What Cursor and I Built in One Night: The SEO Tool That Had My Colleague Fooled

Hans Brattberg
Co-founder, Product & AI Strategy
·4 min read
What Cursor and I Built in One Night: The SEO Tool That Had My Colleague Fooled

In this article

I was supposed to read a book, but got frustrated with SEO tools and accidentally built my own with AI. When I showed it at work, my colleague thought it was a commercial SaaS product. It wasn't — it was just my evening.

One evening after work — that window when your brain is still buzzing but your body has checked out — I sat in my armchair to read a book. But my mind kept spinning. I should fix up the website. Better SEO, an AI tool to help out. How hard can it be?

So I opened Perplexity and went looking for an SEO tool that uses AI in some clever way.

SEO Tools: Gateway to Hell

What greeted me was... a smorgasbord of professional monster tools. Every single one looked built for people who eat, sleep, and breathe SEMrush tables. I wanted a quick look at what was showing up on Google. But every option came with a learning threshold that felt like an entire course. Guides, dashboards, API keys, onboarding flows, billing tiers — the works.

Then I tried to make sense of Google Ads Keyword Planner.

That's when I wanted to scream. How can this tool even exist? Someone designed it to keep consultants employed. Keywords, campaigns, groups, settings — a labyrinth. I clicked around and gave up.

How Hard Can It Be to Build Something Yourself?

I opened Cursor. My favourite IDE these days. It has several ways to work with AI. I run it in Agent mode — where you chat with an assistant that understands your entire codebase and, frankly, sometimes catches things I'd have missed. The beauty of Agent mode is that you don't micromanage every step. You describe what you want to achieve. The AI works toward that goal on its own — through all the tedious steps you'd otherwise slog through yourself.

I typed: "I want this deployed on Vercel. With a database. Automatic deploy via GitHub. What's the easiest way?"

And then it was off. Next.js. PostgreSQL. Docker. A GitHub Actions pipeline. It suggested, I nodded, it installed. I sat in my armchair surrounded by flowers in my bay window, looking out over Stockholm. It was late, and I was only supposed to be reading. Instead, I fell into two hours of vibe coding.

Git: Now With Human Dignity

Let's pause for a second to talk about Git. Git is like programming with your elbows. Every command feels like a riddle from a hostile mentor. It's not built for users — it's built for anyone who already understands how it's implemented. I always liked Mercurial more, but nobody uses that anymore.

But here's the beautiful part: in Cursor, you don't need to understand Git anymore. You type: "How do I push this?" And the AI handles it. Git, for the first time in my life, became usable. Hallelujah.

Early Morning Tea and an AI That Builds UI

I woke up around five. Silence. I made a cup of tea, settled back into the same armchair, and kept going. Now I started iterating. Every tweak took minutes. When I wanted to change the interface, I took a screenshot, dropped it into Cursor, and wrote: "This is what it looks like now, but I'd like it to look like this instead." And it did. Just like that.

By this point I started thinking: this shouldn't be this easy. But it was. And I didn't quite trust it.

What Cursor and I Built in One Night: The SEO Tool That Had My Colleague Fooled

Henrik and the Off-the-Shelf Product

A few hours later, back at work, I posted a screenshot in our Slack. Henrik, a long-time colleague, looked at it and asked: "What does the table mean?" Not gonna lie — that stung a little. I stayed up all night building this, and that's the question?

Turns out he thought it was a commercial product. It wasn't critical — a straightforward question about how it worked. So my overnight hack had passed for an actual off-the-shelf product.

So What Actually Happened?

I went from total confusion in the SEO jungle to a working prototype in under five hours. All it took was the right AI, the right setup, and the stubbornness to not go to bed. Cursor got me past dead ends, generated code I hadn't asked for but needed, and produced a UI I'd have spent half a day trying to style myself.

And Git? My old enemy? We're not at war anymore. We may not be on great terms, but at least we're talking. And that... is a miracle in itself.

Read more

How to not AI Slop

How to not AI Slop

Illustrated by two extremes: an unedited voice transcript of Henriks sitting on a rock talking, vs an AI-written article

Henrik Kniberg
May 9, 2026
The Trust Ladder: how to onboard AI agents like new colleagues

The Trust Ladder: how to onboard AI agents like new colleagues

What data should we let them access? What if the agent gets it wrong? What holds organisations back is rarely the technology — it's the uncertainty. Think of AI agents as new colleagues and extend their remit one rung at a time, along the Trust Ladder.

Hans Brattberg
April 24, 2026
The First 100 Days as an AI Lead: The Playbook in 5 Minutes

The First 100 Days as an AI Lead: The Playbook in 5 Minutes

You got the role. The mandate is clear: make AI agents work across the organization. Now what? A short teaser of the four conditions, three phases, and the patterns that separate AI initiatives that ship from the ones that stall — with a link to the full 100-day playbook.

Nils Janse
April 24, 2026
Demo: The Human + AI-Agent Dev Team

Demo: The Human + AI-Agent Dev Team

How we use AI agents internally at Abundly - both for coding, backlog management and release management. It's a system of AI agents & human engineers working together, building on each other's strengths. This not only enables us to release a new version of our platform every day, but also makes the work really fun.

Henrik Kniberg
April 23, 2026
Webinar slides: AI Powered Software Development from the Trenches

Webinar slides: AI Powered Software Development from the Trenches

Slides and recording from Henrik's webinar "AI Powered Software Development from the Trenches.

Henrik Kniberg
April 10, 2026
One File to Rule Them All — A Lesson in AI Agent Unsafety

One File to Rule Them All — A Lesson in AI Agent Unsafety

2.3 million AI agents on a social network, most running on people's personal computers with full access to files and credentials, reading and obeying instructions from a single public web page. What can possibly go wrong?

Henrik Kniberg
February 10, 2026