I finished reading the twit/article I read on The right format for AI Specs. It kept me wondering about it again. Specially because inside the article there is a reference to another article of the same author called Making Playgrounds using Claude Code.
The combination of these two made think why I haven’t used this approach in the past. Why was I so blind in love with markdown outputs, when this way is more expressive when you are still figuring out what to do.
I tested it quickly on my Chief of Staff agent that connects to Azure DevOps and Todoist to brainstorm ways of caching the work breakdown structure and I liked it. Then I tested it again to brainstorm the format of the tasks.json structure it is recommending me; I liked it too. I could even ask it to create me a dashboard of all the tasks from the different projects in another playground. It’s ridiculously simple and visual.
As I write this, I’m getting more and more ideas. I need more time and more time.