I Built a Tool to Tell Me What to Write About
I spent two weeks building a system to tell me what I already knew I wanted to write about.
That's not quite the gotcha it sounds like. But it's close enough to be worth being honest about.
The idea was simply enough. I have a habit of making git commits, opening issues, scribbling things down, and then completely failing to turn any of it into a blog post. The gap between "I had a thought" and "I wrote a thing" is where most of my ideas go to die. So I built a tool — something that would look at my recent commits, pull in some context, and use Claude to suggest what I might actually want to write about next.
It works, mostly. You run it, it reads your git history, chews through some context gathering, hits the API, and spits out a handful of blog ideas with pitches and hooks. The commit history tells a story if you squint at it right, and it turns out Claude is pretty good at squinting.
Here's the awkward bit though. When I actually ran it for the first time and looked at the output, I recognised every single idea immediately. Not because they were obvious — because they were already rattling around in my head. The tool hadn't discovered anything. It had just written down what I already knew and hadn't gotten around to saying.
Which made me wonder what I'd actually built.
There's a version of this story where I'm being clever and efficient. Automate the boring parts, keep the creative parts for yourself. Let the machine handle the biolerplate while you do the actual thinking. That version is probably true in some sense.
But there's another version where I'd spent two weeks building a system to avoid the actual problem, which is that I find starting hard and having a list of good reasons not to start is very useful. Building the idea tool was a great way to not write anything while feeling extremely productive about it. The commits were piling up. The system prompt was getting refined. The Claude integration was working. Everything was great except for the part where I was writing blog posts.
The question I keep coming back to is whether outsourcing the "what should I write about" is actually cheating. I'm not sure I have a answer. Creativity, at least the kind I do, is mostly just noticing things you've already half-noticed and then forcing yourself to say them out loud. The git commits were already there. The ideas were already there. The tool just held a mirror up.
But there's something a bit funny about using AI to tell you what you think. Or maybe there isn't, and I'm being precious about it. People talk to friends to figure out what they believe. They keep journals. They go for walks. Maybe running your commit history through Claude is just the nerdier version of the same thing.
What I do know is that the post it suggested I write — this one, about the tool itself — was not something I'd have chosen on my own. That one I'll give it credit for. I was too close to it to see the irony.
I don't think I've resolved anything here. The tool exists, it's useful, and I used it to generate the idea for a blog post about whether I should be using a tool to generate blog ideas. That's either very efficient or completely mad and I'm not sure which.
I'll probably keep using it. It's just making it slightly harder to pretend I don't know what I want to say.