miniblog.

After further playing with my LLM project, I'm surprised how hard it is to tune with system prompts. My agent kept saying "obviously" even though my prompt said "helpful, professional". Eventually I found that a "courteous" prompt gets "of course" which is better but not ideal.

I've been tinkering with an LLM interface with my calendar and it's really nice to use. "Delete my April trip" is a lot less fiddly than opening calendar, switching to the month view, scrolling to April and so on. I should probably build an undo as well though.

I certainly see the appeal of an LLM system with full context and tool use (OpenClaw), but the lethal trifecta puts me off deploying it. I do like the idea of an agent with a heartbeat though. A bunch of nice cron-style workflows can be built on top.