One of my standard agent uses is to create “mini-books” on historical topics, such as countries I visit. I have an app that allows me to name a topic or place. This feeds into a pre-written prompt that generates a syllabus written by 5.5 Pro, which is itself composed of prompts for each topic. These prompts are then fed into 5.5 Thinking, and the outputs are put into a nicely formatted website. The full output is usually the length of a short book. All I have to do is write the topic and press enter, and the whole thing appears a handful of minutes later. This is a very vanilla use of coding agents and LLMs, but it’s a remarkable capability that continues to blow my mind. I have never felt more joy in using computers than I do these days.

One of my standard agent uses is to create “mini-books” on historical topics, such as countries I visit. I have an app that allows me to name a topic or place. This feeds into a pre-written prompt that generates a syllabus written by 5.5 Pro, which is itself composed of prompts for each topic. These prompts are then fed into 5.5 Thinking, and the outputs are put into a nicely formatted website. The full output is usually the length of a short book. All I have to do is write the topic and press enter, and the whole thing appears a handful of minutes later. This is a very vanilla use of coding agents and LLMs, but it’s a remarkable capability that continues to blow my mind. I have never felt more joy in using computers than I do these days.

tweeting about this caused me to think about the structure of this project, which was one of the first things I did when coding agents crossed the good enough threshold circa October/November. as currently conceived, the project is very much rooted in a 2024/5 conception of ai: decentralized api calls to llms. chatbots.

but after tweeting this, I realized you can now plausibly just create a digital organization filled with teams of agents, since the agents are getting so good at orchestrating teams of agents. if you're willing to spend the tokens, that is.

so what I am attempting is a kind of digital publisher, with 'submitter' agents who pitch book proposals, editors who pick them, research agents, writing agents, fact-checking agents, editors, and similar--hundreds in total. the token usage will be outrageous. we will see how much better the quality is.