The console is not downstream of the platform

Most teams building "agentic experiences" are running into the same wall, and no one is naming it clearly.

Engineering is shipping primitive APIs. That's their mandate, and they're doing it well.

Design is being asked to ship agentic console experiences. Multi-step workflows, conductor agents, tool chains, human review surfaces.

The gap between these two is where the work actually lives, and no one owns it.

A primitive API is not an orchestration layer. An orchestration layer is not an agent. An agent is not a judgment surface. Each one is a distinct discipline, and shipping the next console generation requires all four to show up at the same time.

What I keep seeing: teams treat the orchestration layer as a downstream problem that someone will pick up later. But the orchestration contract is what determines whether the agentic experience is buildable at all. It determines what the human can review, what the agent can recover from, and what the console can actually show.

The design surface for human oversight has to be specified before the orchestration layer gets built. Not after.

Otherwise you end up with a plan review screen that reflects what was convenient for the model to output, rather than what a human can actually evaluate.

The console is not downstream of the platform. In agentic systems, it is the contract the platform has to satisfy.