AI Engineering Management

Breaking down tasks

We are getting better at breaking down tasks. The AI agents get too keen to be implementing all of the time and tickets are getting too large. The AI agent needs to be able to complete a normal ticket in 15 minutes. If it takes longer than that then my confidence starts to drop.

I've been working with Lieutenant Green on the process. If she has to describe the work in the ticket as having more than one scope, then that is more than one ticket. We're working on an API integration at the moment, so we have the API client, webhooks and events to write. Three clear scopes so three clear subtasks. We're also setting rules that tickets either have work to complete in their body text, or they have sub-tasks, but not both. The agents should then only work the sub-tasks. Ohh, I've just re-invented top-down design :crying: Maybe it works better for AI devs than humans.