HN Explorer

Ask HN: Has anyone used Devin for web-dev?

21 points by vsroy ↗ HN
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.

Has anyone used Devin for this type of thing?

16 comments

[ 2.5 ms ] story [ 32.9 ms ] thread
azlan692
learn programming before using devin it doesn't make websites but just simple instructions, you need to know what you're making.
cheema33
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.
neom
I think https://www.tempolabs.ai/ (YC S23) might be closer to what you're looking for.
solardev
Is it normal that I can't scroll at all on either their homepage or their product hunt page? Other sites work fine.

Using Chrome on Android with Adguard.

uncomplexity_
same here cant scroll
SixClub
Hey fixing this right now!
solardev
Seems to work now! Cool product btw. Good luck on the launch :)
SixClub
Thanks :) Our fancy draggable hero header didn't work on mobile. Fixed now haha
fragmede
Claude w/ Aider will take you plenty far.
csomar
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.

https://github.com/stackblitz/bolt.new/issues

runjake
For others, like me, who don't know what Devin is: Devin is apparently like GitHub Copilot and Cursor IDE.

https://devin.ai/

drudolph914
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
ltravis495
Thanks for the insights. I didn't think it would do better than other LLM but their Github issue stats were quite impressive. It's nice to see some real experience with it so we know its actual capability.