A few months ago, I came across a video of a DJ talking about making music with AI. He described spending hours generating tracks, listening through them, throwing away 99% of them and occasionally finding one that felt magical. The strange part was that he wasn’t describing the process with excitement. He was describing it like a routine. And I kinda related with what he said.
I don’t make music. I write. I design. I spend an unhealthy amount of time staring at Figma files and half-finished documents while convincing myself that this my mobile app had the prettiest UI. But I’ve felt the exact same thing. Generate twenty homepage directions. Thirty copy variations. Ten article structures. A hundred content ideas. Spend hours reviewing them. End the day with more options and less conviction.
The output is objectively better than what I could have produced alone in the same amount of time. The process somehow feels worse.
That’s the part I’ve been struggling to articulate.
I love AI tools. I use them every day. For writing, design, research, random questions that would have been Google searches a few years ago. I’m probably closer to the “AI maximalist” camp than the “AI is ruining society” camp.
At various points I’ve probably flirted with what I jokingly call “AI dependency.” The kind where your first instinct isn’t to think, search, sketch, write or experiment. It’s to ask a model. I’ve even written before about what I think are the five layers of AI usage. The first layer is obvious productivity gains. The last layer is when the model quietly becomes your default interface to the world. I’ve poked that line more times than I’d like to admit.
Which is why this observation annoys me.
Because I don’t think the problem is that AI makes us less productive.
If anything, it often makes us more productive.
I think the problem is that it changes the nature of the work.
A lot of creative work used to begin with boredom. Or confusion. Or staring at a blank page longer than you’d like.
The blank page sucks, by the way. I feel like people (including me) get weirdly nostalgic about it now. Most of my relationship with blank pages has historically been panic. But occasionally, somewhere inside that panic, an idea would show up. Not always a great idea. Mostly horrible. Then another. Then another. And eventually something would click.
AI mostly removes that phase. Which is the entire sales pitch. You don’t need to stare at a blank page anymore because the page is instantly full. The first draft appears immediately.
The problem is that generating ideas and evaluating ideas are completely different activities. Instead of figuring out what I think, I’m reviewing ten versions of what a model thinks I might want to think. Instead of discovering a direction, I’m choosing between directions. I know it’s confusing but trust me it’s not the same.
One thing I’ve always been fascinated by is what I jokingly call sourcing entropy. Most of the genuinely good ideas I’ve had in my life came from one of two places. Either I spent enough time alone with a problem that something unexpected surfaced. Or I stole three unrelated ideas from smart people and smashed them together until they became something new. Neither process is efficient. Both involve wandering around mentally for far longer than any productivity guru would recommend. But I think that’s part of why they work.
Original ideas rarely arrive on schedule. They tend to appear while you’re getting coffee, taking a shower or pretending to work while actually doing something completely different.
AI is almost the opposite. It is incredibly good at collapsing possibility space.
Give it a vague thought and it immediately starts adding structure. Sometimes that’s exactly what I want. Sometimes it’s the difference between shipping something and getting stuck. But I’ve also noticed that if I involve it too early, I often lose the part where I would have discovered something myself.
The final result might be better. The journey there feels less memorable. Less personal. Less mine.
What happens when we get so good at outsourcing the difficult parts of thinking that we accidentally outsource some of the enjoyable parts too? Every new technology moves us one layer higher up the abstraction stack. That’s usually a good thing. Nobody is arguing we should go back to writing assembly code. But every abstraction also creates distance. And I suspect one of the biggest differences between people over the next decade won’t be who uses AI and who doesn’t. Almost everyone will.
The difference will be what they choose to outsource. I’m still trying to figure out where that line is for myself. I just know that every now and then I catch myself generating ten versions of something when what I actually needed was to sit with the blank page for ten more minutes.