The 2025 Stack: How We're Evolving Our Bubble Agency with AI Code

We build with this stack daily in the AI side of our agency and share everything we learn with each other inside the new private Skool community. Idea to App 

We hope this blueprint gives you a powerful head start.

Enjoy!

‍

By Zubair Lutfullah Kakakhel 

‍

Skip 6 months of research & avoid wasting $1000+

Here's the exact stack our Agency is using after 5 years of being a bubble agency.

Including notes about why we are switching & why these tools.

‍

Table of contents

Wake up call by 2 of our clients

The Mindset Shift

Our advantage going from NoCode to Code

The Two Paths Forward - Swiss Army Knife or Professional Toolbox?

What About Bolt, Lovable, and v0 etc

The overall stack direction & Front-end hunt

Database/Backend Hunt

Hosting

Various other bits to complete the picture

What about Security (is Vibe coding secure)

Do I need to learn to code? What is the the Learning Path

Why These Specific Choices Win

The Alternatives I Almost Chose

Some topics for agencies

‍

Wake up call by 2 of our clients

Two client stories forced our agency to grow.

Client #1: The Shock Goodbye

2-year relationship. Multiple rounds of iterations on their MVP.

A bit of silence..

Then this email arrives: "Thanks for everything! We've rebuilt version 2 in Next.js."

Wait, what?

1. He had zero coding experience & was so non technical!
2. We spent ages wireframing & drawing out requirements & then building it.
3. Took him 2 months with AI
4. We only found out after he'd already rebuilt it

I asked him for more background around his insights & this is what he said.

‍

Client #2: We managed to recover from this one.

5-year client since Nov 2021!

• First MVP: We built it (Bubble)
• Business Tool #1: We built it (Bubble)
• Business Tool #2: We built it (Bubble)
• MVP #4: Client: "building this one myself with Cursor"

He showed us his Cursor setup & some stuff he had already built using code for his wordpress website.

The Moment of Truth

That's when we knew. We needed to dive into this head first.

He actually forced us to take the leap. Really hard but looking back I’d say that is probably the best thing that happened to our agency.

We told him we are switching & managed to retain him & are rebuilding his entire internal business in code now.
Eternally grateful for him & his trust in us.

The Mindset Shift 

I need to be honest about something that nobody talks about.

It is really hard to become a beginner again

I remember my first day with Bubble..

I didn't have any web development background. Just some WordPress tinkering. No clue about relational databases, privacy rules, or API workflows.

But I had something powerful: Beginner's mindset. That awe and wonder. That "I'll figure this out" attitude.

Staying in the comfort zone

After 5 years, we had it safe. A decent portfolio, all marketing bubble, all messaging bubble, all internal training & processes & onboarding bubble. All testimonials bubble. Everything bubble bubble bubble..

Then I had to look in the mirror and admit:

I need to start over & everything needs to change.

The hardest part to change is me.

Where to even start?

• Python? Ruby? JavaScript? PHP?
• React? Vue? Angular? Svelte?
• AWS? Google Cloud? Vercel?
• PostgreSQL? MongoDB? MySQL?
• Express? Django? Rails?

The paralysis was real. I spent 6 months and thousands of dollars testing a whole bunch of stuff.

But here's what I realized...

We already know 80% of what matters:

✓ Database design & relationships
✓ API integrations & webhooks
✓ Privacy rules & user permissions
✓ User flows & product thinking
✓ Business logic & workflows
✓ What users actually want

What we need to learn (20% - The New Tech stack & using AI):

• Which framework to use (I'll tell you)
• Where to host (I tested multiple & I’ll share my favourite)
• How to connect the pieces (harder than bubble but necessary)

I don’t like the word game changer but AI is a game changer you know.

It writes the syntax for you. 

You just need to know what to ask for - like learning Bubble's API Connector, not learning to code.

The Uncomfortable Truth About Starting Over

It's soul-crushing sometimes. Really.

But here's what I learned: 90% of this transition is mindset.

Your product skills transfer. Your architecture knowledge transfers. Your business logic transfers.

What doesn't transfer? The comfort. The expertise. The "I know exactly how to do this."

