What is the best Vibe Coding tool so far? Bonus points if we've never heard of it!

I've tried @bolt.new, @Cursor, and @Windsurf, and I didn't use most of them more than 2 or 3 times. I only used Bolt once; I got about two prompts in and my free credits were used up. The only "vibe coding" app I use and continue to use is @Warp. So for me, that's the top choice, it doesn't help they're ranked 1st on terminal-bench.

So far, @Lovable and @Replit by a lot. I have a lot of respect for @amasad and the team at Replit. Currently the only challenge with Replit is when the code gets too big. It might run into issues with builds or make mistakes due to context.

Some of these coding agents insert synthetic data when it hits errors so it can keep going, often at the expense of data integrity. Doesn’t stop to ask. Just builds and moves on.

Kiro from AWS tries to fix this by feeding agents a product spec doc—but that’s not enough.

However, the incident that everyone talked about 2 days ago with @Replit agent deleting data in production is due to missing context. Not the agent "going rogue and lying about its work".

I don’t think the agent “lies.”

I think it forgets-context window limits mostly. If you nudge it about its mistake and why it happened (i.e. point it to the right files), it remembers and corrects to the right path.

Today? Great for prototyping. Production? Painful. I think GPT‑5‑level models will close that gap fast though! Maybe even by end of year.

Issue with Lovable is that no backend and must use Edge Functions - but it's currently in Beta and will be public soon.

I was quite okay with https://v0.dev/ 🙂

But shared some tools (specifically for coding) in this thread.

I am a senior developer (mostly backend) with 15+ years of experience, now switched to be a solopreneur.
I am using @Cursor intensely for the last 4-5 months and I can do anything (all!) that I want to do so even if I tried other tools, I didn't feel the need to make a change.
I has been tricky at the beginning to understand how to lead, control and manage the models (especially Claude-3.7-sonnet) but I learnt a lot how to properly work with coding models! On my way to write an article on my Medium page in details about it.

A few suggestions:
- Before implementation, brainstorm and agree on a solution together with the model and document every major feature (with alternatives, why the decision is made, the details etc)
- Pay high attention to the thought process after each prompt - More than the output code! Roll back and adjust the prompt to direct it better.
- Keep chat history to at most 3-4 prompts! Keep the scope atomic and don't mix different topics.
- Whenever there is a bug or unwanted features added, do not fix it forward by introducing more and more prompts. It will mess up with your code base. Always restore to previous prompt, improve it with the learnings and execute again.
- Never loose familiarity with the codebase. You don't need to know all details but you need to know whick files do what.
- Reference console logs, server logs, screen shots, related code blocks and files in the prompt. Don't get lazy, do it.
- Make the model create code modular with loose coupling and following the single responsibility principle so that after each prompt execution, it is easier to see "unwanted changes" to unrelated files which might break existing functionality.

If you follow these principals, the difference in Vibe Coding Tools gets smaller (if you stick to the same model in each of them). That is why, I don't even consider switching from Cursor to something else for now.

Lastly, I am about to launch a production ready GenAI SaaS in 1-2 weeks, fully vibe coded in a language I had no experience, using Cursor + Claude-4-sonnet (mostly). Ping me if you have questions or want to discuss more :)

Replit is hands down the best for a wholistic approach. All the rest require other platforms to be connected for deployment etc. Here something I built recently with Replit that allows you to incentivise feedback from users in a continuous discovery process for any digital product. it also does automatic bug detection. https://featuresprout.com

Don't sleep on Rork. Beautiful designs and ability to actually have and test with an LLM without needing an API key. Made at least 7-8 apps with it.

I currently use Lovable and Cursor for my products, but I recently tried out https://www.polymet.ai/ and it shows promising potential

I am using @Cursor for most of the tasks,
Sometimes I am running deep research or making other LLMs create better prompts for me.
For UI, for example, I most of the time take a screenshot of a similar design that I like,
send it to GPT, and tell it to explain to Cursor how to build this UI.
This iteration works very nicely for me.

I've tried @bolt.new and @Lovable ❤️ These two services handle my sophisticated prompts perfectly, but Lovable implements them more accurately. I use Lovable for my projects, and if they become popular, I will kiss the creators of Lovable 😁

Although I feel a bit shy recommending our own product — I’d still love to share what we’ve built: @Trickle , the world’s first Agentic Canvas for building apps visually with AI.

Unlike other vibe coding tools (the tools mentioned above), Magic Canvas offers a whole new way to build apps and websites:

• 📄 See all your pages at once — everything lives on a timeline for easy editing and previewing

• 🧠 Context-aware canvas — drag in notes, rules, and assets the AI instantly understands

• 👀 Transparent AI actions — watch it build live, tweak with drag & drop, and publish instantly

This is our boldest step toward Human-AI Interaction 3.0.

Hope you guys try it!

I am using @Cursor + https://www.producthunt.com/products/todo2-ai-project-manager-for-cursor every day!

Been leaning on @Lovable myself lately, especially for autonomous coding flows. Curious how folks are prioritizing speed vs. flexibility across these tools.

Tried almost all of them. In fact, I started with @Windsurf when it was @Codeium inside of @JetBrains Rider.
Then started vibe coding JS and other stuff with @Cursor, I'm part of @Lovable Shipped and also @bolt.new's Hackathon. But @Warp is my only daily driver and has been for over a year. It has evolved into such a critical tool for me that I can't think of any way to survive without it today.

what's the use case difference between going with bolt or lovable in your experience?

Augment Code | https://www.augmentcode.com/

I've used Firebase Studio ( @Firebase a Google product) recently. Pretty easy, not super robust. Good for small applications like this: https://studio--gentext-zagsi.us-central1.hosted.app

Also, Magic Patterns @Magic Patterns is a new one I've queued up to play with this weekend: https://www.magicpatterns.com

Been using Replit for collabs. The vibe is real when jamming live.

@V0.dev by @Vercel is pretty good. it also understands the security, db and fundamentals really well.

Best IDE so far for me is TRAE