I just launched AllPub.co and built pretty much the entire thing using Vibecoding with Claude Sonnet 3.5. I'm a software architect with just some basic Python knowledge, so this was my first full web app build. Wanted to share my experience!

What I Built

Hey everyone

No CS background. Started vibe coding 3 months ago.

16+ hours daily. 15 billion tokens later I'm still learning.

I was using a Claude Code leaderboard service made by another dev.
Submitting my daily stats became my end-of-day ritual. It was my fuel for vibe coding.

Hey everyone, been a Product Hunt visitor for years but never signed up, so I thought I'd start by sharing some recent learnings that hopefully other builders will find useful.

For context, we're building FanBase Copilot, an AI assistant for content creators that learns their voice and context over time. The memory layer is critical. It's what makes the AI actually useful after the first conversation.

Every vibe-coding session starts the same:

Here s the context again
Here s what I already tried
Please don t repeat this

That friction killed my flow, so I built Blocpad (CLI).

It keeps context with the project, not trapped in chat history tasks, decisions, notes, all local.
AI reads the state. No re-prompting.

Building my app with AI tools, zero coding background. The magic part - I can ship features in hours. The scary part - I have no idea if the code is actually good.

Right now my "QA process" is:

- Does it work when I tap around?

- Did anything break that worked before?

AI and no-code tools are evolving insanely fast right now. Every few weeks there s a new tool that changes how quickly you can go from idea to product.

I ve been experimenting a lot with different vibe coding platforms lately, trying to find the right balance between speed, control, and flexibility. What s surprised me most is how far you can go today without a traditional engineering setup.

For context, I recently built @Sendrise , an all-in-one cold email outreach platform, using a no-code + AI stack. I used @Lovable for building the product flows and UI, combined with AI tools for writing, automation, and logic. What started as an MVP quickly turned into a fully working product with lead management, campaigns, CRM, and analytics.

See this?

Step 150 of debugging why a payment does not get saved to a database. Two days on this one bug. And there are plenty more. If you can build somehing that will do the back-and-forth, the "now try this and tell if it... no? Okay, le's do this thn, and this, and that..." Do what Claude Opus 4.5 is tellling me to do, the tens of hours, to get to the solution. Automate that and you have a winer - becuase there are 100K full-stack devs who will do all this more effienctly themselves, yes. but there are 10M non-developers who love what they built, but are getting killed in the debugging, the last 5%.

I am a product designer, not a coder, and so "lowcode" code editors leave me lost. It simply takes too long to ask Claude to step-by-step me through anything I want to do.
I have projects in the works in Lovable.dev and Medo.dev, Anything and Dazl - but yet to get things to work after migrating them to Vercel breakl out of the co-branding that these app-builders like to sneak in (notice the rainbow you see here and there on Lovable-built sites, even whn they use their own domain...?)
Gemini's suggestions:
AI code generation is prone to the "90% Problem," where the final 10% of polish requires significant manual intervention.  

• Bolt.new: Users report that while blazing fast (app skeleton in 20s), it often "shits the bed" when projects get complex, leading to massive token consumption to fix errors .

• Lovable: Praised for UI aesthetics, but its 7-message memory limit leads to "recency bias," causing the AI to get "drunk" on different approaches during long debugging sessions .

• Tempo Labs: While it provides excellent planning features (PRDs and diagrams), some users have reported that its export to GitHub is unreliable, providing only "fragments of code" rather than a full app .

• Blink.new: Emerging as a smoother alternative for full-stack builds, handling authentication and backend integrations with fewer breaks than Bolt .

Strategic Implementation Path

1. Selection (Phase 1): Use Dyad.sh or Anything if you want the most stable backend/auth foundation .

2. Generation (Phase 2): Build your MVP on the Pro tier for 1 month (~$19-$25). Use Discussion Mode to plan changes without burning credits .

3. Sync & Export (Phase 3): Sync your project to a private GitHub repository immediately .

4. Deployment (Phase 4): Connect the repo to Cloudflare Pages for free, unlimited bandwidth or Coolify on a $5 VPS for total control .

5. Refinement (Phase 5): Use Cursor (locally) for further edits. It provides a full IDE environment with better project context for $20/month, allowing you to maintain your "clean codebase" without builder overhead .
   I then asked, about a cheap solution starting with Baidu's Medo.dev - because Medo gives us many credits to start, 100 more/day free, 20 more for 20 clicks on other's projects... So you can build a lot, download, but then have to sort out login and other backend problems. Possible, but takes some doing.
   Gemini's response:

   1. In Supabase Dashboard: Go to Authentication > URL Configuration. Change the "Site URL" from http://localhost:3000 to your actual production domain (e.g., https://your-app.com).

   2. In Vercel Dashboard: Go to Settings > Environment Variables. Add a new variable called NEXT_PUBLIC_APP_URL and set its value to your domain. This tells the backend where to send users when they click the email link.

   3. In Cursor: If the error persists, open your project in Cursor and ask the AI: "Search for all instances of 'localhost:3000' in the authentication logic and replace them with the environment variable for the site URL.".
      But what about you? What is your process if you are no-code building and shipping?

I'm Kevin, founder of EnginifyAI. Over the last 12 months I've been working on a prompt engineering tool. I started off using Bolt but quickly I found myself unable to completely use it effectively. I then tried Cursor and was using it early on but maybe I had the wrong settings as I found myself using all my credits in 2 days. So, I switched to Windsurf, and to my surprise credits were a bit more manageable and it took about 2 weeks before my credits were consumed. Keep in mind that I am new to coding so I spent a lot of time asking questions in Windsurf instead of using my paid subscription to Anthropic.
I'm coding using React/Typescript and Supabase. I have learned that coding is a lot of work, a lot of fine-tuning, and a lot of testing and fixing. The one thing I haven't quite figured out completely is how to fix migrations. Early on I was running SQL scripts directly in Supabase without migration scripts because I found it to be faster. But now Windsurf uses very old migration script references and I find myself fixing the same issues because of an old migration script reference. Any idea how to fix this?
I assume that vibecoding is an experience everyone can enjoy but comes with frustrations about asking the right questions, or being very specific about my requests. I am using Claude Sonnet 4, 4.5 and 4.5 thinking. Are these the best coding ai models for Typescript?

Are you someone who's built > 10 apps using Lovable, Replit, Mocha, or another vibe coding tool? If so I'm very interested in what is your answer to these 2 questions:

Assume you're the "benevolent dictator" of one of these apps for the day, and can choose to add one thing, and remove one thing:

• What's your most desired feature? Thing holding you back the most?

• What really annoys you and would you change or remove?