Skip to main content

My Vibe Coding Stack So Far

 


I've been doing a lot of vibe coding lately, and one tech stack emerged for me as the most practical. Spoiler alert 1: it's Firebase + Codex + Google Cloud. Spoiler 2: tomorrow it could change.

Vibe coding is my latest obsession. I tried many tools, from pioneers like Lovable, Replit, Bolt, to coding assistants like Cursor, Windsurf, Antigravity, to foundation-model-backed Codex and Claude Code, to all-in-one platforms like Firebase and v0/Vertex.

First things first - there is no "one best tool"! The best tool for you will depend on many factors: your skillset, the type of product you're creating, your budget, and your taste. And then, it could change really quickly. A big part of the vibe coding fun is how rapidly the tools evolve. Something that was not possible yesterday might be easy today and implied tomorrow. You can observe the model learning and the tools getting better almost in real time.

So the first consideration is the code-writing model itself. But then, and it might be even more important, is the infrastructure and supporting tools. Assuming you're not only playing around with prototyping, but want to publish your work and have other people use it? Then you need hosting, database, 3rd party services and APIs. In my experience, infrastructure always takes the most effort to get right. That's why choosing the right platform is of vital importance for any vibe coder.

Firebase

Firebase is the modern app-building platform closely integrated with the Google Cloud. It provides app builders with all the tools needed to kick-start their app, bring it to production and grow it. 

Firebase Studio

Like most other tools, Firebase has its own coding agent called "Prototyper" (most likely based on Gemini) that is quite capable with the front-end of the app. It is the first thing you're greeted with when you go to Firebase Studio. The interface is simple to work with, and initial token limits are very generous compared to other, better-known vibe coding tools. How generous? I've spent hours in the "Prototyper" and went through at least a hundred prompts without ever seeing a token warning of any kind.

Firebase Studio is a wrapper around VSCode, so people who want to get into the meat of things can do that easily. And inside the code editor, there is another coding agent. A few, actually, you can choose from, as well as connecting to a custom model you might have on the side.

Firebase Console

That's the main place where you can add and configure services you might need for your app. Need a database? Here's Firestore. Need authentication? Here's Authentication. Need hosting? Here's Google Cloud Hosting. Need Gen AI? Here's Genkit. Need analytics? Here's Google Analytics. Adding and configuring those services is fairly easy, even for someone without prior DevOps or Engineering experience.

Firebase offers a few billing plans: a free tier and a pay-as-you-go, premium plan. It's perfectly possible to create a fairly complicated app, publish it, and have dozens or even hundreds of people try it without paying a cent to Google or Firebase.

Google Cloud

I came to Google Cloud as a total amateur. Previously, I never managed a cloud infrastructure and had only a fragmented, theoretical understanding of AWS. My first impression of Google Cloud was "it's a lot!". It took me a few days to get my head around basic building blocks such as hosting, Cloud Run, IAM and observability. I still only discovered a tiny bit of what Google Cloud offers, but it was enough to publish my three projects. And after publishing, it's pretty much self-managed. The cloud scales automatically, and you can have flexible controls over your costs.

Codex

Codex is OpenAI's coding agent available on the web and most recently as native MacOS and Windows apps. The interface (of the web version) is pretty straightforward: you describe a task, it codes or analyses, you review changes and merge them into your codebase through the GitHub integration.

I built three different projects with Codex: howexpensive.in, 3smarter.com and advantagescout.com. All those are pretty different in scope and fairly backend-heavy. Codex was able to handle those projects competently, but not without issues. Some of those issues are self-inflicted due to the rest of my stack. Because I use Firebase, Codex doesn't have access to my local environment and can't effectively test its own code. For debugging, it has to rely on static code inspection or/and extensive logs I had to add to my apps.

Also, a few times, I ran into a classical vibe coding hallucination loop when Codex tried to solve the same problem over and over again, producing tons of code but no real solution. Now I learned that if a few attempts at the same prompt didn't provide a satisfactory solution, I need to reframe the problem. Here's where /plan mode proved to be very helpful in Codex. In the planning mode, Codex doesn't write any code but can do analysis and create tasks to solve a complex problem in a sequential manner.

Another reason I am sticking with Codex for the time being is the speed-to-results ratio. A lot of other tools I tried were producing similar results to Codex, but working for longer or costing more.

My vibe coding process so far

Here's what I do when I have a product idea I want to build.
  1. I go to Firebase Studio and use Prototyper to sketch out the idea
  2. If I still like the idea and want to continue, I create a new GitHub repo and go to Codex
  3. I run Codex in plan mode, explaining on a high level the product I want to create. I ask Codex to suggest an architecture and development plan, providing it with the set of constraints based on my Firebase setup
  4. Sometimes I would cross-check Codex suggestions in another coding agent (making AI agents fight each other on ideas could be fun)
  5. I would task Codex with backend work and adjust the front-end with the Prototyper inside Firebase Studio
  6. When I want to share my work, I'll publish my project to Google Cloud and configure Firebase Console for production use
  7. When I want changes or new things in my app, I'll create a new branch and use Codex... and the cycle continues until I run out of credits or ideas :)

Pros and Cons

What works about my process? Well, I was able to ship seemingly working apps on my own, within a few weeks of occasional work and without spending a fortune. A year ago, that wouldn't have been possible for me - I couldn't code and had only limited technical understanding of app architecture and cloud infrastructure. This single fact, that someone like me can now fairly easily realise almost any of their ideas, is amazing!

Other pros include

  • Opportunities to learn how apps are being built and hosted
  • Control over my code and hosting
  • Separation between the coding agent and the database/infrastructure
  • Visual editor and code editor in one
  • Amateur-friendly Cloud infrastructure setup, I can figure out
  • A glimpse of a chance to create something of value
And what about the cons? The main one is the expectations. If you expect to start with the "build me a million-dollar app" prompt, you'll be disappointed. For vibe coding to be useful beyond prototyping, you need to have some foundational understanding of how software is built. Or at least you should be eager to learn. Prepare your first project to be a throwaway. And the second. And maybe the third... and...

Some other cons I've encountered

  • Debugging can be a pain when Codex cannot build your app locally
  • Some Firebase and Google Cloud services are tricky to configure (looking at you, IAM and Firestore permissions)
  • Console is the most powerful tool on the Internet, but pain to use
  • A pay-as-you-go plan is great until you make a mistake and suddenly hundreds are taken from your credit card (yeap, that happened to me, some lessons hurt)
  • An existential crisis when you realise creating valuable products is super hard...
... but still fun.

Do I stick to my stack?

For now, yes. But I keep my eyes and ears open for new suggestions. AI tools are developing so quickly that you just cannot stop exploring the new, potentially better tools out there. 

It's a great time to be a product builder right now - all you need is time and a command of a language to realise your long thought-through ideas. And the great time for consumers is coming as well - vibe coding will result in much more choice for users and will drive the price of software down. 

Have fun vibe coding.  

Popular posts from this blog

Product management and operations tools - Jira Product Discovery review

  JPD is a new player in the market of product management software. Jira (and the whole Atlassian suite) has been one of the most popular tool stacks for teams to deliver software products. Now they're adding a missing piece - product discovery.

Product management and operations tools - Productboard review

  The second product management tool I've decided to review is Productboard . It is widely regarded as one of the most popular tools for product teams and the main competitor of Craft.io that I reviewed in the previous post .

PM's AI toolkit - OpenAI Codex and Google Jules

  If I'd try Jules first - I'd have a totally different impression of the state of coding agents today. A much worse impression. Here's why.