with your AI partner |\n| **Ch.01–04** | Foundations: one tool call → the loop → tools as contract → prompts & cache |\n| **Ch.05–08** | Memory and state: short-term → long-term → writing & curation → persistence |\n| **Ch.09–11** | Coordination: planning → multi-agent delegation → the harness |\n| **Ch.12–14** | External surface: human-in-the-loop → connectors\u002FMCP → skills\u002FMCP\u002Fsubagents |\n| **Ch.15–17** | Production scale: backend → observability → cost, latency, model strategy |\n| **Ch.18–19** | Quality and ops: safety\u002Fadversarial inputs → operations and forward-deployed |\n| **Ch.20–21** | Agency: proactive agents → self-evolving agents |\n| **Ch.22** | Design canvas: designing your own agent |\n\nThe chapters are ordered so each one only assumes what came before. If you have a clear project, skim chapters that do not apply yet and come back when they do. Suggested learning paths by project shape (coding agent, personal assistant, multi-tenant tool, research agent, just exploring) are in `CLAUDE.md`.\n\nThe goal is not to finish the course. *The goal is to ship something you wanted to ship anyway, and to understand every line of it.*\n\n---\n\n## Optional: reference systems\n\nThe course occasionally points at four open-source systems for grounded examples:\n\n- **OpenCode** — coding agent (terminal-first, typed tools, sessions, compaction)\n- **Hermes Agent** — personal assistant (memory, skills, cron, channels)\n- **OpenClaw** — self-hosted personal-assistant gateway (channel adapters)\n- **Paperclip** — workflow control plane (multi-agent orchestration, durable Postgres state)\n\nYou do **not** need to clone any of these to get value from the course. Most readers — especially non-engineers — will finish without ever looking at the source. These four are sanity checks.\n\nIf you're an engineer and you find yourself wanting grounded answers (*\"how does X actually do Y in code?\"*), your AI partner will offer to clone the relevant repo on demand, or you can run setup beforehand:\n\n```bash\n.\u002Fsetup.sh\n```\n\nThis clones all four reference repos to `references\u002F`. Idempotent (safe to re-run). Env-var-overridable if you want to point at forks or pinned commits. See the script for details.\n\nIf you want to ask questions about other sources, feel free to put them under `references\u002F` and chat about it.\n\n---\n\n## Repository layout\n\n```\ncourse\u002F         — 22 chapters (read-only by design — this is the skeleton)\nCLAUDE.md       — Behavioral guide for your AI partner (read by Claude Code)\nAGENTS.md       — Same content as CLAUDE.md, named for Codex \u002F Cursor \u002F other agents\nREADME.md       — This file\nsetup.sh        — Optional: clone the four reference repos\ndocs\u002F           — Written notes your AI partner produces on request (created on demand)\nquestions\u002F      — Daily Q&A log your AI partner writes automatically (created on demand)\nworkspace\u002F      — Code, exercises, and project builds (created on demand)\nreferences\u002F     — The four reference repos if you (or your agent) clone them\n```\n\n`CLAUDE.md` and `AGENTS.md` are duplicates — same content under two filenames because different agents look for different files by default. Both are structured the same way the course is: a durable, opinionated file any AI assistant can read in one pass and turn into useful behavior.\n\n---\n\n## One last thing before you start\n\nYou do not need permission to start. You do not need to read every chapter before opening your agent. You do not need to know which framework you will end up using. The first thing you build will be small enough that none of those decisions matter — and once it's in front of you, working, the next decision becomes obvious.\n\nThe people shipping the most interesting agentic systems today are not the ones with the most experience. **They are the ones who got into a tight loop with their AI partner first and stayed in it longest.** This course exists so that when you are in that loop, **you know what to ask**.\n\n\n---\n\n## License\n\nSee `LICENSE`. The course content is open for educational use. Reference systems retain their original licenses.\n","该项目是一个关于如何设计、构建和运营生产级AI代理的22章节骨架课程。核心功能包括通过规划、决策、工具使用、基于反馈的适应以及记忆等功能，使AI系统能够自主追求目标，而不仅仅是响应单一提示。技术特点在于强调架构模式而非具体框架，旨在与学习者的AI伙伴共同完成内容填充，以保持知识的新鲜度和实用性。适用于医疗保健、客户服务、供应链管理等场景，通过协调多个AI代理来提高效率和质量。该课程适合希望深入了解AI代理系统设计的技术人员，并鼓励使用个人选择的技术栈进行实践。","2026-06-11 03:57:22","CREATED_QUERY"]