The components exist. The product is in the seams between them.
Each individual layer of a production agent runtime is well served by open source. There are mature, capable projects for durable execution, for policy evaluation, for identity, for isolation, for model routing, for observability, for evaluation, for workflow, and for audit storage. A team assembling these parts is starting from good material, not a blank page.
No single component is the difficulty. The difficulty is composition: making the layers aware of each other so that a decision in one becomes a constraint in the next. A production agent runtime is defined by these dependencies, not by any part in isolation.
Each dependency is a seam between two components. The seams are where the work lives, and where an assembly of otherwise-excellent parts either holds together or does not.
Wiring the components together produces an internal platform, and that platform does not finish. Once agents run against it, it becomes a product with its own maintenance surface, carried indefinitely by whoever built it.
Neither answer is universal. The right path depends on where the production runtime sits relative to your product, your customers, and your team.
The question is not whether your team can build this. It is whether the agent production layer is where your team creates the most value, or whether the agents on top of it are.
Bring an agent, its deployment target, and the production questions your customers or security team are asking. We will walk through the runtime, control boundary, deployment model, and evidence architecture.