I built a full-stack product in a week. Here's what I'm still thinking about.

Let me back up. I know some basic coding — mostly CSS and styling-related things. I've never built anything close to a production application. No databases, no APIs, no deployment pipelines. That world was always on the other side of the handoff.

What I do have is years of designing products. Thinking in systems. Structuring information. Making decisions about what to build and what to cut. It's the kind of thinking that becomes second nature after a while — you stop noticing you're doing it.

Turns out, that thinking is exactly what building with AI requires.

What's actually under the hood

I want to be specific here, because "I built a product with AI" can mean a lot of things. Here's what SkillHub actually is:

Next.js 15 with TypeScript. A PostgreSQL database on Supabase with Row Level Security. shadcn/ui components styled with Tailwind. Framer Motion for page transitions and scroll animations. A WebGL hero built with Three.js and custom shaders. Zod validation on forms. Email notifications through Resend — both to me as admin and as confirmation to submitters. An admin dashboard with magic link authentication. Three automated cron jobs running on Vercel: one updates GitHub star counts daily, one discovers new skills weekly by scanning GitHub repos and npm registries, and one enriches pending submissions with metadata pulled from READMEs. Privacy-friendly analytics via Plausible. Dynamic Open Graph images per skill page. 83+ skills across 8 categories.

That's 11 tools in the stack. When I look at that list now, it feels almost absurd that it came together in a week.

But the part I'm most proud of is that the platform sustains itself. Every week, a cron job automatically scans GitHub and npm for new Claude skills. When it finds them, they appear in my admin dashboard, ready to review and approve with one click. The directory grows without me having to manually hunt for new entries. I just curate.

The part that surprised me

Here's what I keep coming back to: none of the technical work was the hard part. Not the code, not the infrastructure. Configuring DNS records for email delivery, understanding four different Supabase client configurations, getting middleware to refresh auth sessions, debugging cron job timeouts — Claude helped me through all of it. Every problem had a solution if I framed it clearly enough.

The product decisions came easy too. What pages to build. How to structure the data. Where to invest in polish and where to ship rough. I've been making those calls for years — just in Figma instead of code.

The hard part is what comes next. The product is live. It works. And now I need people to actually find it.

That's the frontier I'm standing at now. Building the thing turned out to be the more straightforward challenge. Getting attention for it — that's a different skill set entirely. And it's one I'm still figuring out.

What I'm still figuring out

I used Claude as a development partner throughout the entire build. Not as a tool I copied from blindly, but as someone I could think with. And the thing I keep reflecting on is how much the quality of what came back depended on how well I framed the problem.

Give it vague input, get vague output. Give it clear constraints and specific context — the way you'd write a good design brief — and the results were remarkably good.

There's something interesting in that. The skills that made me effective weren't technical. They were taste — knowing what feels right and what feels off. High agency — not waiting for permission or the perfect conditions, just starting. And the same design thinking I use in every project: break the problem down, define the constraints, know what you're solving for, communicate clearly. Systems thinking. Information architecture. Scoping. Prioritisation.

I don't think this means designers should become developers. But I do think the gap between having a vision for a product and actually shipping it has gotten much smaller than most of us realise. And if you've spent years thinking in systems, structuring information, and making decisions under constraints — you might be better equipped for this new workflow than you think.

The product

SkillHub is live at skillhub.pm. It's free. You can browse skills, search by category, and submit your own if you've built one. I also wrote a complete build guide documenting the full stack, setup, and code patterns.

It's version one. It's not perfect. But it exists because I stopped waiting for the right moment and just started.