WHY HARNESS ENGINEERING EMERGED AS ITS OWN FIELD

CHECK YOUR UNDERSTANDING • Why did longer agent tasks make harnesses more important? ◦ Answer: Because with no human reading every answer, small errors can spread silently, so the surrounding system must catch problems. • What kinds of causes did production teams report for most coding-agent failures? ◦ Answer: Context management and tool design more than the model's raw reasoning ability. • Why should you treat reports of agent-written code with care? ◦ Answer: Because they are usually self-reported and hard to compare or verify independently. • What is automated harness optimization? ◦ Answer: Tuning harness settings through automated experiments rather than changing the model. • Name one common mistake when comparing two harness configurations. ◦ Answer: Using too few tasks, so that a difference is just luck.