Saturday Sessions: Becoming an AI Builder

Hey Prompt Entrepreneur,

You've tried to learn coding before.

Bought the Python textbook. Started the JavaScript tutorials. Made it through "Hello World" and maybe a few loops before your eyes glazed over and you quietly closed the laptop and walked away.

That's normal. You're not broken. The system is.

Traditional coding education is built for computer science students who need to understand memory allocation and algorithmic complexity. They have a couple of years of being dragged kicking and screaming through a curriculum. You just want to build the damn thing in your head.

I speak Mandarin. Not perfectly, but well enough to live in China, chat with locals, and occasionally make terrible jokes that get polite laughs. How did I learn? Not through Duolingo. Not through grammar books. Not through flashcards.

I moved to China and started talking to people. Badly at first. Really badly. But I practised the actual thing I wanted to learn: having conversations.

If you want to get good at conversation, you have conversations.

If you want to get good at building, you build.

Studying a subject is not the same as practising the subject. One gives you knowledge. The other gives you skill. And skills pay the bills.

How many millionaire academics do you know? Yeah…….there’s a reason for that. Information without action ain’t enough.

I'm going to show you a completely different path to becoming dangerous with code. Not through tutorials and exercises, but through vibe coding - the gateway drug that leads to addiciton.

And I’m your pusher.

What The Hell Is Vibe Coding?

Vibe coding is writing software through conversation, not syntax.

Instead of typing function calculateTotal(items) { return items.reduce((sum, item) => sum + item.price, 0) }, you type "calculate the total price of all items in the cart."

The AI writes the code. You direct the outcome.

And importantly we use natural language rather than computer language. It’s hard to overstate what a shift this is.

If you tune into my vice coding streams you’ll see I literally talk to the tool. Through my microphone. I don’t even type. Because i) then everyone will see how bad I am at spelling and ii) I can just talk out loud to my computer and get the results I need.

This isn't "cheating" any more than using a calculator is cheating at maths. It's using tools to achieve outcomes.

Here's what makes vibe coding different from traditional learning:

Traditional path: Learn syntax → Understand concepts → Build practice projects → Maybe build something real → Probably quit before this

Vibe coding path: Build something real → See it work → Get curious about how → Learn the concepts that matter → Keep building

Is the traditional path purer and “better”? Yeah maybe. But who gives a damn if you just stop. The perfect path is pointless if we don’t get to our destination.

The tools that make this possible didn't exist two years ago. Hell, vibe coding was only coined as a term at the beginning of this year (2025). Now we have:

• Lovable - “The last piece of software” (that’s fighting talk!)

• Bolt/Replit/v0/Base44 - Similar to Lovable, slightly different approaches

• Cursor - A full code editor, backed with AI

• Claude Code - Full development environment with AI built in. Technically it’s a agentic command-line tool", a set of indecipherable words which is precisely why we aren’t starting here!

Each serves a different stage of your journey. More on that in a moment.

Why Traditional Learning Fails Entrepreneurs

Here's what happens when business people try to learn coding the "proper" way:

• Week 1: Variables and data types. Okay, makes sense.

• Week 2: Loops and conditionals. Getting confused but pushing through.

• Week 3: Functions and scope. What the f*ck is scope? Why am I learning this?

• Week 4: Object-oriented programming. Close laptop. Never return.

You just spent a month learning and built... nothing. Nothing real. Nothing you can show anyone. Nothing that solves a problem. Certainly nothing anyone will pay for!

Meanwhile, with vibe coding, by week 4 you could have:

• Built a working landing page with email capture

• Created a simple tool your audience actually uses

• Launched an MVP of that idea you've been sitting on

• Made your first £100 from something you built

In fact I’m running through cycles of 30 days ideating, building and launching projects - one hour a day for 30 days - to show what this process looks like: Building AI Apps in Public Playlist.

The "buT ViBe CODinG isn'T ReAl cODiNg" Myth

Let me address the elephant in the room. The gatekeepers who say vibe coding isn't "real" programming. They turn up on my daily livestreams to tell me what tools and frameworks I should be using. It’s exhausting!