You have to be OK with:

• Charging less initially (while you learn)
• Delivering late sometimes
• Messing up completely
• Feeling stupid daily
• Missing Bubble constantly

"This would've been SO much easier in Bubble" - I say this daily. 

But that's only because I knew Bubble, not because it's actually easier.

Need to rewire our thinking to be more like I just don’t know how to guide AI to do it in code.

The Mindset You Need

The same one you had on Day 1 of Bubble:

• "I don't know this YET"
• "I'll figure it out"
• "This is hard but not impossible"
• "Everyone started somewhere"

The difference? This time you're not learning from zero. You understand databases, logic, user flows, and what makes great products. You just need new tools.

Our advantage going from NoCode to Code

I’ve tried hiring developers for our agency & recently realized something. We have a distinct advantage in some aspects over them.

Whoever I’d hire/interview, I’d get a very very specialized skill set only.

I’d ask questions about which platforms they had built & get strange answers. 

• Who made the SaaS
  
  • I made it
• Great walk me through end to end.
  
  • It was a team effort of 6 people.
• Huh, why?
  
  • I only know back end development
• Who else on the team
  
  • Separate designer
  • Separate frontend
  • Separate Business analyst
  • Separate Dev Ops 
  • Separate Architect

I literally hired a ‘Technical Project Manager’ who had done A - Z work with clients but it turned out he only knew how to get others to do the work. He had gotten too far from the code/servers/real deal.

Bubble devs inherently have a lot more breadth in skill & that I think is an inherent advantage.

The Two Paths Forward - Swiss Army Knife or Professional Toolbox?

What Bubble Really Gives You (and it is a loooot!):

• Visual development
• Front end
• A rich plugin ecosystem that works
• Database built-in
• User auth included
• File storage ready
• Hosting managed
• Version control automatic
• Dev/Test/Live environments
• One login, one bill
• A lot more too

The Hard Truth: In the code world, you need 5-10 different services to match this.

But there are two valid approaches that I have found:

Path 1: The Swiss Army Knife (All-in-One)

Replit - The true Bubble equivalent

• Everything in one platform
• AI agent included
• Editor + Deployment + Infrastructure
• ~$25/month starting

Perfect for:

• Solo founders
• Learning AI code
• Quick MVPs
• Simple projects

Why we didn't choose it:

• The agent can start getting expensive quickly
• We’ve got some complicated business tooling apps & it didn’t fit the bill. For simple MVPs, would have probably been fine.

Path 2: The Professional Toolbox (Best-in-Class)

Separate specialized tools for each area.

• Separate front end
• Separate backend
• Separate auth
• Separate hosting
• Separate ide
• Specialized tools that I need to glue together myself.
• Feels slightly harder, but way way more powerful

Perfect for:

• Agencies
• Complex projects
• Maximum flexibility
• No limitations

The Recommendation:

Starting out? Try Replit first. It's the gentlest transition from Bubble.

Running an agency? You need the toolbox. When clients ask for RAG pipelines, WebRTC, or complex integrations, "our platform doesn't support that" isn't an option anymore.

What About Bolt, Lovable, and v0 etc

You've probably seen the demos. "Build a complete app in 5 minutes!"

I tested them all. Here's the reality:

The First 10 Minutes: Pure Magic

• Describe your app
• Watch it build
• See it running
• Mind = blown

The Next 10 Hours: Reality Sets In

• Bugs multiply
• Changes break things
• Tokens cost a lot
• Performance issues
• "How do I deploy this?"

What These Tools Actually Are:

• Bolt/Lovable: Incredible prototyping tools
• v0: Amazing UI component generator
• Not: Production-ready platforms
• Not: Bubble replacements

Our Actual Use Case:

• Day 1: Prototype in Bolt/Lovable
• Day 2: Export to GitHub
• Day 3-30: Build properly in VS Code

The Technical Reality:

• They generate Vite + React (bad for SEO)
• Code quality varies wildly

When We Use Them: 

✓ Client demos
✓ Rapid prototypes
✓ UI inspiration
✓ "What if" experiments

When We Use Don't: 

✗ Production SaaS
✗ Client projects
✗ Anything needing SEO
✗ Complex features

The Verdict:

These aren't tech stacks. They're stepping stones.

