From GPT to Claude: A Vancouver Developer's Journey Through the Wild West of Vibe Coding

By Chris – Building AI products between Uber rides

---

Six months ago, I thought I'd found the holy grail. GPT was going to change everything – hell, it was life-changing. But here's the thing about revolutionary moments: sometimes you don't realize how clunky your revolution was until you find something better. And sometimes you don't realize how dangerous your new toy is until it builds you a house of cards.

Enter Anthropic (/an-THROP-ik/), the company founded by former OpenAI employees, and Claude (/klohd/), their flagship model. Working with Claude made me realize that my early GPT days were like trying to code on a 1990s Nokia compared to an iPhone – except the iPhone might also set your garage on fire if you don't read the manual.

I'm not here to completely trash GPT – it still owns the mainstream narrative and has broader tooling support. But mostly I'm here to talk about something that's consumed my life for what feels like decades (even though it's only been six months): vibe coding.

What the Hell is Vibe Coding? (And Why It's Not Magic)

If you haven't discovered vibe coding yet, drop everything and get into it – but keep your critical thinking helmet on. The term was popularized by Andrej Karpathy, who described it as "fully leaning into the vibes, letting the AI generate code, and just taking it for granted." It sounds liberating. It is liberating. But it's also fragile.

Here's the beautiful paradox (/PAIR-uh-doks/): you don't need to know anything about code to start vibe coding. What you need is belief in yourself and zero fear. That's literally it. No computer science degree, no bootcamp certificates, no Stack Overflow reputation points. Just vibes.

Think about it – we're living through something that would have seemed like magic to developers just five years ago. Remember when coding meant memorizing syntax (/SIN-taks/) and debugging (/dee-BUG-ing/) for hours over a missing semicolon? Now I'm having conversations with AI that can infer what I mean even when I don't know what I mean myself.

I can't promise vibe coding will make you rich, but I guarantee you'll have more fun than you've had since you discovered you could order pizza at 2 AM. Just don't expect the pizza to always arrive hot.

The Open Source Revolution (For People Who Don't Code)

Let me paint you a picture that probably sounds insane if you're not in this world: I can run a command that pulls code from some developer's server – maybe they're in Estonia, maybe they're in their mom's basement in Ohio – and boom, that code downloads to my machine. Then I can use it to build something that could theoretically make money immediately if I have distribution.

This is the modern miracle of open source (/OH-pen sohrs/). It's like having access to every tool that's ever been invented, sitting in your digital garage, ready to be combined in ways their original creators never imagined.

The democratization is real. A designer with no backend experience can spin up a full-stack application. A marketing person can build their own analytics dashboard. A small business owner can create custom automation tools without hiring a development team.

The catch? Building is the easy part. Distribution – getting your creation in front of people who actually want to use it – that's the real challenge. The internet is littered with brilliant projects that nobody knows exist.

Why Claude Changed Everything (For Me)

Here's where things get personal. GPT felt like having a really smart intern who sometimes gave you brilliant insights and sometimes confidently told you that Paris was the capital of Italy. You learned to work around its quirks, but you never fully trusted it.

Claude feels different. More... reliable? It's like the difference between a friend who's always game for whatever crazy scheme you're cooking up (GPT) and a friend who's game for your crazy schemes but also asks thoughtful questions that make you realize maybe you should wear a helmet (Claude).

The technical difference is in the training methodology. Anthropic uses Constitutional AI, which trains the model to be helpful, harmless, and honest through a process of AI feedback rather than just human feedback. In practice, this means Claude is more likely to say "I don't know" when it doesn't know, rather than confidently hallucinating an answer.

When I'm vibe coding with Claude, I spend less time debugging AI-generated nonsense and more time actually building. The code it produces tends to be more maintainable, better documented, and follows actual best practices rather than just "practices that compile."

The Dark Side of the Vibe

But let's talk about what the hype machine doesn't want you to hear. Vibe coding has created a generation of "developers" who can ship products but can't maintain them. Who can build features but can't optimize them. Who can create but can't troubleshoot when things inevitably break.

I've seen brilliant non-technical founders build impressive MVPs through pure vibe coding, then hit a wall when they need to scale, optimize, or integrate with complex systems. The AI can't explain why your app slows to a crawl when you hit 1000 users, or why your database locks up under load.

This isn't meant to scare you away – it's meant to keep you grounded. Vibe coding is a superpower, but like any superpower, it comes with responsibility. The responsibility to learn, to understand, and to know when you're approaching the limits of your knowledge.

The Vancouver Advantage

Living in Vancouver has given me a unique perspective on this AI revolution. We're close enough to Silicon Valley to feel the pulse of innovation, but far enough away to avoid getting completely swept up in the hype cycles. Plus, the exchange rate means my OpenAI and Anthropic bills hurt a little more, which keeps me honest about which AI interactions are actually worth paying for.

The tech scene here is hungry but not desperate. Optimistic but not delusional. It's the perfect environment for experimenting with vibe coding while keeping both feet on the ground.

Building Between Rides

Here's the thing about building AI products between Uber rides (yes, that's literally my life): constraints breed creativity. When you have 15 minutes between dropping someone off in Burnaby and picking someone up in Richmond, you learn to make every prompt count. You develop an intuition for what will work and what will send you down a rabbit hole.

This fragmented building process has actually made me a better vibe coder. I can't afford to pursue every interesting tangent the AI suggests. I have to stay focused on the core problem, iterate quickly, and ship incrementally.

Some of my best ideas have come while stuck in bridge traffic, frantically typing prompts into Claude on my phone, hoping to capture an insight before my next passenger gets in the car.

What's Next?

The vibe coding revolution is just getting started. We're seeing the emergence of AI-native development workflows, where the entire process – from ideation to deployment – assumes AI collaboration. GitHub Copilot was just the beginning. Tools like Cursor, Replit Agent, and Claude's new Computer Use capabilities are pushing the boundaries of what's possible.

But the real revolution isn't in the tools – it's in the democratization of creation. We're moving toward a world where the limiting factor isn't technical skill, but imagination and distribution. Where the question isn't "can this be built?" but "should this be built?" and "who will use it?"

The developers who thrive in this new world won't be the ones who can memorize the most frameworks or optimize the tightest algorithms. They'll be the ones who can bridge the gap between human problems and AI solutions, who can vibe code with confidence while staying humble about what they don't know.

Conclusion: Embrace the Vibe, Respect the Craft

After six months of living in the vibe coding world, bouncing between GPT and Claude, building products between Uber rides in Vancouver traffic, here's what I've learned: this technology is simultaneously more powerful and more fragile than the hype suggests.

Vibe coding will democratize software creation in ways we're only beginning to understand. It will enable people with great ideas but limited technical skills to build real products that solve real problems. It will compress the timeline from concept to MVP from months to days, from days to hours.

But it won't eliminate the need for deep technical understanding – it will make that understanding more valuable, not less. The vibe coders who succeed long-term will be the ones who use AI as a springboard to learn, not a crutch to avoid learning.

So dive in. Start vibe coding. Build ridiculous things. Break stuff. Ship products that probably shouldn't work but somehow do. But stay curious about why they work. Stay humble about what you don't know. And remember that behind every magical AI interaction is a complex system built by humans who very much had to understand their code.

The future belongs to the vibe coders who never stop learning. The question is: will that include you?

---

Chris builds AI products and drives for Uber in Vancouver. You can find his latest projects and rants about vibe coding on his blog, assuming he finds time to update it between rides. Follow along for more adventures in the wild west of AI-assisted development.