A conference talk arguing that despite the massive leap in AI capabilities, the interaction protocol — prompting — is still fundamentally a batch model inherited from punch cards. The speaker introduces three concepts: channel (the physical medium), expression (the richness of meaning it can carry), and protocol (the rules of exchange). While LLMs have dramatically expanded expression, the protocol remains unchanged: assemble a complete request, submit, wait, repeat. This is compared to punch card batch processing. The talk argues that prompt engineering is just learning to package batch jobs better, and that the real opportunity is building AI interfaces that participate in real-time human conversation — interrupting, backchanneling, tracking speakers, and choosing their moment — rather than waiting passively for a submitted turn. Demos of NVIDIA's Personal Plex and JoinIn AI's multi-party meeting assistant illustrate what truly conversational AI interfaces could look like.