For about a year, I watched Cursor from a distance.

I'd see the demos on Twitter. Videos of a sentence in English becoming three files of working code. The Cursor account shipping something I couldn't ignore every few weeks. I'd nod. I'd tell myself it was cool. And I'd keep my editor exactly the way it was. Copilot still running in the background, obviously. I told myself that was different.

I was afraid. Not of the tool. I was afraid of what I'd become on the other side of installing it.

I've been making things on a computer since 2013, the year I got my first PC. It didn't start with code.

When I was a kid, I wanted to be a 3D artist. I'd spend nights inside Cinema 4D, following YouTube tutorials step by step, making little intro animations that nobody needed and nobody watched. I made posters in Illustrator. I stacked effects in After Effects until my laptop fan sounded like a takeoff. I learned Photoshop the way kids learn video games: by poking every menu until something broke.

None of it was useful. All of it was the point.

At some stage, the curiosity leaked into code. I started with VB.NET. Forums, YouTube, copy-paste. I'd download tiny freeware tools, open them in a disassembler, and try to see how they worked. I read hacking threads late into the night, not because I wanted to hurt anything, but because I wanted to understand. How does this actually run? What happens if I change this byte?

I was just a kid who couldn't stop poking. The internet was enormous, and the source of almost everything was readable if you tried hard enough.

A few years later, diplomas, first jobs, production codebases.

Web development is the thing that stuck. The reason is simple. With a laptop and a browser, one person can ship to any user on any platform in the world. No factory. No publisher. No app store, most of the time. You type, you push, it exists.

For someone whose whole childhood was about making things for the sake of making them, that's the best deal I've ever been offered.

The first AI tools didn't worry me.

GitHub Copilot and Tabnine were neat. You type the first two words of a function, the editor guesses the rest. Sometimes it was right, often it was wrong, mostly it saved a couple of seconds. Fancy autocomplete. I used it the way I used snippets.

ChatGPT showed up and for a long while it felt the same. No web access, stuck on its training cutoff, confidently wrong about a library the moment you asked anything version-specific. I asked it a few things. I closed the tab.

None of that threatened me, because none of it thought. It completed. I was still the one making decisions.

Cursor shipped in 2023 and was already on my radar, but the noise caught me through 2024.

It was Twitter, mostly. The Cursor account itself, plus a few builders I followed, all posting clips. Nobody in my circle was using it yet. I was watching it alone, which somehow made it worse. People weren't posting autocomplete anymore. They were posting a small agent that read a whole repo, proposed a refactor, wrote tests, and explained itself.

That's when I froze.

I thought if I installed it, I'd stop learning. I'd stop feeling the small pleasure of writing a function from scratch, of arguing with myself over a variable name, of the 30 minutes it takes to move a tricky piece of state to the right place. I thought I'd become a passenger. And I really liked being the driver.

There was another fear under that one, and it took me a while to name it. I was scared that if I leaned on AI now, I'd never be able to go back. That I'd forget how to think about code without it. That the muscle I'd been training for a decade would go soft in six months.

So I waited. I let the hype wash past me. I told myself the tools weren't mature. None of that was true. I was just nervous.

August 2024. I don't remember the exact trigger. A deadline, probably. Curiosity, eventually. A year of watching from the outside had worn the fear thin.

The first week was uncomfortable in a way I didn't expect. Not because Cursor was bad. Because it was good, and every line it wrote correctly was a tiny piece of my identity I had to give up.

The second week, something flipped.

I stopped treating it like a writer and started treating it like a pair. One that thinks fast and never gets tired. I'd describe what I wanted, argue about trade-offs, ask why it picked that pattern, learn one thing I didn't know, push back, try again. I was reading more code than I'd read in years. I just wasn't typing it.

The prompts themselves? In French, by the way. I tried English for a week. It didn't buy me anything.

A few months in, Cursor bought Supermaven. The autocomplete I'd kept running in the background quietly became part of the editor in my foreground. I never had a reason to open Copilot again.

Cursor@cursor_ai

We are excited to announce that @SupermavenAI is joining Cursor!

Together, we will continue to build Cursor into a research and product powerhouse.

8:02 PM · Nov 12, 2024

2k

Open on

The tools kept arriving after that. Claude Code, agents, whatever shipped next month. I tried them all without the flinch I used to have. The hard part had already happened.

And creatively? I've never shipped more.

The shape of my work moved one level up.

I don't type most of the code anymore. It comes from a prompt I refined. What I do instead is think about shape. Architecture. Where a boundary belongs. Which abstraction is going to hurt us in a year. What the user will feel when this loads on a phone on a bad train connection.

Those are the parts I always loved most. The parts I used to rush through so I could get to the fun of typing. Now they're the job.

I won't pretend it's all clean. There's a dependency I didn't ask for, and I notice it. Some days I catch myself reaching for the agent to rename a variable, and I stop and rename it myself. Some weekends I rewrite a whole design pattern by hand, just to prove to the panicky part of me that I still can.

But here's the twist I didn't expect. The AI sometimes writes lines I don't fully understand on first read. I ask it why. It explains, clearly. And I walk away knowing something I didn't know five minutes ago. I'm learning as much with it as I ever did on forums late at night. Maybe more.

So the rule I try to hold isn't write more code without AI. It's smaller, and harder. Never stop asking how it works, why this pattern, whether there's another way. That's the muscle worth keeping. The rest follows.

I don't have a clean answer for what developers are for in 10 or 20 years.

I suspect a lot of us will be orchestrators. Editors of intent. People who decide what should exist, in what order, under which constraints. I also suspect there will always be legacy codebases messy enough that no model wants to set foot in them, and humans will keep getting paid to wade in.

But that's a guess. Anyone confident about this timeline is selling something.

What I do know is this. The reason I loved Cinema 4D as a teenager, VB.NET a few years later, React in my first real jobs, and Cursor today, is the same. I like being able to imagine something and then have it exist on a screen a few hours later. The tools kept changing. The drive didn't.

Install it.

Not because you'll be left behind if you don't. The industry will be fine either way. Install it because the thing you're afraid of losing, the joy of creating, is the exact thing waiting for you on the other side. Typing was never the point. Making things was.

And the fear of dependency is worth taking seriously, but not worth obeying. Use the tool, then put it down sometimes. Write a function by hand on a Saturday. Read a codebase with no agent in the sidebar. Keep the muscle warm.

We're still early. The tools are still clumsy in ways we'll laugh at in five years. Whatever is coming is going to rearrange more than our editors. But if the last decade taught me anything, it's that curiosity is the only thing that travels well between eras.

Install it. Stay curious. Keep making things nobody asked for.