Has anyone here actually Vibe Coded a product all the way to production?
Posted by Badgergeddon@reddit | ExperiencedDevs | View on Reddit | 25 comments
I use Cursor plenty for simple stuff, but I'm just not buying the hype train. For me, AI always falls over when some complex bug comes along or requirements get complex. It feels like all vibe coding full apps is going to do is make a load of tech debt and a complete trash fire of a codebase. Today's models seem way off the promises. Am I wrong?
Curious-Stomach8957@reddit
I found this solution very helpful and I am currently working on implementation:
Prompt " Now I want you to help me transitioning from a Prototype (what I have now) to a Production App. Consider that - if possible - I want to use only Google's services."
This is a significant step. Moving from a "Prototype" (a single file running in a temporary preview) to a "Production App" (a real, live application that saves data and has users) requires a shift in architecture.
Since you want to stick to Google Services, the answer is unequivocally Firebase. Firebase is Google’s app development platform that provides the database, hosting, and authentication all in one bundle.
Here is your roadmap to production, visualized.
The "Google Stack" Architecture
Currently, your app is just the "Frontend" (React code). To make it production-ready, we need to connect it to the cloud.
Here is the 4-Step Plan to migrate your Code Assistant prototype to a real Google-hosted app.
Step 1: The "Great Refactor" (Break the Monolith)
Step 2: Data Persistence (Firestore)
Step 3: Authentication (Firebase Auth)
Step 4: Hosting (Firebase Hosting)
Your Immediate Next Step
You cannot do this entirely inside the current "Code Assistant" chat window because we need to install software packages (Firebase SDKs).
I recommend we start by cleaning up your code structure so it is ready to be moved.
CleverThunder87@reddit
I felt the same way for a while. Cursor is great for quick scaffolding, Replit AI was fun for prototyping, and Lovable was solid for UI drafts, but none of them survived once the project hit real requirements.
The only time vibe code didn’t implode on me was when I paired it with tools that already had the rough parts prebuilt. For a healthcare prototype I did this year, I used Specode for the intake and onboarding flows so I didn’t have to glue together PHI logic myself. Then I used Cursor and Replit for the custom parts.
That combo actually made it to production without turning into a junkyard of tech debt. The trick for me was not letting AI generate the entire codebase, just the parts that don’t matter as much.
youroffrs@reddit
My entry into vibe coding was through Blink.new. I was using tools like Airtable and Coda, but Blink let me build a full web app just by describing it. No code, no hassle. It handles everything design, backend, auth, database, hosting like a pro. I had an MVP up and running in under an hour. Super beginner-friendly and surprisingly powerful. Definitely worth a try if you're diving into vibe coding!
ngtwolf@reddit
A little old discussion since I stopped by to check this topic as well. I do have a lot of little project i 'vibe coded' either for myself and at least half of them went south after a bit and had to be restarted. Oh, and lots of yelling at the AI, so if the AI does become sentient, it's likely coming after me first. :)
That said, my site projectrollouts.com was completely vibe coded. I originally just built the tools for myself but then since I had the domain name, i put them up with a full UI and it's 100% complete.
AppointmentDry9660@reddit
Yes. I am a 13+ year dev who "vibe-coded" a blog with the ability to write comments to production (SSL certed) and I plan on documenting it soon. I am going to make some more projects once I figure out a rather glaring security flaw that I don't want to become a problem when I document / open source the whole thing
I had never launched anything app related before, they were all at home "for fun" projects. For work, I work on a cloud that is unrelated to anything I vibe coded in, other than it can use JavaScript sometimes.
Dry_Atmosphere_8029@reddit
Absolutely. Just built a full react app for my strata (violation system) that allows council, and onsite management to create violations in real time, take the evidence submit it to council for approval and if a fine should be issued or not. It automatically emails the owner and tenants allows them to dispute.
Pretty fun took me a month? Lots of QA, lots of refactoring, lots of making it secure, CORS JWT, everything security related was a god damn nightmare. Being relatively new to coding I thought everything was being stored in a database ready for production. Nope - SQLite which needed to be migrated to Postgres.
I learned so much. I’m learning how to code on my own, I’m learning how to debug on my own. Vibe coding for me is a way to augment my learning - and it’s been working amazingly!
Just deployed everything to a dedicated Smurf server at ovh secured it, deployed caprover and it’s running smoothly!
barney_stinsonj@reddit
i think its evolving
kaorzildrTheWise@reddit
The way that's worked best for me when building a start up app was to treat it as a starting point, a pair programmer for ideas, and overall productivity booster - not a replacement for me.
For example, say I got a new design that's using graphs from Chart JS and cards from material UI together to present the info.
I usually start with sending the design into chatgpt, ask it to help me create some of the base component structure to match the design.
If there's a new API, I asked it to create the dto objects that I need based on sending it an example of how I create them in the code base.
By doing this, I save a lot of time to get some of the base structure of a new feature set up quickly. Next, I might make some edits to clean things up, get things closer to where the design is, or ask ChatGPT to modify something specifically that it missed or doesn't look correct.
