What model did you use?

If you plan on using AI for refactoring I can recommend the following:

- Make very small changes
- Be very explicit with your prompts
- Best case, have a strong test suite to run between any AI change (in most IDEs the AI can run those tests without your input)

To be fair, without careful considerations LLMs seem to be least useful for refactoring. IMO it makes sense as refactoring is a very complicated task which is usually best done in very small iterations (with lots of testing/ type checking etc. between them).

Yeah that's a good use. I also really like to use it for scripts!

It's definitely more complicated, but AI helped me a ton creating a custom render object for my 'flutter_canvas' package implementation. Of course, it was more trial and error but bigger chunks of code (complicated transforms) were fully AI built!

I haven't tried self-hosting but I'd assume it performs pretty well. But I agree, it's mind blowing to think about all of that intelligence fitting on consumer hardware 😃

You can always self host (for example Deepseek which is really good!).

I like to think about AI in a way where it allows you to do more in the same amount of time. Think of it this way, the AI takes care of the nitty gritty details while your brain can focus on the real hard challenges!

In the end, it all boils down the task at hand. Some tasks AI is great for, for some it's not. You just have to figure out when to use it to what extent!

From my experience, AI usually goes the extra mile. For example when building an API endpoint you might just throw some error to get the feature shipped. The AI usually tries to follow best practices, throw telling errors etc.

I often found the AI handle cases that I didn't think about myself!

Definitely! There are a couple of layers to this IMO.

First - Less mental strain. It just requires less brainpower to tell the AI what to do instead of doing it yourself. This only works for tasks the AI confidently solve, but human iteration may be required.

From my experience, the more specific you are the better. Agent mode really shines when you instruct it to also write + execute tests as it can correct itself before yielding to you. That's also exactly what I'm working on for Vide.

Haven't publicised it too much as it's still _very_ early. Spending a bunch of time getting the web version out of the door so everybody can try it out with just a few clicks. It's getting there ...

Thanks for coming to the stream & all the great questions! :)

Unfortunately not! :(

I need to keep the dependencies to a bare minimum as any dependency in vide might conflict the users dependencies

Waitlist link: coffee-television-59c.notion.site/1af88f114594...

Vide hooks into Flutter, giving you a dev focused environment — deeply integrated with AI.

More info & web demo soon!
Join the waitlist below 👇🏼