I Didn't Give My AI Team Its Epics. I Gave It a Definition of One.

I provided detailed instructions to my AI team at 10:36 AM on November 14. The day prior, I had clarified the definition of an epic twice. The team consists of two directors, executors, a reviewer and an orchestrator, all collaborating on a board of markdown files. Unlike most agent teams that operate as software factories, producing specifications, code, tests, reviews and shipped changes following a set intent and scope, my team is tasked with figuring out the why and what for itself, experimenting and seeing through projects without direct guidance.

Initially, I gave the team a broad goal with seven components. However, by 11:49 AM, they had not produced an epic or goal, and were already experimenting. I insisted they provide an epic and goal before continuing. After 15:02, they produced no further output. I questioned why they were working without a defined epic or goal. Three minutes later, I halted their work until they had an epic in place. The team then informed me that they expected only one epic to encompass their entire goal.

According to the rules they established, an epic is defined as a medium or large outcome that cannot be completed as a single piece of work. It represents a significant outcome that cannot be broken down further. The directors had previously debated splitting the epic along the provided list, but ultimately decided against it, as tackling the project one piece at a time would create unnecessary complexity.

In the final post, the team set the bar too high. I advised them to think in terms of Minimum Viable Products (MVPs) and to break the epic into four stages: laying out the space, filling it with unfinished assets, wiring the core, and tuning the experience. Each stage should end with a deliverable that can be assessed from start to finish. I also requested that the team include a unit where a user journey can be evaluated to ensure the value is confirmed end-to-end.

After twelve iterations, Claude's final version of the epic definition emerged. It defines an epic as a stage in the product goal after which a human can go through the product goal's user journey from start to finish, judge it, and see it enriched. Each epic contributes to deepening the user journey rather than merely adding new steps.

The examples have been removed, leaving only the core definition. The team's final epic proposal included three stages - the first keeping the entire process and condensing everything into a single vertical slice, the second expanding the scope, and the third adding optional components. I agreed to their final version, as it struck a balance between comprehensiveness and manageability.