What tools are you using and what is frustrating you?
Hello everyone, Sumit here from the Himalayas. I hope everyone is having a relaxed weekend. My workflow with vibe coding has settled pretty well as I get more and more time out of desk while Claude Code builds the software.
I wanted to offer any assistance to fellow founders. I have been vibe coding full-time for a little over 4 weeks. Wrote about it here. Please share your tools, or workflow and in particular what is not working for you. What is frustrating you in building software with vibe coding?
269 views
Replies
Hi Sumit,
Technical pain – context. When I hand-code I instantly recall why a line exists; with vibe-coding I’m handed blocks I’ve never seen and nobody “remembers” the intent. Long-context models help, but architectural reasoning still evaporates.
Design friction – my loop is
ChatGPT → Cursor → Devin → Cursor → Claude Code.
Each is amazing, yet I still have to hunt the newest docs, paste them in, and babysit the model. Now that raw coding speed is solved, the next frontier is working creatively away from the desk—finishing tasks on a phone (or any new device) in seconds.
I’m tackling that with LLMonster: a mobile share-sheet that triggers generative-AI workflows. Would love feedback if you’re curious:
https://www.producthunt.com/products/llmonster-mobile-ai-agents
@ekusiadadus Thanks for sharing! I have taken steps away from the code review process. I feel it is a bottleneck now. I am still not able to follow a test driven approach but I am improving the test coverage. CI workflows are built by Claude Code and run automatically on GitHub, giving me a lot of confidence.
I use Claude Code only. Well I have used Gemini CLI and Google Jules a few times but nothing comes close to the quality of code generated by Claude in my opinion. I even get GitHub tickets written by Claude Code now.
@ekusiadadus The mobile flow would be a nice to have. For the near future, I am thinking about getting a more portable laptop than what I have (a heavy gaming laptop). I have been thinking about a self-hosted server based solution with mobile apps but I have not decided about it yet.
LLMonster is not a coding agent, right?
@brainless LLMonster is not so much an agent itself, but rather just something for managing and triggering agents...
After all, the agents are being built by people who are far more talented than me.
Our sole purpose is to create the most user-friendly UI possible. If we can operate solely through smartphones without writing prompts, we hope it might resonate even a little with the 96% of unpaid ChatGPT users.
@ekusiadadus Do you think not writing prompts is the way in future? I thought prompts are the future, instead of the current UI/UX. By prompts, I do not mean heavily crafted texts for language models, rather just text that (hopefully) models understand easily.
@brainless I believe that not writing prompts is the future direction.
This is because our ideal, as depicted in many movies, is that we don't deliberately give prompts to AI.
From a technical perspective, when you give too many prompts, the accuracy of transformers drops dramatically at a certain point.
The following is somewhat unorganized thoughts in my head, but...
I think about this by separating tools and agents.
Tools work like slaves - humans give instructions and they follow. And it's preferable that this stays within the range of human cognition.
Agents are premised on being treated almost like humans.
If people in the future treat AI like humans, then prompts would become effective. Of course, there are many technical challenges.
However, in most cases, people fundamentally think of AI as tools.
Related, Vercel felt like a super power at first but lately nothing useful is coming out desing wise, deploying a demo mockup to show my design team
@howell4change do you mean v0? I tried it once or twice. It felt good for a landing page but I feel I have better control with the CLI tools. Claude Code is pretty much my full-time partner in code. Gemini CLI sometimes.
UI/UX Design, taste and understanding of human psychology lacking in my opinion. Would be great if there was a tool or add-on that would hold the LLM to a higher stander of design while building "closer to the spec or brief the user has in mind".
@chris_hage yes in visual design I guess this needs to get better. It is not stopping lots of products embracing LLM generated UI though since so much of UI/UX is standardized and most products want/should stick to common patterns. LLMs have seen more of these patterns than any human can learn in a lifetime and this changes the productivity side of things. Maybe not the most creative but definitely productive.
