So like especially the auto routing piece, right? So when I'm selecting it, I'm selecting it, right? But as an engineer, if I don't select it, like outside of in Cloud Core or outside of Cursor essentially, like where is the actual routing happening here?

So it depends on the harness.

I'm a firm believer that I think harnesses should dictate to some degree the model routing, model behavior.

You need to understand the task workload.

And quite honestly, as we're designing our harnesses, we're trying to make them more geared towards discrete tasks.

So some of our cloud harnesses have very discrete tasks that are kind of known.

So you'll have kind of an orchestrator agent that will, you know, run a discrete task, like a fixed CI task, for example.

What are the actual use cases in practice? What actually happens in cloud harness? How much of an elaborate setup that is? What kind of tasks can be delegated to those relatively safely? So tell me what works great, what kind of is okay, and what kind of doesn't work.