THE ROLE OF THE AI HARNESS ENGINEER

EXAMPLES Example 1: A typical week. On Monday a harness engineer reads the weekend's traces and finds an agent repeating one call. On Tuesday she writes an evaluation case for it and adds a guard. On Wednesday she reviews the permissions of a new tool another team wants to add. On Thursday she tunes a compaction threshold using a before and after test. On Friday she writes a short status note. Example 2: Combined skills. Diagnosing a slow agent requires profiling code and reading logs, which are software skills, combined with understanding that longer contexts increase delay, which is AI knowledge. Example 3: Prompt engineer and harness engineer. A prompt engineer crafts the wording of one instruction block. The harness engineer decides when that block enters context, and what happens if the model ignores it. Example 4: Harness engineer and platform team. The harness engineer defines what data an agent emits, such as spans with timings and costs. The platform team maintains the storage and display of that data for the whole company.