Musings of the Day, 2/14/26:

Spending countless hours (and tokens) over the last week playing with both OpenClaw and Claude Code projects has really clarified one of the big challenges to AI:

How to get enough work done in an acceptable context window before costs explode and results degrade. I feel like I’m dealing with Guy Pearce in “Memento” where I have to constantly leave breadcrumbs for my AI agents to overcome a debilitating loss of daily memory and context.

Playing with OpenClaw truly quantified this problem for me because I see the explicit token use and cost, whereas this phenomenon is hidden to most LLM users through hallucinations and degraded results.

Opinion:

A machine can’t truly have a “soul” until it has a human-like context window, and that’s gonna take some doing.