What is your AI Vibe coding workflow? + Money saving tip.

paddy_proxies

Registered Member
Joined
May 24, 2025
Messages
67
Reaction score
41
Is anybody here using the new AI tools for building stuff?

I'm talking about tooling such as Bolt, Lovable and Cursor etc.

I find myself rapidly prototyping stuff in using the first two above, then when I'm happy with it.. I link it to GitHub, clone it locally and tweak it as I want. Sometimes by hand, and sometimes with Cursor.

I've also been playing around with Cursor, using the free tier and providing my own Claude / OpenAI API keys .. you end up with most of the cursor functionality but on a PAYG basis.

How are you leveraging these tools?
 
i use bolt or lovable for bots that too good i am not doing advertising just share my experience.
 
keep at it just ensure you're not asking a website/framework developing tool to handle content, design, whole sitemaps' webpage contents etc. yet

leaves your tools with a lot more room for actually getting what you need done

try v0.dev, keep in mind these tools seirously favour react typescript

i spent a month getting the two tools to create roblox games for me, works, but tools really did nearly have mental breakdowns
 
keep at it just ensure you're not asking a website/framework developing tool to handle content, design, whole sitemaps' webpage contents etc. yet

leaves your tools with a lot more room for actually getting what you need done

try v0.dev, keep in mind these tools seirously favour react typescript

i spent a month getting the two tools to create roblox games for me, works, but tools really did nearly have mental breakdowns

Good shout.

One thing I did notice is that the sites these things spit out aren't always the most SEO friendly.. or sometimes you might feel they don't want to be indexed LMAO.
 
Is anybody here using the new AI tools for building stuff?

I'm talking about tooling such as Bolt, Lovable and Cursor etc.

I find myself rapidly prototyping stuff in using the first two above, then when I'm happy with it.. I link it to GitHub, clone it locally and tweak it as I want. Sometimes by hand, and sometimes with Cursor.

I've also been playing around with Cursor, using the free tier and providing my own Claude / OpenAI API keys .. you end up with most of the cursor functionality but on a PAYG basis.

How are you leveraging these tools?
I've noticed the same with some AI-generated frontends: they look fine but end up outputting bloated markup or poorly structured layouts that confuse crawlers. Especially when div soup replaces semantic HTML, or when key tags like <title>, <h1>, and meta descriptions are either generic or missing.

If you're aiming for organic visibility, it's worth doing a quick manual pass post-generation, or even hooking the output to a static site generator where you can enforce cleaner structure and meta tagging. Tools like v0.dev are amazing for prototyping, but definitely not production-ready for anything SEO-heavy without cleanup.
 
Interesting how stable do you find Bolt for larger projects. Im still on VS Code + Copilot but curious if Bolt can handle multi file workflows properly.
 
Interesting how stable do you find Bolt for larger projects. Im still on VS Code + Copilot but curious if Bolt can handle multi file workflows properly.
craft something up using one of the tools bro, craft it up, open through gh desktop and use vscode through there

copilot on your top-performing websites :)
 
I've noticed the same with some AI-generated frontends: they look fine but end up outputting bloated markup or poorly structured layouts that confuse crawlers. Especially when div soup replaces semantic HTML, or when key tags like <title>, <h1>, and meta descriptions are either generic or missing.

If you're aiming for organic visibility, it's worth doing a quick manual pass post-generation, or even hooking the output to a static site generator where you can enforce cleaner structure and meta tagging. Tools like v0.dev are amazing for prototyping, but definitely not production-ready for anything SEO-heavy without cleanup.

100% agree .. for any serious stuff I build, I do plan on have a post production process, where I can audit the structure, sitemaps and all the common stuff that crawlers look for.
 
100% agree .. for any serious stuff I build, I do plan on have a post production process, where I can audit the structure, sitemaps and all the common stuff that crawlers look for.
That’s exactly the way to go. Once the prototype is solid, exporting to static and auditing structure, heading hierarchy, and internal linking makes a real difference. I’m currently testing a pipeline with v0.dev → Astro → SEO pass via Screaming Frog to catch crawl issues early. Still a bit clunky, but much better than shipping AI output raw. If you’re running JS-heavy components, you might also want to pre-render critical pages to avoid rendering gaps.
 
That’s exactly the way to go. Once the prototype is solid, exporting to static and auditing structure, heading hierarchy, and internal linking makes a real difference. I’m currently testing a pipeline with v0.dev → Astro → SEO pass via Screaming Frog to catch crawl issues early. Still a bit clunky, but much better than shipping AI output raw. If you’re running JS-heavy components, you might also want to pre-render critical pages to avoid rendering gaps.

Yea +1 on the JS -> Static.

I need to solve for that, I think what I have is ok, but needs improvement.

Got any details on your pipeline.. is there a pattern online that I can read about?
 
Yea +1 on the JS -> Static.

I need to solve for that, I think what I have is ok, but needs improvement.

Got any details on your pipeline.. is there a pattern online that I can read about?
Yeah, most of it I stitched together manually. There isn’t a clean tutorial for this exact combo, but if you’re already using Astro, check their docs on partial hydration and @astrojs/sitemap. From v0.dev I usually export raw HTML, clean it up, then feed it into Astro pages with metadata blocks. For crawling, Screaming Frog just runs in scheduled mode on the rendered output. It’s basic but gets the job done for small to medium projects. Happy to share a stripped config if you’re testing something similar.
 
I think ai isn't good enough to handle the kind of volumes of code I would be flashing through it and asking it to expand features. I think every project has that aspect to it. Ai only can still do the basics. It does them pretty good though. But I wouldn't trust anything beyond my own supervision, and copy and paste only. I wouldn't just let it loose and say go on improve this or find that issue. lol i'd be out of API credits by the end of the day! lol and above all I don't trust it. There are too many 1's and 0's being blasted to not get a few wrong, and every program apart from a chatbot output, needs to have correct code.
 
It depends a lot on the information you provide it as well. Context engineering is what that's called now.
Imo it still takes a good software engineer to steer the agent into a direction that doesn't result in an unmaintainable mess.
 
It depends a lot on the information you provide it as well. Context engineering is what that's called now.
Imo it still takes a good software engineer to steer the agent into a direction that doesn't result in an unmaintainable mess.

100% Agree. While I think a seasoned engineer can build something decent that can be put into production.. obviously with a lot of experienced guidance.

I do still think it's a fantastic tool for non coders to get something up and running as a PoC etc.
 
I use copilot and i think is good enough but still what will be better?

I actually haven't used Copilot.

I've been using Bolt a bit and it's decent.. It has been decent enough at building a web service in Go also, along with Postgres migrations and docker infra.

I got a bunch of free credits during their hackathon last month and now they have a new customer in me lmao.
 
Back
Top