These are the same people who told us using WordPress wasn't "real" web development. WordPress now powers 40% of the internet.

They said Webflow wasn't "real" design. Webflow companies are worth millions.

They said no-code wasn't "real" building. No-code startups are getting acquired daily.

Guess what? These buT ViBe CODinG isn'T ReAl cODiNg people aren’t shipping anything. They aren’t building. They aren’t creating.

Let’s look back:

• Instagram's first version was a bourbon check-in app built in 2 weeks. The code was, by all accounts, a mess. Sold for $1 billion.

  

  https://www.feedough.com/history-of-instagram/

• Facebook started as a PHP hodgepodge that Mark Zuckerberg hacked together in his dorm. "Real" programmers would have been horrified. Worth $1 trillion.

  

  https://en.wikipedia.org/wiki/Facebook (and yes that’s the highest resolution available!)

• Twitter was built in Ruby on Rails in 2 weeks. It could barely stay online. The "fail whale" was legendary. Changed the world.

  

  https://www.firstversions.com/2015/07/twitter.html

They were things that worked well enough to prove value. The are lots of jargon terms for this like MVP (minimum viability product) or PoC (proof of concept).

Basically it’s a shitty first draft. Importantly though it’s something that is DONE.

Perfect is the enemy of shipped. And shipped is the only thing that matters. None of these were "production-ready scalable enterprise-grade deployments."

And here's what the gatekeepers are really scared of (even if they refute this vehemently): Vibe coding is dramatically better than it was 6 months ago.

Six months ago, vibe coding tools struggled with complex state management. Now, they handle it fine.

Six months ago, they couldn't do proper authentication. Now, it's a solved problem.

Six months ago, the code they produced was genuinely bad. Now, it's better than most junior developers. (And is having the effect on hiring junior devs that you would expect…)

The tools are evolving faster than the criticism. Yeah there are still issues. But guess what will happen to those known problems? They’ll get ironed out and eliminated.

The Progression Path

If you spend enough time on Twitter or Reddit you’ll be inundated with exhortations to use Claude Code or Codex or whatever the new hotness is. Remember that the people talking about this stuff (mainly) know what they are talking about and are more advanced.

These are NOT good starting points.

Let me demonstrate using Claude Codes’s official guide:

The good news is when you do this in a cafe like I just did you look like some L33t hax0r:

The bad news is if you’ve never used the command line or Terminal (think DOS but modern) then this is utterly incomprehensible.

So we’re not starting here. It’s a dumb place to start. Let’s be more reasonable.

Phase 1: The Gateway Drug (Lovable/Bolt)

First up we’ll use a pure vibe coding tool like Lovable, Bolt, Replit, v0, Base44…the list goes on.

I don’t care which you use. I’ve used them all and prefer Lovable but use whichever you like the look at. Spoiler: they are all Claude or ChatGPT under the surface anyway!

If you want to use Lovable use this link and you get an extra 10 credits. I also get 10 credits so that’s nice for me. 😝 

Start here. Not with Cursor. Not with Claude Code. Anyone telling you to start with those is like a gym bro telling you to deadlift 200kg on day one. Could you do it? Maybe. But why make life harder and risk injury? And yes…Claude Code can cause bodily harm.

Lovable and Bolt are your gateway drugs. They remove ALL friction between idea and reality. You literally say “hey I want to build such and such” and it builds it for you. Don’t overcomplicate.

Your first project will be embarrassingly simple. A calculator. A timer. Who cares - you're building. Your second will be something you actually need. Your third, something for your audience. Your fourth, something people might pay for.

I wrote a whole guide on getting started with Lovable which will help with the first steps. It’s applicable to the other tools too. That discusses what to build and helps you create a project plan to get up and running.

Along the way, you'll bump into things like Supabase (where your data lives) and GitHub (where your code lives if you want to back it up). You don’t need a full course on these things - you don’t need to understand the full workings. You'll use them because you need them. Click the button, connect the thing, move on.

