Connect the dots.
Trends are claims to examine. Each perspective links to supporting events and acknowledges limits and counterevidence. Coverage is not adoption.
Our editorial method ↗The interface becomes an entry point to answers and work, while the host must remain in control
From Gemini, ChatGPT and Blazor to Binance AI, Workday Sana, CopilotKit Soar and OpenDuet, generative UI is entering everyday, financial and core enterprise workflows. OpenDuet turns attention, work in progress, approval, missing input, retry and resolution into a persistent task interface, showing that proactive agents depend as much on host-managed state, authority, approval and recovery as on generated content.
11 supporting events · Our analysis02 ↗Protocols, catalogs, native rendering, inference formats and code generation are becoming separate layers
Event transport, wire protocols, component catalogs, host embedding, native state, identity boundaries, model syntax and client construction code are evolving independently. A2UI Express compiles a model-friendly representation into standard messages. The first A2UI CLI now turns catalog schemas into typed Python builders across 0.9, 0.9.1 and 1.0, carrying catalog contracts into application code without changing the renderer protocol.
20 supporting events · Our analysis03 ↗Specialized models, local interaction and evolving interfaces
Specialized training, bounded decisions and local runtimes are separating the cost layers of interface intelligence. GUI‑JEPA shows that when a host can enumerate trusted action candidates, a 0.50B discriminative policy can replace token-by-token action decoding by a 7B model with one latent-space scoring batch, while Agent S4 evolves the interface itself. Model capability, output modality, candidate catalogs, planners and rendering runtimes are becoming independently optimizable layers.
8 supporting events · Our analysis04 ↗Evaluation and engineering move beyond the first screenshot into task outcomes, causal history and active verification
Generative-UI and GUI-agent acceptance is shifting toward hidden-target recovery, multi-step outcomes, verified progress, causal message history, and active evidence collection across interfaces, files and structured state. Two unrelated papers named VeriCUA respectively learn when to invoke Judge/Probe verification and use a verification agent to explore read-only GUI, accessibility, file and database evidence. Malva confirms 335 tool-mediated defects across 40 open-source applications, showing how observation, action and feedback contracts can silently separate an agent from the live interface.
20 supporting events · Our analysis