Hi everyone, I’m the IX lead for Flutter & Dart team and I have a request! We want to create a blog post that looks at all of the cool things you can do with Flutter, Dart, and Antigravity. Some members of our team have found cool ways to use Antigravity and we’re going to write about that, but I’m also wondering if the community has found some cool things, too!
So, if you have a tip that you’d like to get into that blog post we are working on OR if you have a specific ask or topic you want the blog post to include, please let me know in this thread.
I’ve used it to take the code from an app I vibe-coded in TS in AI Studio, reverse-engineeer that for a PRD file, then update the PRD to use Dart/Flutter, and build the same thing in Flutter going forward from the PRD. Fairly satisfying experience, until I run out of credits. Hurry up and bring us BYOK so we livestreamers can do a demo without fear!
There a currently four things I would like to see:
Keep the Implementation_plan, not just switching it out for a new one on each prompt, spent 1/2 a day on getting the planning done, and then on the next prompt and new plan.
Keep the Task a bit longer than just one prompt, it behave the same as the Implementation_plan
Create an PRD from the first prompt.
Toke limit beyond the amount that you use in 1 hour.
Look at the lint rules : withOpacity - withValues the one that grates me.
Other wise very impressed with the capabilities, build a game controller and it did do a very good job.
PS. For it to understand what is secretes and not just push it to GitHub
@antfitch I think something in the lines of how AI is a helper and not something that will spill out perfect code to make your app in 15 minutes and you be rich in the next day. But I’m pretty sure this is not the image Google wants.
I guess your question was not about features we want, but, here’s what I want anyway =P:
Do you know how we can upload PDFs to Notebook LLM, so it can generate content solely based on those documents? I would like to see custom models trained solely with Flutter from documentation and known repositories (dart_mappable, http, bloc, etc.). Maybe some specialized models (Flutter with Bloc or Flutter with Flutter It). I was trying to convert an old project using some state management package to flutter_it (get_it, watch_it, command_it) and Antigravity did a pretty lousy job (it even try to add a old 4 years package named get_it_hooks because it thought that was get_it related, I guess). So, having models that uses the official documentation would be very nice. And you already have all the material you want on pub.dev. Make the flutter favorites to be “modelized” or create some new feature that allows package owners to provide AI instructions on how to do things. Something like that.
I find that the Dart MCP’s ability to look up documentation in pub is most of what you want. You just need to inject instructions to prefer that over “innate knowledge”.
Thanks everyone. We’re working on something now. If anything else comes up while your coding, add it here and we’ll see if we can get it into an upcoming post!
Don’t get your hopes up. As impressive, frustrating and/or misleading as A.I.s can be. Understanding is not computable. A.I.s are tooling built upon weighted graphs structures. The term neural network is a misnomer designed to market A.I.
Using contents of this forum for the purposes of training proprietary AI models is forbidden. Only if your AI model is free & open source, go ahead and scrape. Flutter and the related logo are trademarks of Google LLC. We are not endorsed by or affiliated with Google LLC.
Hurry up and bring us BYOK so we livestreamers can do a demo without fear!