TL;DR — what I want you to know
- You don't need to know how to code. I didn't write a single line.
- The speed is genuinely insane. Hours, not months.
- It will break. That's where the learning actually happens.
- Every PM, designer, or non-technical person has an edge here — you just don't know it yet.
Why I started
I've worked in product and tech my whole career. I've written specs, run sprints, and sat in a hundred engineering standups. But I never learned to code. Not properly. A bit of SQL here, some Notion formulas there — but not build things coding.
That started to bother me more as AI got better. I kept hearing "the barrier to building is gone" and I kept thinking: is it though? So I decided to test it for myself. No bootcamp, no tutorials — just me, Claude Code, and a blank folder.
The first thing I built was this website. Then a job application tracker on Replit. Then an AI daily brief agent. Then experiments with Suno-generated music. Each one taught me something the last one didn't.
What I actually built
Here's the honest timeline — roughly in order of complexity:
This website — agilefamily.com.au
A full Next.js site with EN and ZH versions, custom design, responsive layout, modal components, portfolio section, and a quiz. Built and deployed on Vercel. Iterated over multiple sessions. Took a few hours from nothing to live. This article you're reading right now was also added this way.
Smart Application Tracker (Replit)
A job application tracker with a real database, status columns, and filtering. Built entirely on Replit. No local setup — I just described what I wanted and iterated. The whole thing went from idea to working app in under 2 hours.
AI Daily Brief Agent
An automation that pulls together news, my calendar, and pending tasks — summarised and sent to my inbox every morning. Built in 30 minutes. Runs every day. I haven't touched it since.
AI Music with Suno
Not exactly vibe coding, but vibe creating. I wrote prompts and Suno generated full original songs. Spent an afternoon exploring where generative audio is at. Surprisingly good.
The tool I used: Claude Code
I tried three different tools when building this site. Here's the honest comparison:
| Tool | What it's good at | Where it fell short | My rating |
|---|---|---|---|
| Antigravity | Fast visual prototyping, drag-and-drop | Limited control, hard to customise beyond templates | |
| CodeX | Great for line-by-line help if you know code | Assumes you already know what you're doing | |
| Claude Code ✓ | Full file management, understands context across the whole project, CLI-based so very fast | Setup takes 20 min if you've never used a terminal |
Claude Code won for me because it could hold the whole project in its head. It didn't just autocomplete a line — it understood that I had a ZH version and an EN version and that changing one thing needed to be reflected in both. That's not something you get from a simple AI chat assistant.
The best way I can describe Claude Code: it's like pair programming with someone who reads everything you've written, never forgets it, and types 10x faster than you.
What nobody tells you about vibe coding
The hype makes it sound like you just describe something and it appears. That's mostly true — but here's what gets left out:
It breaks. Constantly.
I hit layout bugs, TypeScript errors, hydration mismatches, components that worked in dev but not in production. Every single project had at least one moment where I stared at the screen thinking "I have no idea what's wrong here."
But here's the thing — I described the error to Claude Code and it fixed it. Every time. The debugging loop is: see error → paste error → watch it get fixed → move on. It's not frustrating. It's actually fast.
Context matters more than code
The times I got the best results were when I was specific. Not "make it look better" but "the hero section needs more white space above the heading on mobile, currently it's 16px and it should be 40px." The more precisely I described what I wanted, the better the output.
That's a writing and thinking skill, not a coding skill. And it's one I've been developing for years without knowing it would be this useful.
You still have to make decisions
AI doesn't decide what to build. It doesn't know whether the contact form should go above or below the portfolio section. It doesn't know that the Chinese version of the site needs an extra featured card for Xiaohongshu. Those are judgment calls. And they still require you.
The thing that surprised me most: my PM skills made me better at this
This is the part I didn't expect.
I kept assuming that "not knowing how to code" was my disadvantage. What I discovered is that everything else I knew from years of product work became an advantage:
Mobile-first thinking
I knew to check the site on a phone immediately. I caught a broken layout on mobile that would have embarrassed me if I'd shipped it. Developers sometimes forget to check — I never do, because I write mobile QA into every spec.
Bug triage and reproduction steps
When something broke, I knew how to write a clear bug report: what I expected, what happened instead, what I'd already tried. That's exactly how you talk to Claude Code effectively. Clear bug reports → faster fixes.
Prioritisation
I had a long list of things I wanted to add. But I also knew which ones mattered for launch. I shipped the core site first, then iterated. Most first-time builders try to build everything at once and get stuck.
Accessibility and contrast
I knew to ask: "Does this text meet WCAG contrast standards?" I knew to add ARIA labels. I knew the colour #A3A3A3 on white is borderline. These aren't coding questions — they're product quality questions.
Knowing when to stop
There's always one more feature to add. I knew to draw the line, ship the thing, and let reality tell me what actually needed fixing. Most things I was worried about turned out not to matter. One bug I'd ignored turned out to be the first thing someone noticed.
The gap isn't between people who can code and people who can't. It's between people who can think clearly about what to build — and people who can't. AI has made the second group irrelevant. The first group just got a superpower.
What I'd tell someone starting today
If you're a PM, designer, consultant, or anyone who thinks "I'm not technical enough to build things with AI" — I want to push back on that.
Start with something real. Don't build a tutorial app. Build the thing you actually want to exist. A tracker for something you do manually. A page that describes your work. A tool that saves you 30 minutes a week. Real problems give you real motivation to push through when it breaks.
Use Claude Code, not just ChatGPT. ChatGPT in the browser is great for questions. But for actually building, you need something that manages files, runs commands, and understands your whole project. Claude Code in the terminal does that. It takes 20 minutes to set up. Worth it.
Expect it to break — and be okay with that. The first time you get a red error screen, your instinct will be panic. Resist. Paste the error. Ask what's wrong. 90% of the time it'll fix itself in one message.
Bring your own expertise. Whatever you know from your field — write it into your prompts. "Make the mobile nav accessible" is better than "make the nav". "The CTA should feel urgent but not pushy" is better than "make the CTA good". Your taste and judgment are the inputs. The AI handles the implementation.
Ready to start building?
I run workshops on exactly this — from complete beginners to people who want to build real agents. Come try it.
See upcoming workshops