Latest / 1 More Paper / Closing Time: A Theatre of Augmentation — Njå et al. (2026)

Transcript

Closing Time: A Theatre of Augmentation — Njå et al. (2026) · 1 More Paper

• You’re listening to “Closing Time: A Theatre of Augmentation,” by Ola Njå and colleagues. Published in 2026. Abstract. When AI removes the friction of building software, the practitioners doing the building are the first to learn what freed capacity becomes. Between 2024 and 2026, a CPO and a CTO in a small Norwegian PropTech startup lived through a capability shift that compressed tasks estimated as months into days. They built more, faster, and with fewer people than their team structure had required. They felt augmented. They also stopped sleeping well. This paper stages what they saw as a duoethnography written as a play in seven scenes. The practitioners meet historical and contemporary thinkers on augmentation, acceleration, participatory and human-centered design, cognitive ergonomics, model epistemol-ogy, and calm technology. Dialogue is drawn from three rounds of recorded reflective conversations and two years of shared practice; author notes between scenes carry the citations and the argument. We ask one principal question: when AI removes the friction of building, what does the freed capacity become? We let the charac-ters on stage reach for words that fit. Some words are old. Some are new. We leave the question open, and the words with it. CCS Concepts. • Human-centered computing → HCI theory, concepts and models; Empirical studies in HCI; • Software and its engineering → Collaboration in software development. ACM Reference Format. Ola Njå, Arne Eirik Nielsen, Harald Holone, Suhas Govind Joshi, and Jo Her-stad. 2026. Closing Time: A Theatre of Augmentation. In Adjunct Proceedings of the 14th Nordic Conference on Human-Computer Interaction (NordiCHI ’26 Harald Holone Østfold University College Halden, Norway the email address Jo Herstad Department of Informatics University of Oslo Oslo, Norway the email address Dramatis Personae The Anchor Characters (present throughout). CPO Designer turned coder. Chief Product Officer of a Norwe-gian PropTech startup. Speaks from data. CTO Engineer. Chief Technology Officer. The one who said “the CTO disappears.” Speaks from data. The Bartender Does invisible work. Notices everything. Speaks last. The Visitors (in order of first appearance). Licklider (1960) Dreamed of human–computer symbiosis. Measured the overhead at 85 %. Returns at closing. Engelbart (1962) Built augmentation systems. Brings the brick-pencil. Recognizes his regenerative loop running in two people, not in an institute. Rosa (2013) