Hello everyone,
We just shipped a round of improvements to the new AI system:
- Fixed the infinite loop issue that was burning through tokens
- Added a “stop” button to interrupt the AI mid-generation
- Tuned the system prompt for better visual quality on frontend generation
- Fixed bugs that were causing logic generation to fail
As much as this new release is a step up, today’s AI is still far from where we want it to be. The next big leap is a true full-stack experience: one agent you talk to that handles both frontend, backend and multi-pages changes in a single flow. Beyond the productivity win, this full-stack agent will also resolve many of the generation issues we see today. Frontend tweaks frequently require backend adjustments (and vice versa) and because our AI can’t yet work across both layers, it sometimes introduces breaking changes on either side.
Another key improvement will be full element coverage. This alpha only supports a handful of our elements today and we’re aiming to have the AI supporting all of them by the end of next week (including advanced elements like charts, calendar, etc.).
To sum up, here’s what’s next:
- Full element coverage in the AI (by the end of next week)
- A true fullstack AI agent (a few weeks out, expected for mid-May)
- Once 1 & 2 ship: laser focus on generation quality and token cost
We are optimistic the AI experience will be in a very good place within 4 to 6 weeks.
