Cheezy World

Traditional Agile development teams composed of a Product Owner (PO), a Scrum Master (SM), several Developers, and one or two Testers are incompatible with AI-driven development. The old team structure was built around the assumptions that things happen at a human developer pace and achieve human developer quality. We built all of our processes around this human pace - everything from POs working with stakeholders to define and elaborate user stories to SMs facilitating Sprint Planning and Daily Standups and even our approach to testing.

When you build an API designed for AI agents, you quickly discover something humbling: agents make the same mistakes over and over. Not because they’re incapable, but because they’re working from memory instead of reference material. I spent a few days experimenting and ultimately fixing this in Stride, and the results surprised me.

The Problem

Stride is a kanban-based task management platform built for AI agents. Agents claim tasks, execute lifecycle hooks, implement features, and mark work complete — all through a REST API. The system works well when agents format their requests correctly. The trouble is, they often don’t.

I see a lot of people posting about how bad AI is at creating code. I, along with everybody I work with, am having the opposite experience where AI is producing high-quality code that is well factored and tested. For a long time it has been a mystery how people can have such opposite experiences.

I now have three cases where close friends were making such statements and I offered to have a chat to understand their experience. In each case, they were using tools that were far from ideal and the tools they had were not configured at all. When I told them about options to provide AI with more instructions and constrain / direct how it works they were surprised.