This is an interactive visualization that shows what happens behind the scenes when you ask an AI coding tool (Claude Code) to do something. It replays the actual sequence of API calls — messages sent back and forth between the software on your computer and the AI model — as an animated factory assembly line.
The visualization is self-referential: the task it depicts is the task of building this visualization. The user asked Claude Code to create a visualization of its own API calls, and the assembly line shows every step the AI took to build the page you're looking at.
When you click Play, a glowing box (representing the user's request) moves left-to-right along a conveyor belt. Each stop is one API call — a single round-trip where the software sent a message to the AI and waited for a response.
Click any station to see details. Toggle between Technical view (tokens, cache stats, tools called) and Plain English view (human-readable descriptions).
This accompanies a presentation called "Everything You Know About AI Is Wrong," which uses a central metaphor: AI models are chickens (enthusiastic but not bright — think HeiHei from Moana), and the software systems around them are sheepdogs that keep the chickens moving in the right direction.
A typical API call is roughly ~75% guardrails and instructions ("dogs"), ~24% previous AI output ("chickens"), and less than 1% actual user text ("you"). The architecture — not the model — is what makes the chickens productive.