Like ChatGPT for writing - great for first drafts, but you wouldn't publish the raw output.

Feel free to use them to go from 0 to 60. Then switch to real tools to go from 60 to production.

However, once you have the real deal, we go straight to the terminal now.

It’s okay to feel overwhelmed. You’re not alone.

The next section is dense. 

It's natural to feel that way when you're learning something new.

Learning this stuff solo is tough.

That’s the main reason I created the Idea to App community. 

We learn and build together.

For example, I’ve put together a 70-minute crash course that walks through this entire stack. 

It covers the end to end complete workflow:

• Creating a Front-end
• Connecting the BackendDatabase
• Production Deployment

It’s just one of the many resources, live sessions, and Q&As we use to make progress faster, together.

You can find it waiting for you inside the Idea to App community.

‍

The overall stack direction & Front-end hunt

I explored everything. Here's my real journey:

Path 1: The "Traditional" Stacks ❌

The Two-Language Problem

Here's what killed these options for me: The web runs on JavaScript.

With Django/Rails/Laravel, you end up managing:

• Backend: Python/Ruby/PHP
• Frontend: JavaScript (because all modern UI needs it)
• Result: Your team needs to know TWO languages
• Mental overhead: Constantly switching contexts

Python/Django

• Promise: "Battle-tested, Instagram uses it"
• Reality: Amazing for data/AI, overkill for web apps
• The "batteries included" is nice (like Bubble)
• But: Python backend + React frontend = complexity

Ruby on Rails

• Promise: "Developer happiness, convention over configuration"
• Reality: It's the WordPress of web frameworks
• Still solid, works great
• But: Would a fresh CS grad choose to become a Rails developer?
• The talent pool is shrinking

Laravel (PHP)

• Promise: "Modern PHP is amazing now"
• Reality: Try explaining to clients why you chose PHP
• Incredibly powerful, great ecosystem
• But: PHP stigma is real, hiring is hard

The Trend Reality Check:

It's not that these are bad - they're excellent. But ask yourself:

• What are bootcamps teaching? JavaScript
• What are CS grads learning? JavaScript
• What does AI understand best? JavaScript
• What runs everywhere? JavaScript

I spent a month here before accepting the obvious.

Time to embrace JavaScript.

Path 2: The JavaScript Decision ✅

Once I decided on JavaScript (AI understands it best), the real choice began:

Pure React

• Feels too minimal..

Vue.js

• Loved by developers, smaller community

Next.js ← The Winner

• Full-stack out of the box
• SEO that actually works
• Vercel's backing = constant improvements
• What successful startups use

Database/Backend Hunt

Here's what I realized too late - we weren't just looking for a database. We needed what Bubble gives us: database + authentication + file storage all in one place.

The nightmare scenario? Imagine cobbling together:

• PostgreSQL for your database
• AWS S3 for file storage
• Clerk/Auth0 or Okta for authentication

Now you're managing three different services, three different bills, three different APIs, and when something breaks, good luck figuring out which service is the culprit.

The Backend-as-a-Service Hunt:

Firebase Google's all-in-one solution. Works great until you realize you're locked into Google's ecosystem. After five years of Bubble lock-in, I was allergic to vendor dependencies.

Xano We have a client paying $300/month for Xano. For a database. That's more than their entire Bubble plan. The math doesn't add up.

Convex.dev This one got interesting. Database + auth + storage + background jobs (the holy grail!). But then I saw "authentication is in beta" and they recommend using Clerk instead. Beta authentication? For client projects? That's a hard no.

Supabase:

Supabase won because it's basically:

• PostgreSQL database (battle-tested, not proprietary)
• Authentication that actually works (social logins, magic links, the works)
• File storage with proper access controls
• All open source (if Supabase dies tomorrow, you can self-host)
• Dashboard that doesn't require a PhD to understand

The only thing missing? Background jobs. There are edge functions but not really proper background jobs..

For that, we added Trigger.dev. Amazing powerful platform for background jobs. Open source too & thriving community.

