When I was an undergrad studying animal acoustics, I was designing the data collection part of our research project and I had a question about how to structure the data. I asked my PI if we should do it sequentially or devise some creative plan and he just told me: “Ian, if you don’t have an evidence-based reason to do something in a particular way, just do it randomly” and that stuck with me. This week my manager was asking me in our 1 to 1 to share how I use AI and after trying everything from writing custom sub-agents, tidying my CLAUDE.md, having custom skills, I now do… nothing.
Quite literally, I open claude from my Home directory and just go to bat with it in short-lived sessions that do not last more than 8 hours. The context is in my head. I can type a 6-sentence paragraph with my ideas in 2 minutes better than I can maintain custom software.
All of these folks that you see talking about their complex AI setups, do not have 1) the statistical rigor to back their claims up with proper testing and 2) are just trying shit out and seeing what sticks.
This last year has pushed me more to accept the absurd in life as Camus called it, and I feel like this is one more manifestation of that. We’ll see how it plays out.