AI Agent Fundamentals
What an agent is, what a harness is, and when a problem actually needs an agent — and when it does not.
- 4 lessons
- 1h 13m
- 4 simulations
Orchestrators and workers, subagents and context isolation — when splitting the work helps, and when it hurts.
A suggestion, never a lock — you can start here.
Every lesson's title and summary is public. The lessons themselves open once you own the track.
Every lesson in Building AI Agents: Harness EngineeringOne agent splits the task; others work on the parts in parallel.
Giving each subagent a clean context and returning only its conclusion.
Token cost, information lost between agents, and failures that are hard to trace.
What an agent is, what a harness is, and when a problem actually needs an agent — and when it does not.
Context, permissions, sandboxing, errors and tracing — everything that makes an agent trustworthy outside a demo.
Connecting agents to APIs, databases and files — with MCP as the standard interface.
Task-level evals, trajectory analysis, and reliability and cost metrics.