How is CAN addressing different from I2C or SPI?

CAN is message-based, not node-based. An I2C transaction targets a device by its address and SPI selects a device with a chip-select line, both are destination addressing. A CAN frame instead carries an identifier that names the message content (e.g., "engine RPM" or "brake status"), and the frame is broadcast to every node on the bus; there is no source or destination field. Each node uses hardware acceptance filters to decide which IDs it cares about and ignores the rest. This decouples producers from consumers: a sensor broadcasts its measurement under a known ID, and any number of nodes can consume it without the sender knowing or caring who's listening, which suits a distributed control system. The design implication is that you must define a system-wide ID map (which ID carries which signal, and thus its priority), and that's exactly what higher-layer standards like CANopen and J1939 standardize on top of raw CAN. So with CAN you think in terms of messages and their priorities, not device addresses or select lines.