Velocity
@chris_hage Hey Chris, I'm a product designer and I've tried out Lovable, Bolt, Cursor, VO and Figma Make and have found Figma make the best for UI/UX Design and taste then V0 was the next best. I still think these tool do have a way to come but the speed at which they are improving is really promising.
@chris_hage @rachel_long_smith I will check out Figma make, thanks for sharing.
Velocity
@chris_hage @brainless No worries, I suppose the only downside is the cost because you pay for the full Figma suite which is great for designers as we are alredy using Figma. But I'd then suggesting trying V0.
@brainless @rachel_long_smith Curious to hear, what do you think is lacking in the space for designers (i.e what would you build and why). Given the pain points you have seen that are being unserved.
Tools like Claude Code and Cursor have absolutely changed the game, but I keep running into one weird frustration:
Too much velocity, not enough visibility.
@suvam_deo yes you are right. We were just chatting about this somewhere else. The hidden parts of production software hits non-technical people hard. Everyone can see "code" generated. But how to test, where to host, deploy, stream errors, manage issues, take backups... We are now getting 10x code but the rest of the pipeline is a bottleneck.
@brainless Oh got you. That's something that we should think. Thank you for rasaing this Sumit.
DaVinci Resolve – still cannot drag and drop some elements in the timeline and move them on the x and y axis 🥲
@busmark_w_nika interesting. I am not sure I have tried that or maybe I do not remember having this issue. I use DaVinci Resolve but my video content creation process is slow and happens once in many months.
BTW, I meant tools and workflow related to vibe coding but glad that you shared :)
@brainless Aaaa, in that case: When the editor will rewrite also good and useful parts of the code (despite you said: Don't change the previous code) :D
@busmark_w_nika LOL this is a painful one. What editor and AI agent combination are you using?
@brainless I used v0... but it was a few months ago :D
Hey Sumit & fellow vibe-coders
I was deep in a side project when ESLint & hydration bugs stacked up,700+ issues. Out of frustration, I yelled at @Cursor to "think outside the box"… and shockingly, it did. It generated layered fix scripts on its own, config, patterns, hydration, best practices, and more. In 2 passes, it cut the errors by 90%. What started as a vent turned into a full-on 6-layer CLI architecture. I moved everything into one orchestrator file, fix-master.js, and that’s how NeuroLint was born.
Then @Lovable tried to “help”... and couldn’t even run it. It gave up and defaulted to mock code 🤦: Read Full Story Here
@alcatec Thanks for sharing. I will read through your post. But how did you get to so many issues at once in the first place? My usual work mode is to try hard and not move on from current issue till codebase looks good. I have some tests, full format and lint checks in backend and frontend apps.
@brainless Thanks Sumit!
Actually, I’m a non-technical founder. I taught myself WordPress, but when I started using Cursor just a few months ago, I had no clue about TypeScript, JavaScript, React, Next.js, CLI, or even how VS Code worked. I just wanted to build my app (Taxfy.co.za) using AI tools.
A few months ago, I wasn’t even using half the jargon I know now! My wife jokes that I’m “getting smarter” because she sees me working in a code editor every day, that’s how new all this still is to me.
So unlike you, with 15 years’ experience, I didn’t actually understand what was happening when NeuroLint started forming. All I knew at the time was that I had 700+ errors, and somehow Cursor generated scripts that fixed most of them. That was all I cared about.
It’s only recently, after learning more about development, that I realized how valuable those scripts actually were and that they’d evolved into a proper system. That’s how NeuroLint happened, without me even planning it. Now I’m learning as fast as I can but honestly, I need someone like you to help guide me through this properly.
Hi Sumit！
The biggest frustration is that when I use vibe coding, it is OK for small projects, but when it comes to medium-sized and large projects that require complex front-end and back-end, you can no longer simply trust AI to achieve your programming goals. In addition, I have also tried to output the document first, and then let AI execute according to the document distribution, but the effect is always a headache; I also let AI test itself and deliver it after it passes, but obviously, it is still not satisfactory. This is the problem I encountered. It has not been solved so far.