Last September, Anthropic gave us a 5-day research preview called “Imagine with Claude” where Claude could build full applications from scratch in real time—then they took it away. In this video, I break down what it means that they just brought it back, but this time embedded directly into every Claude cloud conversation. This isn’t image generation and it’s not a “canvas” gimmick. It’s what I call on-demand generative UI: Claude deciding, in the moment, that the best answer to your question is an interactive app it generates right inside the chat—and you can throw it away when you’re done. I walk through examples like an interactive compound interest explainer (sliders, changing values, instant feedback) and even a paper airplane folding animation. Under the hood, it’s simply writing software—you can download it as an HTML file and save it as an artifact to use standalone. The bigger takeaway is the ecosystem play: these features aren’t isolated releases. Anthropic is clearly trying to become the “everything app” for knowledge workers, and at this point model intelligence gaps matter less than the product harness around the model. That’s also where vendor lock-in starts to look inevitable, just like cloud providers.