Agent harness engineering 101

Everyone is talking about agent harnesses, but what even are they and what is going on? Harnesses are essentially the wrappers around your model to form an agent. So your prompts, your tools, and your memory. Cloud code is a harness and so is Codex and Pi, which is what OpenClaw uses. And all of this started with prompt engineering, had context engineering, and now we have agent harness engineering. And all of this ultimately strives to achieve the same goal, optimising the output that we get using these models. But at this point, there is leverage in building your own harness. A harness is actually proprietary, also capturing and storing user memory, that's also data. But anyway, this is just, I think we're going to see a lot being built around this very soon.