11–20 of 22 posts
It seems to me that AI-based coding assistants are better at writing code for more established frameworks, which has led me to defaulting to vanilla or tried-and-true/"boring technology" because AI has trouble with newer or more frequently changing stuff. Curious if this has been the case for anyone else.
Because AI is not there yet. Boilerplates are the better option until AI generation catches up. Here is what I do: 1. Explore an idea, research what competitors do, etc. 2. Use my own boilerplate for a greenfield project https://achromatic.dev (sorry for self-promo) 3. Write down what I actually want and paste the text and screenshot to v0 to see if it can generate something useful. 4. Improve/write the initial code…
Should mention I'm non-technical and building a scrappy mvp this way. It could end up super scrappy...
For example I created an open source metadata crm called IceburgCRM iceburg.ca (available in Django / Laravel) recently. Without AI involved you could configure the crm by inserting data in the database. For example you add an entry for accounts, contacts, etc in the modules table and your crm will have menu links and a module section setup for accounts, contacts.
With AI you can now generate metadata needed to create custom crms. For example typing in "create a wine crm" in the AI prompt I build I can generate:
"create a crm for bee keeping": https://beekeeping.iceburg.ca
"create a coffee lovers crm": https://coffee.iceburg.ca
Using AI in traditional applications opens up a new dimension for us crud developers. I think we will see a shift to more metadata applications in the future as the benefits become clear. It's a new field using a different type of creativity and some old school design patterns.
Yes I can help with this! I go straight to code, but I use my design copilot chrome extension [0][1] to constantly give me component-level design feedback as I build. It basically forms my entire design system now. Currently it’s a manual process - you use the built in screenshot tool and it kicks off the design feedback flow. But I’ve been playing with adding data-attributes for automatic feedback whenever the compo…
I won’t go off-topic with actual details (many of which I’m sure you’ve already considered), but I’d be more than happy to share some ideas if you’d like And not on the Chrome store, because I don't want to rate it just yet. (5* for the concept though )
Yes I can help with this! I go straight to code, but I use my design copilot chrome extension [0][1] to constantly give me component-level design feedback as I build. It basically forms my entire design system now. Currently it’s a manual process - you use the built in screenshot tool and it kicks off the design feedback flow. But I’ve been playing with adding data-attributes for automatic feedback whenever the compo…
Hey mate, I tried your extension and I love the idea. I don’t want to be rude, just honest, but in my opinion, it’s in such an early development stage that it can be quite frustrating and overall not very good. I won’t go off-topic with actual details (many of which I’m sure you’ve already considered), but I’d be more than happy to share some ideas if you’d like And not on the Chrome store, because I don't want to ra…
h e l l o p a p i l l o n s o f t w a r e d e v
(also future GenAI friendly frameworks may be not really programmer friendly)
Because AI is not there yet. Boilerplates are the better option until AI generation catches up. Here is what I do: 1. Explore an idea, research what competitors do, etc. 2. Use my own boilerplate for a greenfield project https://achromatic.dev (sorry for self-promo) 3. Write down what I actually want and paste the text and screenshot to v0 to see if it can generate something useful. 4. Improve/write the initial code…
+1 for v0. Good for creating starting base design. I then ask Claude to organise codebase, then take it to Cursor. Should mention I'm non-technical and building a scrappy mvp this way. It could end up super scrappy...
Been working great! I proofread almost everything of course but moving really fast and not having to muck around with which tailwind class is really nice.
My most productive workflow, is take a stab at what I want first and use gpt/claude to generate boiler plate code. I use the base components from tailwind components and ask Claude/GPT to tailor them to my needs. Been working great! I proofread almost everything of course but moving really fast and not having to muck around with which tailwind class is really nice.
Earlier quoted context omitted.
+1 for v0. Good for creating starting base design. I then ask Claude to organise codebase, then take it to Cursor. Should mention I'm non-technical and building a scrappy mvp this way. It could end up super scrappy...
It's good to not have an empty canvas. Like a writer and an empty page. But the output is not that good yet imo.