I'm thinking of purchasing the Devin subscription, so I can give it screenshots of a web-app, and have it just implement the flows. But ... I don't know if it will be capable of doing so.
I have not used Devin. But from what I gather from the reviews, Cursor + Composer/Agent will take you a little further than Devin will, for a lot less money. Neither will take you where you expect/want to be. You will still have to do some/a lot of the work yourself.
My team tried it, my conclusions were:
- it took just as much prompting as a using an LLM to get simple tasks done
- took many hours to come to wrong conclusions for simple problems
- the examples in their site are definitely cherry picked
They are all wrappers around GPT-4/Claude. If you don't know what you and the AI is doing, you will not get anywhere. If you know what you are doing, you'll probably just use a regular editor with an AI auto complete.
Using Chrome on Android with Adguard.
https://github.com/stackblitz/bolt.new/issues
https://youtu.be/927W6zzvV-c?si=uerRK1czUIGZbJBn
https://devin.ai/