Mo, maybe you can talk a little bit more about that and what that looks like in practice then for a developer.

Yeah, absolutely.

And building on what Bumik covered here, first, there is the nature of the agents, how long they run, how compute intensive the work that they're doing is.

Are they mainly pulling data and waiting on data to come in? Or are they, in fact, actively analyzing this data or building code and so forth? And of course, the intelligent always-on orchestration control plane that directs the actions that happen throughout the system, including the creation of sandboxes.

And so in the coding agent context, for example, or when you ask an agent to do some math for you, it will generate some code, and the system will, through APIs, ask for the instantiation of a sandbox environment that's an isolated environment, so code cannot break through.

Of course, security, reliability, key priority there.

And within that sandbox, GKE Agent Sandbox will give you the flexibility to run on various different virtual machine types, which is fantastic given the broad range of choices in Google Cloud and Google Compute.

What's it going to look like for somebody who wants to have a foundation now that they can scale then going forward as these workloads just increase exponentially?