TL;DR How I went from doom-scrolling to vibe coding.
Vibe Coding to the Max đ
I didnât start vibe coding because I had a genius idea. I started because I was procrastinating. Scrolling! Watching people ship websites, Apps, Micro tools and more. Seeing those Lovable ads pop up every five minutes. And then I had this moment:
âWhat is actually stopping me from doing this?â
âWhy canât I just join the crowd?â
So I did. Because the only thing worse than failing⊠is wondering what wouldâve happened if you tried.
The Spark đ„
It started small. I worked with my cousin on my friendâs business website â PureOrigins.uk. He was more advanced with AI tooling at the time. He kicked it off. I picked it up midway and finished it.
That was it! The spark that started this journey!
If I could do that⊠I could do more.
Then I watched another friend engineer his site forefinder.golf purely through vibe coding. Watching him iterate fast, ship quickly, break things, fix things â it was addictive.
And when I say addictive, I mean addictive.
There were nights I shouldâve been asleep but instead I was pushing another increment live. The only real battle now? Managing my ADHD and not juggling 12 projects at once đ
Look To Start Small đ€
My friends have asked me:
âWhy donât you just build one big âseriousâ thing?â
Simple. I wanted to feel what building with AI was actually like. I studied Software Engineering. I worked in DevOps. But over time I drifted more into delivery and programme leadership. Vibe coding has been my way back into building â seeing code again, understanding structure again and once more breaking problems down. Itâs been like reconnecting with a muscle I hadnât flexed properly in years.
So for now I am working primarily on:
- Personal tools
- Experiments
- Friendsâ projects
- Potential business ideas
- Things that started as shower thoughts đż
I find one of the biggest, most rewarding surprises, has been building my own website. I had been putting it off for years. I couldâve used WordPress or a template, but building it increment by increment just hit differently. More control! More ownership!! More reward!!!
The Biggest Lesson: AI Is Not a One-Shot Machine đ§
Early on, I thought if I wrote a comprehensive enough requirements document, AI would build the whole thing in one go. Right?
Wrong!
Even with detailed instructions, thoguh personal experimentation, I find it falls short if you try to do too much at once. Thatâs when something clicked. This isnât magic. Itâs Agile.
Build small. Ship fast. Iterate.
The moment Codex started losing sight of the narrative, ignoring design intent and drifting from the vision, I realised dumping everything into one mega prompt wasnât the way. You have to guide it, just like you would guide a team.
Shipping Small But Fast âĄ
For me, this looks like:
- Build the skeleton.
- Break functionality into parts.
- Deliver one working slice at a time.
- Test behaviour and styling.
- Merge only when it behaves properly.
I even built myself a lightweight starter repo template, something I can copy into any project to standardise how AI behaves as a small delivery team. It frames AI like a Product Owner, Scrum Master and Dev Team; But all controlled by one person.
That structure alone improved the quality of what I was building.
AI Isnât Just a Tool. Itâs an Extension đ€
At some point I stopped treating AI like autocomplete. It became an extension of me.
Does having a dev background help? Absolutely. Is it required? No. But thinking like a Product Owner helps massively. If you can break work down, clarify outcomes, guide direction and refine incrementally, youâll get far better results.
AI builds. You steer.
Thatâs the game!
Futher pitfalls⊠Agents: Theyâre Powerful⊠But Can Often Over-Engineer đ§©
Agents are incredibly useful when youâre stuck on a big hurdle, need structured reasoning, or want gaps identified in your thinking. Theyâre great at decomposing complexity.
But they can over-engineer small problems. Sometimes the solution is simple. Not every task needs a swarm of agents.
What Frustrates Me đ
Letâs be honest. Depending on what model you are using, what prompting you have provided, AI does not one-shot beautiful UI. It doesnât naturally nail UX without direction. At least that has what I have observed. Evidently it is an area which developments are being made. Especially if we were to look at what is coming out of China with Kimi-K2.
If you want something polished, youâll spend time refining prompts, adjusting structure, fixing layout quirks and re-aligning to your design intent. That iteration loop can be exhausting.
But thatâs also where the craft comes in. A new muscle you develop along this journey. The AI prompting muscle. Something I feel will strongly be our future. Learn how to correctly prompt with AI now, will pay dividends for all in the future.
What I Learned (Accidentally)
Just download the tools.
VSCode. Codex. Copilot. Claude.
Stop overthinking it and start small.
Also join a Discord community or start with your friends. Some of my best learning came from conversations about whatâs actually achievable, better strategies for building apps and even discovering local LLM setups like Ollama for projects where buying tokens didnât make sense. Things which only came up in conversations with my friends. Things which helped grow my knowledge and shape the way I attack projects
You donât need the most expensive stack. You need the right stack for the job.
Has This Changed Me?
Yes.
I find it has made me more rounded. Itâs brought back the builder in me. Itâs reminded me how to start small and structure an SDLC that fits the scale of the project, not some enterprise theatre version of it.
Iâm learning more. And Iâm learning faster.
If Youâre Thinking About Starting đ
Start small!
Understand what AI can do. Understand what it cannot do. Yes, you can build a full product. But there will be hurdles: UI refinement, architecture clarity, scope drift and your own discipline.
Treat it like a team member. Not a vending machine.
And if you get stuck⊠Well Iâd say Google it but maybe, form now on, Iâll start saying GPT it đ€Ł Jokes aside, there are massive communities out there which can help you explore and navigate this new world. YouTube videos which really help to break things down nicely. And like I said before, levarage your friends. If they havenât started already, started this journey together. Joint learning is a very powerful tool in and of itself.
The Truth
I believe vibe coding isnât about AI replacing developers. Itâs about building better solutions. Itâs about bringing shower thoughts to life!
For the first time, the gap between âI have an ideaâ and âit existsâ is tiny. And once you realise that, you stop waiting. You stop over-planning. You stop saying âone day.â You just build!
Messy. Incremental. Imperfect. But live.
AI hasnât made me a developer again. Itâs made me a builder again. And thereâs a difference. âm shipping more. Learning faster. Thinking bigger.
And honestly? Iâm only just getting started đ