ChatGPT Apps: Embedded UI no longer renders on web after agentic multi-step exploration

We’re building a ChatGPT App that lets users explore and view our own content. Until a few days ago, a workflow like this worked on web: 1. GPT performs agentic multi-step exploration 2. It chains multiple data-only calls to resolve candidates/details 3. It then invokes a UI-rendering step to surface the result in chat Now, the exploration/data steps still succeed, and the UI-rendering call also appears to be invoked, but the embedded UI never actually appears in the web client. The key pattern is: * data-only exploration succeeds * standalone UI rendering works * UI rendering after sequential exploration calls no longer surfaces on web So this does not look like a retrieval failure. It looks more like a regression in the handoff from multi-step exploration to embedded UI surfacing. Also worth noting: * Web: embedded UI does not appear * Mac desktop: still behaves as expected Has anyone else seen a recent change in ChatGPT Apps where the UI-rendering step is invoked but the embedded UI does not render after agentic multi-step exploration, specifically on web?