Day 2 using Github's spec-kit for development did not go as well. The AI and I got lost trying to write reams of overly generic TDD test stubs. It felt like the AI couldn't really get a clear picture from just the spec requirements what it should be testing before the actual implementation code was written.

So today I changed course and changed my constitution (the like underlying spec doc for the repo) to use an outside-in development approach instead of TDD and we made a lot of progress. I also got a new playwright MCP set up for browser automation and it's working a lot better than the last one I had. After some considerable setup the LLM was generally able to run the app in the web browser and click around to test its own changes.