Track 6: AI Agents, Agent Harness Engineering
Evaluate a Harness: Same Model, Different Harness
Measure task success, pass^k and cost per solved task across harness changes.
Get notified on YouTube Udemy certificate: coming soon
Course
Udemy: coming soon
Tools
- Inspect (UK AI Security Institute) (open-source, free)
- Docker (freemium)
Watch first
Then try
In this module
- The Eight Parts of an Agent Harness
- How Coding-Agent Harnesses Are Built (Claude Code, Codex)
- Skills and Plugins: Package Know-How for Agents
- Hooks: Run Your Own Code Before and After Every Tool Call
- Human Approvals: Permission Modes and Allow/Deny Rules
- Context Engineering for Long Tasks: Compaction and Notes
- Sandboxing Agents: Containers, Files, Network, Secrets
- Project: Build a Mini Coding-Agent Harness
- Evaluate a Harness: Same Model, Different Harness
- Cost and Latency Budgets for Agents
The course this lesson belongs to