This is incidental learning - the best kind. You learn what you need when you need it, not before. Just in time learning not just in case learning.

Instead we want to get addicted to shipping. Each project gives you a dopamine hit. You want more. You start seeing possibilities everywhere. Every problem you hit you’ll suddenly realise “oh I could build a tool for that!”.

When you find yourself thinking "I wish I could just change this one thing" but the tool won't let you - that frustration is good. It means you're ready for more control. That’s the next phase. You are ready to graduate.

Phase 2: Open the Hood (Cursor)

Cursor is a fantastic tool. I just don’t suggest starting with it if you’ve never coded.

It looks like this:

Not as scary as Claude Code and the command line (we’ll get there!) but definitely intimidating.

Cursor is your bridge tool. It's a real code editor but with AI deeply integrated. Now you can see the code. Edit it. Break it. Fix it.

Cursor gives a lot more power. You have more control. That’s a good thing and a bad thing. Lovable and similar tools protect you from yourself by hiding away the more complex elements of building. Basically to stop you mucking it up. Cursor opens up the engine and lets you tinker around. Amazing if you know what you are doing - problematic if you don’t!

I this phase you might try rebuilding something from Phase 1, but better. Or build something entirely new. The point is you're starting to understand what's happening under the hood.

What you're actually learning:

• File structure (where things live)

• Basic syntax (how things are written)

• Dependencies (what needs what)

• Deployment (taking it live)

You're not just studying these. You're encountering them as you build and learning just enough to proceed. I’ll talk about how we learn alongside in a moment.

Phase 3: The Big Boi (Claude Code)

Claude Code (and similar tools) is where you graduate. Full development environment. AI assistance when you want it. Full control when you need it.

You're no longer just describing apps. You're architecting them. Building something real for real users, with authentication, payments, a database - actual value. You can do this at the other levels but Claude Code gives you much more power.

I won’t go into much more detail because you are either i) using it already and don’t need me to reiterate or ii) so far from using it that anything I say will be gobbledegook. Just know for now that it’s up the chain and something you’ll want to look into later! Don’t feel stupid for not starting your journey here.

The Critical Reality Check

On this…you might not need to progress past Phase 1.

If Lovable builds what you need, why go further? You're an entrepreneur who can build, not a programmer. You hire programmers and engineers when you need them. You don’t need to spend years becoming a professional programmer. In fact doing so right now is a questionable endeavour considering what’s happening to the job market!

The ability to quickly prototype and validate ideas might be all you need. Going deeper is a choice, not a requirement. The point isn't to become a developer. It's to be dangerous enough to build what you need, validate your ideas, and speak the language when you do hire professionals. We use building to shape our offer and value, not to become the technical co-founder. Super important distinction!

Some entrepreneurs may stay in Phase 1 forever, using vibe coding to:

• Test ideas quickly

• Build MVPs for validation

• Create internal tools

• Shape their offer before hiring developers

ESPECIALLY as the tools themselves get better. Which they are doing.

Others go deep because they enjoy it or want full control. Both paths are valid. And don’t let any one tell you otherwise. Do what works for your building process not your codebase. Code is a tool to build with, nothing more.

Turning AI Into Your Personal Tutor

There’s a very dangerous trap we do need to avoid with vibe coding. It’s tempting to just give instructions to our tool and then accept everything it’s doing. When it’s running its processes maybe you pull out Tiktok and watch your favourite luscious-eyebrowed British AI creator. Who can blame you? You hear the ding that tells you Lovable has finished and move onto the next part of the build.

That’s risky. Real risky. Why?

First…the tools aren’t perfect. They’ll make mistakes.

But more importantly it’s robbing you of learning opportunities. Every time the tool builds, edits, publishes is a chance to learn something.

We can ask “What are you doing?”, “What is React??”, “Explain this piece of code”, “Why did you do it this way and not that way?” etc.

