09 Building Agents
Phase 2 - Agent Core
Stage 9 of 14. Previous: Stage 8 - Agent Architectures. Next: Stage 10 - Multi-Agent Systems.
Goal
Build agents from direct APIs first, then learn when frameworks add value.
Learn
- Building an agent loop from scratch
- Direct LLM API calls
- Parsing model output
- Error and rate-limit handling
- Provider-native function calling and tool use
- LangChain, LangGraph, LlamaIndex, CrewAI, AutoGen, and the OpenAI Agents SDK
- How to choose a framework without losing debuggability
Build
Build the same tool-using agent twice: once from scratch and once with a framework. Compare transparency, speed, debugging, and maintainability.
Exit Criteria
- You understand the agent loop beneath framework abstractions.
- You can decide when a framework is worth using.
- You can handle tool failures, rate limits, and malformed model output.
Checkpoint
Use the Stage 9 checkpoint before moving on.