Software is cheaper to make than ever. Maybe the economics will change if the way tokens are being subsidized today goes away, but given that we will probably be able to do more inference with fewer tokens in the future, I would imagine the dynamic will hold true.

I don’t think being able to move faster is all good, however. I liken it to having a fast car. Sure, you can travel a longer distance in a short amount of time, but all that mileage is a waste if the destination isn’t clear. A higher rate of work does not equal progress.

And so, as a Claude Pro user, I’ve found some relief in the usage limits I bump up against every few hours. Yes, it’s incredibly frustrating to have my stream of consciousness development come to a screeching halt out of nowhere. But I’ve realized over time the limits aren’t a break from Claude. It’s a break from me. Me and my brain, which has inevitably run off with serious momentum into territory that may or may not be beneficial. The limits are a circuit breaker for that intoxicating feeling of materializing software at the speed of thought. That break is important in this environment, because it’s easy to feel like you’re going somewhere when the wind is blowing in your hair.

I’ve seen people write about the benefits of friction. Of having that barrier between desire and gratification that helps you realize what is really important instead of just instinctively reacting to an immediate want. We’re in a world now where the difference between want and need is getting real fuzzy thanks to technology. That might be great for reinforcing addictive behavior in the name of profits but not so great for the benefit of humanity.

I really like using AI for making software. But I also realize its limits, and the fact that because these machines will make anything for me, it’s probably a good idea to take more time to think about what exactly I want to make.