AI Coding and Writing Things Down
A few weeks ago an agent quickly solved something in one shot that would’ve taken me hours, or even days-it was one of those convoluted problems that in the past I'd need to resign myself to chewing on all day. I've had a lot of these moments over the last couple of years. I call them "whoa, what's my job" moments.
I was a very early adopter of agentic coding tools. I loved the speed at which I could move and how good the models were at the mundane stuff. As they've improved, those "whoa" moments have stacked up, and somewhere along the way I mostly stopped writing code by hand. This isn't unique to me. Boris Cherny, the creator of Claude Code, has talked extensively about giving it up himself. Plenty of others have said the same. At the startup where I work, I'd estimate 90% or more of our code is now written by an agent.
I'm not pointing this out as a problem, and I'm definitely not anti-AI. However, while working more and more with the models and with people using the models, I've started to notice something happening to me, and to people around me—something Lars Faye recently put words to. If I spend all my time managing agents that do the work I used to do, and none of it keeping my own skills sharp, I eventually lose the ability to manage those agents well. I also lose my connection to the craft I've grown to love. I don't think I've gone far down that path, but it's clear how easy it would be. And the person at the end of that path has lost the expertise and taste that make them valuable in their work and, more than that, make them a fuller human. That's the part I take seriously.
Here's the thing I keep coming back to, though: the same technology that can hollow out your thinking can be pointed the other way. LLMs are incredibly powerful tools. They aren't going to change things; they're actively changing them now. What I do at work today is radically different from what I did two years ago. Anthropic recently released Claude Fable 5, a step change better than what we'd been using for months, and we're still early on the growth curve of AI. The question isn't whether the tools reshape us. It's whether we’ll be shaped to think more or less.
For me, the solution has been almost embarrassingly low-tech: I write things down. On paper. When I'm coding now, I write out the problem I'm working on, what I expect the outcome to look like, the structure of the system, before I ever hand it to an agent. The simple act of putting pen to paper forces me to understand, to probe, to strain my mind in a way that keeps it sharp. The agent does more of the typing; I do more of the thinking. That's the trade I want.
That “pen to paper” ethos is why we built Writually. It asks you to spend some time each day putting thoughts on paper and investing in your own brain and creativity instead of outsourcing them. And yes, there's AI under the hood, but leveraged deliberately: it gives you prompts to write about, questions to follow up on, and themes pulled from what you've written to push your thinking further. It's one example of a different kind of app: one that doesn't ask you to surrender your thinking or creativity to AI, but uses it to make you think more, create more, and stay cognitively fit.