An AI agent is the most demanding user your framework has ever had, and by far the most honest. It can't ask a coworker, can't lurk in your Discord, can't intuit that "it just works that way." Every single thing it gets wrong is something your framework failed to communicate.

For the past year the Critter Stack team has been adapting to our new AI world order. This is our report of all the spaghetti we've thrown up against the wall. From leaning even harder into our already terse coding style that saves on token usage, to curating AI Skills files to make AI agents lean into that style, to building reflective capabilities into every part of our system so AI agents have a mix of command line diagnostics or MCP tools to understand how a Critter Stack application is configured and behaving. We'll also talk about how we've tried to build AI friendly test harnesses that allow an AI tool insight to a system under test -- especially for the type of Event Driven Architecture that our tools specialize in.

An AI agent is the most demanding user your framework has ever had, and by far the most honest. It can't ask a coworker, can't lurk in your Discord, can't intuit that "it just works that way." Every single thing it gets wrong is something your framework failed to communicate.

For the past year the Critter Stack team has been adapting to our new AI world order. This is our report of all the spaghetti we've thrown up against the wall. From leaning even harder into our already terse coding style that saves on token usage, to curating AI Skills files to make AI agents lean into that style, to building reflective capabilities into every part of our system so AI agents have a mix of command line diagnostics or MCP tools to understand how a Critter Stack application is configured and behaving. We'll also talk about how we've tried to build AI friendly test harnesses that allow an AI tool insight to a system under test -- especially for the type of Event Driven Architecture that our tools specialize in.