Saw a video of Kia van where the presenter shows of a 3 piece bumper, a food vendor can buy a specific piece if he bashed one corner. Now car bumpers contain sensors and cameras and flush lights and a light hit can cause very expensive damage, and its brushed off with "compared to a 90s car, you can walk away from ...".
A good harness should not only make agents more capable at completing tasks, but also make their outputs much easier to review.
For example:
A good harness constrains the action surface, context, and task boundaries.
An agent’s failure isn’t always due to “writing incorrect code” — it can also result from “doing things it wasn’t supposed to do.”
Tests and lints can verify part of the correctness, but they often fail to validate task scope.
A well-designed harness should shift the review process from “reading the entire diff” to “verifying whether the changes stay within the defined task boundaries.”