And the AI will (with infinite patience) explain everything.

But you need to put it in teaching mode. You need your AI that you are learning and don’t want it to just do everything.

Here's a prompt you can use inside your vibe coding tools to make it explain and teach:

You are operating in TEACHING MODE for someone learning to code through building.

Core principles:
1. Over-comment everything - explain WHY not just WHAT
2. After each code generation, provide:
   - Plain English explanation of what this code does
   - Why we structured it this way
   - One syntax pattern to notice
   - One common mistake to avoid

3. When I ask questions:
   - Validate that it's a good question
   - Answer at my current level
   - Show me 2 ways to do it (simple now vs better later)
   - Connect to the bigger picture

4. End each session with:
   - 3 key concepts we used today
   - 1 small experiment to try
   - 1 question to research

5. Progressive disclosure:
   - Start with working code
   - Explain current implementation
   - Suggest "what if we tried..." experiments
   - Point out patterns I'll see again

Remember: I'm building to learn, not learning to build. Keep momentum high, theory minimal. Explanations should be like you're explaining to a smart friend, not a computer science student.

Adjust this to taste. Because it WILL be annoying. When you are just trying to build something and it’s asking you questions and walking you through topics it’s frustrating! The key is to balance learning and building so that you don’t get frustrated.

How to activate teaching mode:

In Lovable: Settings → Project Settings → Custom Instructions → Paste the prompt

In Cursor: Settings → Features → Rules for AI → Add the prompt

In Claude Code: Set the Output style to “Explanatory” or “Learning”

And when it gets too much in one session - just remove the instructions! But remember to add them back in and not get lazy!

The Learning Stack That Actually Works

Forget Codecademy. Forget FreeCodeCamp. Forget Coursera.

Definitely forget the textbooks. Ugh.

Instead here’s a totally flipped script:

Primary Learning (80% of your time)

Build in public

Choose a project and get to work. THIs is the majority of your time. You get better at building by building. Making errors, working out what went wrong, fixing it and keeping on moving.

Post your progress on Twitter or whatever your preferred platform is… "Day 3 of building my invoice app. Added PDF export. Broke everything twice. Fixed it three times."

This creates accountability. It documents your journey. It attracts help when you're stuck. It builds an audience for what you're building. This will be useful when you are going to launch.

Check YoniMan for a masterclass in this approach. This is also what inspired me to live stream my building sessions on Youtube.

Your errors are your curriculum

Error message → Copy into Claude/ChatGPT→ Ask for an explanation and solution → Try solution → Understand why it worked

This is how real developers learn. They used to use Google and Stackoverflow. Now we can do this with AI.

Secondary Learning (20% of your time)

Fireship on Youtube

Fireship on YouTube explains any concept in 100 seconds. Perfect for entrepreneurs.

Need to understand APIs? 100 seconds. Databases? 100 seconds. Authentication? 100 seconds.

Here’s the channel https://www.youtube.com/c/fireship. Only watch ones that are relevant to the topic at hand. Otherwise it’s noise.

The NotebookLM Hack

For LONGER format learning don’t just passively watch videos.

Instead use Notebook LM (free) as a study partner.

Here's how:

1. Find a good YouTube tutorial on a concept you just encountered

2. Plug the Youtube url into NotebookLM

3. Generate:

   • Study guide

   • FAQ about the topic

   • Audio podcast discussing it

   • Practice questions

NotebookLM turns passive watching into active learning.

For example if you plug in this video about Cursor you get this in NotebookLM:

You can now “chat to” the video, create flashcards, generate a quiz and much more. It takes a few moments to generate everything and your retention will 10x

Your Action This Week

1. Go to Lovable (use this link for +10 credits)

2. Build something

I had some more points but realised they are unnecessary. Just got here and type something in to get started.

That's it. Don't read another tutorial. Don't watch another video. Don't buy another course.

Do this and you’ll have done more than 90% of people "learning to code."

And shipping beats studying every single time.

Keep Prompting,

Kyle