The Promise vs The Reality
Every CEO has heard the pitch: AI can write code now. Just describe what you want, and the computer does the rest. No more expensive developers. No more technical debt. Just you, a chat interface, and infinite possibilities.
I bought into this vision. As someone who built AI systems at Marks & Spencer in 1989, I figured I had the technical chops to make it work. I started writing games to prove I could still code. That simple experiment turned into a proper education in what's now called "vibe coding" – instructing large language models to write software through natural language.
What I discovered surprised me. It's not that the promise is false – it's that the reality is far more nuanced than the hype suggests.
My Journey Through the Vibe Coding Landscape
The Firebase Beginning
I started with Google's Firebase suite. The appeal was immediate: I could write code, deploy databases, and handle architecture all through conversation. For someone who last coded seriously decades ago, this felt like magic.
It worked. Until it didn't, the system would make simplistic mistakes – the kind a junior developer makes on day one. Worse, it would make the same mistake repeatedly, no matter how I rephrased my instructions. I also hit cost overruns twice, forgetting to set spending limits and ending up with eye-watering bills for minimal work. That felt like a Google problem, not mine.
The Search for Alternatives
I tried Cursor next. Brilliant tool, but it's built for people who already know how to code. If you're working in a complete development environment like Firebase Studio, it doesn't quite fit.
Then Claude Code, which came highly recommended by fellow entrepreneurs. Solid performance, but I kept searching.
The Famous.ai Disaster
I rebuilt the entire game in Famous.ai. The development environment was genuinely enjoyable. The charging model was absolutely ridiculous.
Here's what happened: I'd give an explicit instruction. The system would fail to implement it correctly. Then it would charge me for the development time anyway. The bills became astronomical for work that was half-done or wrong.
My conclusion: you'd be better off posting the job on Elance and hiring a human developer. Full stop.
Back to Firebase with Gemini 3
When Google launched Gemini 3, I returned to Firebase for a new version of the game. This time, it's been highly successful.
Why? Two possible reasons. First, Gemini 3 might be better at coding within the Firebase suite. Second, and more likely, I've learned how to speak the language of these systems. I know how to write correct logic now. I understand what works and what doesn't.
There's also a third factor: I now use another large language model to help me write better briefs for the vibe coding system. This has massively reduced my cycle time.
The game is live at Britgrid.com if you want to see what vibe coding can actually produce.
What I Learned About Vibe Coding
It's Not Push-Button Simple
The seductive idea is that you chat naturally, and code appears. That's partially true, but misleading.
You don't need to know PHP, JSON, or React. But you absolutely do need to learn how to give instructions properly. You need to understand logical flow. You need to be precise about requirements. You need to spot when the system is about to go down the wrong path.
This is a skill. It takes time to develop.
Cost Control Is Critical
I learned this the expensive way. Twice.
Set spending limits before you start. Monitor usage obsessively in the early days. Different platforms have wildly different charging models – some charge for successful output, others charge for compute time regardless of quality.
Famous.ai taught me that the most sophisticated interface can hide the most predatory pricing.
Platform Choice Matters Enormously
Not all vibe coding environments are equal. Some are genuine productivity multipliers. Others are expensive science experiments.
Firebase works for me because it handles the full stack – code, database, hosting, and deployment. For quickly building and maintaining live applications, that integration is invaluable.
Your mileage will vary based on what you're building and how you think.
What This Means for CEOs
Don't Dismiss Your Development Teams
The biggest mistake I could have made was walking into the office and announcing, "AI writes code now, so we'll do everything that way."
What I've learned from falling flat on my face is that vibe coding is a genuine skill. Your developers need to know it. They need time to experiment. They need a budget to fail safely.
The people who will be best at vibe coding are the people who already understand software architecture, user experience, and system design. In other words: your existing development team.
It's Not as Cheap as You Think
Yes, vibe coding can be faster than traditional development for specific tasks. But the cost savings aren't automatic.
You're trading developer time for compute time plus learning time. You're trading predictable hourly rates for unpredictable usage-based pricing. You're trading known quality thresholds for quality that varies with how well you prompt.
Budget accordingly. This is an investment in capability, not a cost-cutting exercise.
The Skills Gap Is Real but Different
Your team doesn't need to learn new programming languages. They need to learn how to be extremely precise in natural language. They need to understand how to break complex requirements into achievable chunks. They need to recognise when the AI is going wrong before too much time is wasted.
These are coachable skills, but they take practice. Give your people space to develop themselves.
What I'd Do Differently
If I were starting this journey again, I'd make three changes:
Start with guardrails. Set spending limits, time limits, and scope limits before the first line of code. Test the platform on a small, disposable project first.
Invest in learning. Use another AI to help write better prompts. Study what works. Build a library of successful instruction patterns. This compounds quickly.
Choose platforms carefully. Optimise for integration and predictable costs, not for the flashiest interface. The tool that makes you feel clever might be the one that bankrupts your experiment.
The Bottom Line
Vibe coding is real. It works. I've built a functioning game using it, and I'm someone whose serious coding days ended when the Berlin Wall fell.
But it's not magic, and it's not free. It's a new skill that requires investment, practice, and disciplined thinking.
As a CEO, your job isn't to replace your developers with chatbots. Your job is to help your team add this capability to their toolkit, with realistic expectations about costs, timelines, and the learning curve.
The companies that will win with AI-assisted development are the ones that treat it as a skill to master, not a shortcut to avoid. That means giving your people time, budget, and permission to fail safely while they learn.
I fell flat on my face multiple times on the way here. That's the tuition fee for this education. The question for you is whether you'll learn from my expensive mistakes or make your own.
The game built through this vibe coding journey is live at Britgrid.com – feedback welcome.
Submit a comment