I'd rather the system start managing models rather than each app doing their own thing. Whether that's engineering for and supporting KServe and inference snaps, an Ollama systemd service, or something else I don't care. But it's time to get serious about system tooling around AI.