I read code every day. Then I tried writing some.
What AI gave me
For about a year now, I've been using AI for pretty much everything in my workflow. And honestly, I've loved a lot of it. I get to work on bigger problems, think about architecture, and build things I wouldn't have had time to build before.
I think we should be using these tools. The expectations from companies, investors and the people around us make sense to me.
What surprised me
But little by little, I handed over more of the work. Every now and then, I caught myself handing over some of the thinking too.
The other day, I thought I'd try a LeetCode problem. Just for shits and giggles. See how it goes after a year of working this way.
I was genuinely shocked by how much I struggled to write the code.
I still read code every day. I review what the models produce. But sitting in front of an editor and coming up with it myself felt very different.
It reminded me of learning a language. You can understand someone speaking. You can read what they write. Then it's your turn to speak, and you can't find the words.
That was the feeling. And I wasn't expecting it.
What I want to find out
I don't know how much of that was rust, the particular problem, or the way I've changed how I work. But it made me curious enough to build Muscle Memory.
I want to see whether you're noticing something similar. I have a hunch some of you might be as surprised as I was. Maybe you'll find it easier than you expected. I want to know that too. Where do people struggle, where do they do well, and how does that change with practice?
Why the data matters beyond me
There are plenty of opinions online about what AI is doing to developers. I don't trust the confidence of those opinions very much. I'm someone who needs data.
- Developers can look at their own experience, not just read about everyone else's.
- Educators can see where practice is needed.
- Teams and tool builders can ask better questions about which abilities AI-assisted work exercises, and which it leaves unused.
What the data can and can't say
Challenge data can describe how people solve these problems. It won't, by itself, prove that AI caused anyone to lose a skill. Strong performance and surprising non-results matter just as much as struggle.
Once you accept the Terms, we record how you work through a challenge: your keystrokes and edits in the editor, your runs and their results, and the optional answers you give about yourself. We keep what happened apart from what we conclude from it. Nothing is shown to anyone else next to your name, you can go stealth to stay out of anything we publish, and you can delete everything any time. The Privacy Notice lists it all. How and when findings are shared is still being worked out, and we'll say so here when it is.
The editor has syntax highlighting, type errors and hover types. Completion, chat and generated code are off. Python, JavaScript, TypeScript, SQL, HTML, PHP, Ruby, Go, C, C++, C# and Java run in your browser.
What I want to keep
I enjoy working on the bigger problems AI has opened up for me. I also want to be able to take a smaller problem, understand how it fits together, and write the code myself. I want that choice to stay available to me. I still want to speak the language.
Try one
Pick a coding challenge, write the solution yourself, and see how it goes. Maybe it feels rusty. Maybe it comes back easily. Either way, I'm curious.
— Ekku Leivonen