Read

the biggest danger with long-running agents following an elaborate plan is the same as always: When it works, magic. When it makes a small mistake early on, you now have 6000 lines of code that needs a rewrite. Waterfall didn't work for humans, why would it work for agents?

he most important thing to know about labor substitution, the place where any serious analysis has to start, is this: labor substitution is about comparative advantage, not absolute advantage. The question isn’t whether AI can do specific tasks that humans do. It’s whether the aggregate output of humans working with AI is inferior to what AI can produce alone: in other words, whether there is any way that the addition of a human to the production process can increase or improve the output of that process. That’s a very different question.

With AI you can DDOS your enemies through a legal system that has no defense infrastructure to protect itself or its citizens.

When a system gets overrun, the people with means find the side door. When public school systems decline, the rich kids go to private schools. When cities get dirty and violent, people with means to move go to the suburbs.

The people who can afford to step outside the system, which has been flooded with zero-signal AI slop, will quickly and efficiently do so. In fact, they will be greeted with open arms by the person on the other side of the transaction.

The most telling thing about the LLM moment is what this technology is actually good at. LLMs write code, generate images, produce music, summarize documents, draft prose… which is to say, they have achieved mastery over the exact domains that were already, by any sane measure, overprovisioned. Was anyone saying that we didn’t have enough digital writing, images, videos, music, video games, or applications, a few years ago? The core triumph of technological growth is taking scarcity and creating abundance. Well, LLMs create an abundance, that’s for sure. But there was already an abundance of text, online, and an abundance of images, and there’s some insane stat like 24 hours of video gets uploaded to YouTube every second or whatever, and yes, there has been an abundance of code, of programs, of apps. And before we got these fancy new tools to produce more code, there wasn’t a lot of people saying “Gee, what we need is more apps, the app store is too empty.”

The only solutions to our problems - the problems of hunger, of poverty, of injustice, of disillusionment, of alienation - are political solutions. I understand feeling totally defeated by that idea, given what politics is like on this planet. But it’s all we have. We start to build the political structures that can enable humanity to take care of all of us or we drown. There is no fate but what we make.

You let them run free, and they are merchants of complexity. They have seen many bad architectural decisions in their training data and throughout their RL training. You have told them to architect your application. Guess what the result is?

And I would like to suggest that slowing the fuck down is the way to go. Give yourself time to think about what you're actually building and why. Give yourself an opportunity to say, fuck no, we don't need this. Set yourself limits on how much code you let the clanker generate per day, in line with your ability to actually review the code.