I'll finish up the 10% of the features from there manually, clean up the code, and make sure I'm feeling good about it.
After that, I'll send it over the components and ask it for feedback like a reviewer, get performance tips, things like that. It'll make suggestions and edit the components, so it's a quick way to get that out of the way then doing each one manually myself.
I then put up the PR, get any final review from our AI reviewer, decide to incorporate it or not, and then merge!
For me, this workflow has gotten me to build some pretty complex stuff in 1/4 of the time I'd say, especially for side hustles/start ups on the side, it's exactly what I want it to help me do - code faster :D.
Prestigious-Leek-948@reddit
Love the workflow — totally agree that using AI as a productivity booster rather than a full-on replacement is the sweet spot. If you’re ever feeling the backend side is slowing you down, I’d recommend checking out Robodev. It’s been a game-changer for me, generating solid backend code directly from clear specs.
There’s no open access yet, but if you sign up via app.robodev.com/signup you’ll get three months free to test it out.
LeadingFarmer3923@reddit
Totally fair take, you're not wrong. AI tools like Cursor are great for quick iteration, but once complexity hits, they often crumble without context or deeper planning. Vibe coding can give you a head start, but turning that into a maintainable product without racking up tech debt needs actual design thinking. What’s helped me is treating AI as a junior pair, not a lead dev. Try planning first with tools like stackstudio.io, they can map out structure from the start so you don’t end up rewriting everything later. The tech's evolving, but for now, it still needs guardrails and real engineering behind it.
DisastrousSecret7062@reddit
Its now become a PM thing as well. PMs are doing vibe coding to build weekend prototypes
These_Translator_488@reddit
https://www.threads.net/@ar7work/post/DHUY5sjyirT/vibe-coding-in-a-nutshell
mylons@reddit
i'm vibe coding through a rails 7 -> 8 upgrade. my client had a test suite that was just abysmal. vibed through parallelizing the tests first, runs in 2-5 minutes instead of 30-60 minutes. could probably vibe harder and make it work in sqlite but not enough time on the contract.
vibes through config changes np.
i personally think this is more interesting than just shipping a product with it. i know i could do that given what it's done with this 10+ year old rails app.
unflores@reddit
Yeah. There's probably a certain point where you should bail. Afterwards, your experience will give you a good intuition on that I think...
loumf@reddit
I helped someone who did. He knows basic HTML and CSS, but is a web marketer, not a programmer.
He vibes programmed a membership site that connected to Luma and Stripe using Replit. It was in Typescript using React/Node/Express and Postgres.
I had 3 1-hour sessions with him to fix bugs he couldn’t get Replit to solve. He cannot read Typescript at all
But, he does understand what goes into making a website like this work. He understands the basic ideas of a frontend and backend, a database, a REST API. He has project managed sites like this.
It was quite impressive.
loumf@reddit
To me the code was not that different in messiness from code written by humans. Lots of WTFs, but not more than usual
coredusk@reddit
After the AI hype dies down, non-vibe developers will be high in demand to fix all the AI code.
rlbond86@reddit
The guy who invented the term said it was for low-stakes weekend products
InflationOk2641@reddit
I'm adapting one of those low-stakes open source weekend projects that tracked missing buses, to slightly customise it for a different region. I figured it would be an hour of effort. There's no separation of concerns in the code, data about buses is statically written in the code, no interface for the database access. I ended up having to rewrite all the logic. Of course that means the time invested to adapt it is several times longer than anticipated. My impression was that it was written by someone who had less than a year, of programming experience and no experience on how to structure code properly
Yourdataisunclean@reddit
Yup, it has great potential for prototyping, but not something that you would really push to prod.
Startup land will get interesting because if you're not careful Minimum Viable Products (MVP) will end up being Minimum Viable prototypes (MVp). This could be good for doing more experimentation, but could create issues once you find one you want to exploit and push out to more customers.
Time-Mode-9@reddit
I'm a senior dev. Copilot is like next level Google, but good luck trying to get anytbing spanking more than a couple of hundred lines of code to work, especially if you're doing anything that hasn't already been done and documented online
justUseAnSvm@reddit
Yea, I mostly vibe coded a weekend project.
xiongchiamiov@reddit
I haven't heard of this before, so for others: https://en.m.wikipedia.org/wiki/Vibe_coding
I'm not sure what this has to do with vibes other than being a Gen Z thing, but (shrug).
As to your question:
Senior engineers have been doing this for a long time. They just used junior engineers instead of LLMs.
Mrqueue@reddit
Obviously not
tyler_church@reddit
I think this very much depends on definitions. "Products" come in all sizes. "Production" for a static marketing site is very different than a complex backend service.
I'd be willing to bet there are static marketing sites live today that were completely vibe coded.
Beyond that level of complexity, you need more and more human intervention for things like deployment, database setup, etc. At which point I think it's disengenuous to call it "vibe coding" anymore.