The Prompt Is Still a Punch Card - Ted Johnson, JoinIn AI
Ted Johnson argues that current AI interfaces, particularly prompting, operate on an outdated "batch processing" protocol akin to punch cards. Despite advanced LLM capabilities, this interface design forces humans to adapt to machines, hindering natural interaction. He advocates for a shift towards human-compatible interfaces where AI actively participates in real-time conversation, leveraging its intelligence to remove user burdens and amplify human potential.
Ted Johnson reframes Artificial Intelligence not just as an intelligence technology, but fundamentally as an interface technology. He opens by highlighting the inherent strangeness of current AI interaction: typing requests into a small box and waiting for a response, a process he aims to make feel unfamiliar again despite its normalcy. His central “disappointment” stems from the observation that despite AI’s immense power, users often still have to “learn AI” and find its use unnatural.
To illustrate this, Johnson introduces three key concepts:
- The Channel: This refers to the physical medium through which intent is conveyed, such as a keyboard, microphone, or a prompt box. He emphasizes that channels dictate the kind of signal they can transmit (e.g., text as discrete symbols versus voice carrying timing and hesitation). However, carrying more signal doesn’t equate to the machine understanding it. Historically, humans adapt to the machine’s chosen channel, even when it’s an arbitrary legacy like the QWERTY keyboard, which was designed under constraints that haven’t existed for over a century.
- Expression: This denotes the range and richness of meaning an interface allows to be conveyed. Johnson traces the evolution from assembly opcodes to modern programming languages, each offering fixed vocabularies. Natural language, facilitated by Large Language Models (LLMs), represents an “enormous” leap in expression, allowing users to communicate “almost anything the way you’d say it to another person.” An “ocean of meaning” — context, nuance, intent — can now be input, which machines can finally take in.
- The Protocol: While the channel has largely remained static (typing into a box) and expression has exploded with LLMs, Johnson argues that the protocol — the rules and shape of the interaction — has not kept pace. He asserts that “prompting is the protocol of a punch card.”
The Prompt as a Punch Card Protocol
Johnson draws a direct parallel between modern prompting and the batch processing protocol of punch cards. In the punch card era, users would pre-encode their entire request, submit it, and wait for results, often for hours or overnight, only to find an error, fix it, and resubmit. Prompting follows a similar pattern: “Assemble the whole request, submit it, wait, read what comes back. Something’s off, assemble it again, submit it again, and wait.” While modern AI offers quicker response times and interactive features like asking for updates, Johnson contends these are merely “interactive sprinkles” on a fundamentally batch system. The speed has “fooled us into thinking that it become interactive. It hasn’t. It’s still batch.”
He criticizes “prompt engineering” as a flattering name for learning the “rules for packaging up good old batch.” He states: “The protocol is the part we’ve had to learn. We just gave it a flattering name. We call it prompt engineering and treat it like it’s a power user skill. Strip the label off and it’s a set of rules for packaging up good old batch.” These rules are akin to the mastery a punch card operator had in assembling a deck to avoid failure. This “mastery” is concerning because it forces humans to adapt to the machine’s outdated protocol, rather than the machine adapting to human interaction.
The Mismatch and the Blame
The core problem is a significant mismatch: “Model capacity is shooting straight up. Reasoning, speech, vision, memory, planning, all curving upwards. The interface protocol, flat.” The human is still doing all the work around the LLM, managing context, timing, ambiguity, and repair. This makes the interface “feel like work,” and when things go wrong, users blame themselves. Johnson emphatically states, “it is not our fault. We are not bad at using AI. We are being asked to operate a brand new kind of intelligence through a protocol of a punch card.” The shortcut taken in enabling AI was to neglect the interface.
Towards a Human-Compatible Interface
Johnson offers examples of how interfaces could evolve:
- Critique of Current Voice Modes: He recounts an anecdote where a voice AI responded to a user’s casual comment (“Hey Ted, come on in”) not meant for it, demonstrating its single-slot protocol and lack of contextual awareness.
- Nvidia’s Personal Plex: This research model showcases true real-time turn-taking and interruption handling, where the AI “stops, it yields, it picks the thread back up.”
- JoinIn AI’s Approach: His company, JoinIn AI, focuses on improving the protocol by giving models a better understanding of human and group conversation. Their system takes action only when appropriate, labels participant statements (question, proposal, answer), and determines intent and speaker reference in a natural, collaborative setting. The AI participates as a member of the conversation, understanding context and choosing its moment, rather than waiting for a pre-packaged prompt.
Mindset Shift and Design Philosophy
Johnson urges a fundamental mindset shift: viewing AI as an interface technology that can remove constraints and amplify human potential. He argues against seeing AI as a “smarter machine hiding behind prompts” and instead as an intelligence that “should be able to meet us partway, if not all the way.”
- The Critical Design Question: “What burden are we still putting on humans only because the machine used to be too limited to carry that burden itself?”
- The Answer: It’s not always chat, voice, or markdown walls, but rather leveraging “the affordance humans already use with each other: Communication, a question, a pause, a sketch, a checklist, a quiet aside, or saying nothing at all.” The goal is an interface where the AI manages timing and modality, making computers more fluent with us and reducing the “translation tax.”
He concludes that if a machine can finally understand more of what we mean, then “we can and should stop reshaping ourselves to be understood by it.”