Three agents, three frameworks, one talk
summary
Compares three concrete implementations for agent‑based search and shows what actually works: a hand‑rolled agent that uses Llama 3.1 (7B) locally and exposes every token so failures are visible, a LangGraph/LangChain approach that wraps planning and tools into reusable graph nodes but pulls many dependencies, and Pydantic.AI that enforces schemas and async run contexts so outputs are validated but brittle with small models; demonstrates two‑turn memory, tool calling, and that local 7B models succeed intermittently.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:07What defines an AI agent
- 4:21LangChain history
- 6:48Pydantic AI evolution
- 9:56Building agent with Ollama
- 16:54Agent in LangChain
- 21:10Agent with Pydantic AI
- 27:59Choosing frameworks based on use case
- 32:52Other frameworks
- 33:34Q&A
related talks
39mTowards Useful, Reliable, and Scalable AI Agents
40mAgents in Action: Advancing Open Source AI Missions
40mBuilding Effective AI Agents Simple, Composable Patterns
49mAI Agents: Collaboration in Real-Life Applications
27mWhy the Best AI Agents Are Built Without Frameworks (Primitives over Frameworks)
34mAI Agents: Transforming Autonomous Decision-Making
02h 08mIntroduction to AI Agents
02h 22mArchitecting and Testing Controllable Agents