Hosting

This decision took months and even a DevOps consultant. Here's the journey:

Vercel - Everyone recommends it. Twitter loves it. Then you hear the horror stories. One developer's infinite loop, one DDoS attack, one viral Reddit post, and suddenly you owe $5,000. It's like Bubble's workload units but possibly worse with AI code. We couldn't give clients this risk.

DigitalOcean - Great for developers who want to manage servers. But we're not server managers. We don't want to SSH into Ubuntu at midnight, manage SSL certificates, or figure out why the disk is full. We want Bubble-like simplicity.

Fly.io - Powerful, modern, container-based. Also assumes you know what a Dockerfile is and enjoy reading infrastructure documentation for fun. Next.

Render - Almost picked this. Good pricing, decent UI. but per developer pricing hurt.

AWS/Google Cloud/Azure - Sure, if you want to hire a full-time DevOps person and read 500-page documentation. We're a Bubble agency, not NASA.

Railway.com

Railway won because it strikes the right balance.

• Push code, it deploys. No Docker knowledge required.
• Set a spending limit. Sleep easy. No surprise bills.
• Scales when you need it, but starts at $5/month
• The UI makes sense to me

Now when clients ask about hosting costs, we can say "$5-20/month" with confidence. Try explaining Vercel's usage-based pricing to a non-technical founder.

Various other bits to complete the picture 

Code Storage: GitHub

Why Not: GitLab, Bitbucket: Everything integrates with GitHub

Development: VS Code

Why Not: Cursor,Windsurf.

• Because it is Free.

Analytics: PostHog. Still exploring this space.

‍

What about Security (is Vibe coding secure)

Tl;dr. Vibe coding is insecure. Vibe planning & vibe engineering can result in a secure product. But you need real human guardrails.

I need to be brutally honest about something that could kill your agency.

Security is Not Privacy Rules 2.0

In Bubble, security felt simple. Privacy rules here, conditions there, done. Your data was safe if you got it right.

Not really. It was easy to make mistakes as well. Petter Amlie wrote a whole book about it.

Code is similar. You could argue it is easier to make mistakes as LLM is writing it. But it is also easier to fix things.

My Zero-Risk Approach

Look, if a button is misaligned on mobile, who cares? A little polish issue. But a security breach? That's unacceptable.

Here's what I did (and what I recommend):

For Agencies: Hire a Security Gatekeeper

We brought in a security contractors. One hour per day, spread across projects. Their job:

• Review every pull request
• Flag security issues before production
• Train the team on what to watch for
• Own the security outcomes

Was it hard to find them? Yes. Is it expensive? Yes

I think of it as insurance, not cost.

What They Actually Do:

• RLS policies
• Middleware authentication
• API security patterns
• Environment variable management
• SQL injection prevention
• XSS protection

The moment our contractor said "just use middleware like this," everything clicked. We could guide Claude properly. Without that knowledge? We were guessing.

For Freelancers: Find a Security Buddy

Can't afford a contractor? Find someone in the full-stack space who'll review your code. Trade services. Pay for code reviews. Join security-focused communities. Do something but do not ignore this!

Because here's the truth: AI can write insecure code that looks perfectly fine.

The Bottom Line on Security:

In bubble, infrastructure was their issue & application layer yours.

In code, you also choose the infrastructure layer.

Do I need to learn to code? What is the the Learning Path

The question everyone asks: "Do I need to learn to code?"

Honest answer: No. Not in 2025. As models get better, this will fade even more.

What we actually need to learn: How to guide AI. How to possibly navigate your way around code & read it roughly at a high level.

This area is being called a few different things. My favourite term is specification driven development. But some are calling it Vibe Planning & Vibe Engineering. The term Vibe coding is a bit muddled up.

A bubble analogy

Think about it - if someone doesn't know Bubble's API Connector exists, they'll be stuck when they need external data. Same with scheduling API workflows on lists. You don't need to understand HTTP protocols, but you need to know these features exist and when to use them.

AI coding is identical. You need to know what to ask for, not how to build it.

What "Learning AI Code" Actually Means:

Instead of syntax:

const fetchData = async () => {

  const response = await fetch('/api/data');

  return response.json();

}

You learn concepts: "Claude, create an API endpoint that fetches user data and handles errors properly."

