The end of manual programming

`1:00:35`

And at different levels of abstraction. When I'm working in Ruby code, and we have a lot of Ruby code because it's in our entire business, and I'm asking agents to make changes to that code, I still sweat the details. What I'm coming to realize is that the economic payoff of that sweat is diminishing rapidly. The reason why I, for 25 years, was sweating every line of code so judiciously was because I knew the payoff of keeping an architecture coherent and malleable was software that could change and evolve- quickly with a small team, and not exorbitant cost, and not introducing a bunch of bugs when you change one thing over the other. That was the driving economic argument for why you should write beautiful code, 'cause beautiful code is easier to understand, it is simpler, it is more malleable. That was premised on humans doing the modifications. I think it is an open question to which degree this still matters. Now, it does matter, and the reason I say that, at least for the moment, is that tokens are still scarce. At this moment in time, we are all token limited. Well, not all, but anyone who doesn't have endless budgets are token limited. So therefore, there is great payoff to writing systems that agents have an easier time dealing with and evolving without having to relearn the entire context. Just like humans, if they can make iterations and changes to the code base without wrecking the architecture, they can make the next change just as cheaply as the last one.

And this is the classic ball of mud where you end up with a system that's a ball of mud-... because it's just put together in a way where nothing is connected and it's just a real mess, right? I've seen that with agents, and I've seen it in our own code bases where, again, the first PR is, like, mediocre of quality, and then if you add another PR on top of that, and then five more down the line, it's not very good, right? So there's still a payoff to that, but that payoff is premised on our current moment. I am now able to... And this is, by the way, where the AI psychosis really comes in, when you try to extrapolate what nine months from now it's gonna look like. What two years from now it's gonna look like. But as an intellectual experiment, I think the first computer I used was a Commodore 64.

`1:03:11`

It had one megahertz CPU and 64K of memory. Everyone who wrote software for that machine internalized certain constraints. Well, all the constraints. That machine was nothing but constraints. If you put that programmer in a time machine and teleported him or her to 2026 and gave them a modern computer today and said, "Now write me a piece of software," they'd be lost for a little bit because all their techniques and all their heuristics would simply be wrong. Or not wrong, because efficient software is still beautiful. They would be out of date with the value that they could create. The amount of optimization you have to apply for a one megahertz computer to produce a video game is just very different from what you have to apply today.