So… What’s in Your Vibe Coding Stack Right Now? (2026)

Right now my vibe stack is pretty lean. I'm building mostly with ChatGPT as my partner, Cursor for fast iteration and Vercel to ship without friction. I connect ideas first, code second. Mostly shipping small, useful tools fast.

My current vibe stack is all about momentum. ChatGPT for product thinking, Cursor for coding in flow and Replit when I want zero setup. I'm shipping scrappy MVPs and internal tools, optimizing for speed and clarity over perfection.

I'm running a mixed vibe stack. ChatGPT helps me reason and plan, Claude helps with clean refactors and Cursor keeps me moving without breaking focus. I'm shipping experiments weekly, learning what sticks by releasing instead of overthinking.

Lately I keep it simple. ChatGPT as my main collaborator, v0 for quick UI drafts and Vercel to deploy instantly. Everything connects around fast feedback loops. I'm mostly shippiing micro-products and validating ideas in public.

I'm treating my vibe stack like a studio. ChatGPT for ideation, Lovable for rapid UI exploration and Replit for quick demos. I connect them loosely and focus on shipping prototypes that feel real enough to test with users.

Below is the full stack I’ve used to build 10+ products so far.
Feel free to reference it. For me, it’s practical, cost-effective, and works well for solo builders

• Main coding: @Codex by OpenAI + @Replit
  Replit helps me quickly get a first working version with a decent UI, which is usually the hardest part to start from scratch with most coding agents.
  Since Replit can get expensive, once the UI feels right, I switch to Codex to continue backend implementation and further development.

• Database & Auth: @Supabase
  My favorite backend tool. It’s not just a database, it also handles authentication, email sending, email verification, and policies out of the box.

• Deployment: @Railway
  Simple setup and auto-deploy when I merge into main. Very smooth for small teams or solo projects.

• Domain & DNS: @Namecheap.com
  Straightforward domain management and DNS configuration.

• Testing / Validation: @Scout QA - AI Quality Companion
  I use it to verify UI and core flows before announcements or launches — especially helpful when building fast with vibe coding.

ChatGPT for brainstorming, Claude Code for implementation, Railway for execution. Just moved my AI app from Replit to Railway because Replit was burning $ real fast!

I'm mainly using Claude Code (I'm on the MAX plan), and use ChatGPT (Plus) for small tasks and concepts. I have tried Jules but that did not really work for me, as my environment couldn't run on the instances (Django with Posgres, Redis, Celery...).

I am quite happy right now, I am only a bit sad that I cannot fully utilise MAX as I'm only working on my own software and SaaS on the side (about 10-15 h/week, sometimes 20h).

I got infected by the whole AI Assisted Coding by Peter Steinberger, Mario Zechner, and Armin Ronacher late spring after having used Claude Web for coding, only. I had my fair share of "I can't sleep good anymore".

I then took an utterly needed break over the summer, and after that attended Answer.ai's/Jeremy Howard's SolveIt course which taught me "how to code slow with AI" (highly recommended! Great for learning and to really be confident with what you produce.).

In fall, I attended the amazing "Elite AI Assisted coding" course by Eleanor Berger and Isaac Flath which helped me get a good balance with AI.

I have some MCPs when needed (eg. Playwright) but actually don't use them much anymore, as I think most MCPs bloat the context too much. If I need something, I often let Claude generate a short Python script that can do the job and note it in CLAUDE.md or simply add a make command for it which Claude usually also just finds.

I also still have active GitHub Copilot and Replit subscriptions but don't really use them right now. I have briefly used GH Copilot as a review agent in a GH Actions Workflow but don't use it for that, right now.

I'm happy with Claude, but if I have a bit time, I'd test Codex CLI.

Just stumbled across this place… bloody loving it, great post mate.

Been properly testing AI models for the last 2–3 weeks now — got a list of about 100 of them compiled and chucked into an app for anyone to mess around with. Still just testing phase so don’t send the whole internet my way yet haha, but here’s the link for your perusal anyway:

Project Pulse – AI & No‑Code Platform Directory

Built the whole thing on v0, and honestly, gotta say it’s one of the most straight-up honest platforms out of everything I’ve been hammering lately. Weirdest one I’ve hit so far is Wix.Vibe — they straight-up fabricate results and tell you everything’s peachy when it’s clearly broken. Call them out on it and the system turns into a frightened schoolboy caught with his hand in the cookie jar.

Then you’ve got places like Base44 and Replit — you fall head over heels early in a project, feels amazing… until the paywalls slam down and suddenly it’s frustrating as hell whether you’re a total newbie or a seasoned dev. Time to start hunting for another route.

Overall though, I still can’t get my head round how fast this tech has moved already. I look at the AI model I’m building right now and just think “shit, I’m miles behind the curve already”. Synthetic AI models are only just starting to properly emerge — fingers crossed it won’t be long before humans aren’t even typing code anymore. Let the machines handle all the boring menial crap so we can sweat the actually hard, interesting bits instead.

Good luck on your projects..

May the 4th and all that hoohar

Dan