The field guide
Agentic AI engineering, chapter by chapter.
A continuously updated handbook for engineers building useful, reliable, and production-ready agents — grounded in real tools, trade-offs, and lessons learned.
From framework choice to evaluation: the essential patterns for dependable agent systems.
Contents
Issue 01 chapters
LangGraph vs CrewAI vs AutoGen: choosing for production
A decision guide comparing orchestration, state, observability, human-in-the-loop support, testing, and failure recovery across three leading frameworks.
Build an agent with memory and tools from scratch
A complete implementation path from a plain model call to a reliable agent with short- and long-term memory, permissions, traces, retries, and cost controls.
Why agents enter loops: five real failure cases
Five recurring production failures — tool loops, hallucinated actions, context overflow, stale memory, and runaway spend — with detection and recovery patterns.
MCP, Function Calling, and A2A explained
Understand the boundaries between model-native tool calls, MCP servers, and agent-to-agent communication, then choose the right protocol for each layer.
The AI agent production deployment checklist
A practical launch checklist covering secrets, permissions, timeouts, retries, queues, observability, data isolation, incident response, and cost limits.
How to test agent accuracy, cost, and stability
Build an evaluation loop with task success metrics, trajectory checks, regression datasets, latency and token budgets, red-team cases, and production feedback.
Practical
Every chapter ends with a decision, checklist, or working pattern.
Connected
Chapters link back to the frameworks, tools, and research in the directory.
Fresh
One focused issue each month, with clear dates and versioned updates.
Keep exploring