The last update was in May. Since then the work has been one theme: tightening the loop from a built app to launch media that shows the real product. Here is what shipped and is live, and what is proven but still gated.
Demo videos now show your actual app
Product demo videos in Growth Agent no longer lean on stand-in imagery. The capture pass visits your deployed app and shoots it the way the format demands:
- Vertical demo beats are captured at a real phone viewport, so mobile-first apps render legibly instead of as shrunk desktop pages.
- Landscape demos give the app the frame: the caption band was cut to roughly a fifth of its former height after side-by-side review.
- Stat and data beats declare a focus region at capture time, and the render zooms into that measured region of the real screenshot — a crop-to-region move in the style of the best screen-capture tools, never a guessed rectangle.
- Captures land in a durable content-addressed store, so re-renders and route screenshots reuse verified shots instead of re-shooting on every run.
Presenters that actually speak
The May update was honest that provider-backed lip sync was not proven yet. That caveat is now retired. After a controlled bake-off across the video-provider roster, presenter beats render on the provider that was the only one to deliver a true talking presenter — synchronized mouth movement, the product name pronounced correctly, vertical framing respected. Silent world footage stays on the cheapest best-framed provider; the roster split is recorded with the measurements that decided it.
In the launch-film proof lane, the full version is further along: a film generated for a real user session opens with the presenter speaking the hook in the provider’s own voice (lips match by construction), hands off to a narration track aligned word-by-word by speech recognition, and composites the session’s actual dashboard — byte-verified before any render spends money — as the product proof. Every stock and generated clip is vision-screened so third-party brands and invented interfaces never reach a customer film. A 16-world concept library keeps different products from converging on the same generic office scene.
The builder got a long reliability wave
- Publish celebration and Guardian readiness were restored end to end.
- Registration attribution is preserved until consent, and publication requires explicit consent.
- GitHub imports open directly in the editor, and imported refinements keep their scope.
- Mobile builder and share flows were polished across a multi-PR pass.
- Bring-your-own deploy credentials now surface invalid tokens and scope problems clearly instead of failing opaquely.
Design range is on by default
The nine-style aesthetic axis is live in production: generated apps draw from distinct design directions instead of one house look, before any visual refinement you do on top.
Simpler pricing, and a new way in
Plan choices were simplified — see pricing for the current packaging. And PromptUI builds can now start from your editor: the PromptUI VS Code extension is published, so an idea can go from editor to build without switching context first.
What we are still not claiming
House rule, same as May: claims follow proof. Motion inside demo beats — live clips with a scripted cursor instead of stills — is merged and production-tested, but it stays behind a flag until one more clean production run, so demo videos ship stills-plus-zoom today. Social publishing proof still covers the X loop end to end, while other networks remain partial. And the full self-voiced launch film described above is production-proven in the proof lane, not yet the default render every user gets.
The loop, stated plainly
Build the app from a prompt. Keep proof, preview, and deploy attached. Let Guardian watch what ships. Then turn the same product context into launch media that shows the real thing — not a mockup of it. That last part is where this quarter went.