Agents7:36 · 3 Oct 2026
Same Model, Lower Score: Why Claude Code Lost on Terminal Bench
The same Claude Opus 4.6 scored very differently depending on the harness around it. What harness engineering is, what counts as the harness, and how you would design one for a coding agent.
In this video
- 0:00 Intro
- 0:25 The easiest definition of a harness
- 0:57 The harness you use daily
- 1:20 How prompt engineering and context engineering relate to the harness
- 2:28 Building blocks of the agent harness
- 3:37 The harness you can try right now for free
- 4:27 The process of building a harness
- 6:02 What is the Ralph loop?
- 6:21 Why harness engineering matters
- 6:41 My opinion
- 7:12 Outro