What You DO Need to Learn:

✓ How to describe features clearly

✓ How to read AI's output 

✓ When something looks wrong

The Claude Code Advantage:

Claude Code in its current form is magical. It just... gets it. You describe what you want in plain English, and it builds it. The only times it gets annoying are with super complex edge cases - and even then, it's more about patience than skill.

Your Bubble Skills Transfer Directly:

• Database design → Same concepts
• Workflows → API routes and functions
• Privacy rules → Row Level Security
• Conditions → JavaScript conditionals
• Custom states → React state
• API Connector → Fetch requests

The syntax changes. The thinking doesn't.

The Honest Truth:

You'll be productive in 2 weeks. You'll be confident in 4 weeks. You'll wonder why you waited 6 weeks.

The hardest part isn't learning. It's unlearning the fear that "real code" is hard.

Concluding thoughts & a request

I was silent for the first 6 months of 2025 because I was knee deep trying to figure all of this out.

Now I’m learning & sharing everything in a new community. 

Join me in this journey

But building & learning together is a whole lot more fun than doing all of this silently.

There are weekly live calls for friendly networking + Q&A

I have already recorded a 70 minute course on

• Setting up your environment
• Understanding the stack
• Building with AI
• Deploying your first app

And lots of content on our new vibe framing discovery process which is super fast & bypasses figma/ui/ux designers.

& I’ve got lots more planned.

Link : https://www.skool.com/idea-to-app-3209/about 

Why These Specific Choices Win

Next.js Wins Because:

• It's full-stack (no separate backend)
• SEO works (Bubble's weakness)
• Vercel backs it (constant updates)
• Every AI knows it

Supabase Wins Because:

• Closest to Bubble's database
• Auth included (like Bubble)
• Open source (unlike Firebase)
• Scales infinitely

Railway Wins Because:

• No surprise bills (looking at you, Vercel)
• Simple as Bubble hosting
• Predictable scaling
• Great for agencies

The Alternatives I Almost Chose

Replit

• True all-in-one like Bubble

Convex.dev

• Combines supabase & trigger.dev

‍

Some topics for agencies

Managing a 12-person Bubble team transitioning to AI code? 

Here's how we're doing it:

Our Transition Strategy:

First Project: Watch and Learn We subcontracted to a code agency. Painful to watch someone else build, but we learned the infrastructure basics.

Second Project: Guided Practice

• 8-week timeline (could've been 4 with experience)
• 4 developers on what should've been a 1-person job
• Daily contractor coaching
• Pull request reviews
• Lots of "aha!" moments

Current State: Mixed Reality

• Some teams still in Bubble (ongoing projects)
• New projects in code
• Weekly knowledge sharing sessions
• Devs teaching devs what they learned that week

‍

The Awkward Reality of an upside down org structure.

Our junior developer who knows Code is now teaching senior Bubble developers. Yes, it's upside-down. Yes, it's humbling. But in AI world, recent learning beats past expertise.

The Surprising Response:

I expected resistance. Instead:

• "When do WE get to try code?"
• "Can I join the next code project?"
• "This is actually exciting"

Why? Because I explained the why. Showed them the client who left. Demonstrated the speed difference. Made it about growth, not replacement.

Practical Training Tips:

1. Over-resource initially - 4 people learning is better than 1 person struggling
2. Pair with contractors - That daily guidance is gold
3. Extend timelines - Your first project will take 2x longer
4. Share learnings weekly - One person's discovery helps everyone
5. Let juniors lead - If they know more, let them teach

The Hidden Advantage:

Our team's CS backgrounds helped, but honestly? The Bubble mindset of "I can figure this out" matters a lot more I think.

For Agencies Considering This:

Start with one small project. Pick your most curious developers. Give them time, resources, and permission to fail. The rest will follow.

Remember: Your team already learned Bubble from scratch. They can learn this too.

Have Questions?

Reach out directly:

• Twitter: https://x.com/ZubairBuilds 
• LinkedIn: https://www.linkedin.com/in/zubairbuilds/ 
• Email: zubair@azkytech.com 

P.S. Still on the fence? I get it. I procrastinated too. 

The best time to learn was 6 months ago. The second best time is today

Link : https://www.skool.com/idea-to-app-3209/about 

For just $25/month

‍