Track 05
AI Engineering & Harnesses
Use agents to build and check software, or design the tools and rules that let a running agent do a job.
Start here
Decide what job the agent will do
Use an agent to build software. Give a coding agent one small change, a concrete example of the expected result, and a check it can run. Review the diff and the check before accepting the change.
Design a system for an agent to operate. Write down one task, such as reading a report. The control plane is the part of the system that checks who is asking, what they can do, and what gets recorded.
Worked examples
Interactive lab
Approval State Machine
See where a proposed change waits for a person and what happens when the rules reject it.
Step through an approvalWorked example · Preview
How to collect ChatGPT Ads reports
Define what the tool can read and what it returns when a result is partial or stale.
Give an agent a bounded reporting tool
References
Reference pattern
Agent control plane
Follow how the system checks an agent’s identity and permissions before it acts.
Trace a report request through the control planeSkills
Code Maintenance Plugin
Run the local example to turn a proposed code change into a plan with verification steps.
Run the coding workflow example in SkillsSkills
Harness Governance Plugin
Run the local example to record how you would evaluate a change to the rules and tools around an agent before adopting it.
Run the harness review example in Skills