A structured learning journey into AI agent systems — from single-agent fundamentals to multi-agent orchestration and dynamic agent generation.
This repo documents my hands-on exploration of the agentic AI landscape, including paper reviews, framework comparisons, and experiment notes. The goal is to build a solid foundation for designing and implementing production-ready multi-agent systems.
Everything learned here feeds into my main project: ai-agent-office — a general-purpose framework that dynamically generates AI teams for any industry.
| Topic | Description | Status |
|---|---|---|
| 01-agent-fundamentals | ReAct pattern, tool use, memory, single-agent design | Ongoing |
| 02-multi-agent | Multi-agent collaboration, AutoGen vs CrewAI, orchestration patterns | Not started |
| 03-local-llm | Running Qwen3 locally via Ollama, free GPU setup (Colab/Kaggle + ngrok) | Not started |
| 04-papers | Paper reading notes with analysis | 🔲 Not started |
| resources.md | Curated list of courses, papers, repos, and tools | 🔲 Not started |
| Paper | Key Takeaway | Notes |
|---|---|---|
| ReAct (Yao et al., 2022) | Reasoning + Acting loop as core agent pattern | → notes |
| MetaGPT (Hong et al., 2023) | SOP-driven multi-agent software development | → notes |
| ChatDev (Qian et al., 2023) | Virtual software company with role-playing agents | → notes |
| AgentVerse (Chen et al., 2023) | Dynamic agent recruitment for emergent collaboration | → notes |
| Generative Agents (Park et al., 2023) | Observation-planning-reflection architecture for believable agents | → notes |
| Framework | Best For | Dynamic Agent Creation | License |
|---|---|---|---|
| AutoGen | Multi-agent conversation | ✅ via code | MIT |
| CrewAI | Role-based task pipelines | ✅ via config | MIT |
| LangGraph | Custom stateful workflows | ✅ full control | MIT |
| Microsoft Agent Framework | Production enterprise apps | ✅ graph-based | MIT |
Detailed comparison: → 02-multi-agent/framework-comparison.md
See ROADMAP.md for the full 5-week learning path and PoC timeline.
Courses (Free)
- AI Agents in LangGraph — DeepLearning.AI
- AI Agentic Design Patterns with AutoGen — DeepLearning.AI
Repos
- microsoft/autogen — Multi-agent framework
- geekan/MetaGPT — Multi-agent software company
- OpenBMB/ChatDev — Communicative agents for software dev
- QwenLM/Qwen3 — Open-source LLM with strong agent capabilities
Docs
- Ollama — Local LLM runtime
- AutoGen Documentation
- LangChain Documentation