We're in uncharted territory. These tools are less than three years old, and nobody knows what happens when someone uses the same AI assistant for a decade. How much should it remember? How should it handle years of accumulated context? Meanwhile, we're seeing a Cambrian explosion of AI apps, each experimenting with their own memory approach, while the underlying models get more powerful every week. There's no playbook, no settled best practices—just everyone trying different things and seeing what sticks.

Claude Memory: A Different Philosophy | Shlok Khemani

This was a delightful and insightful read.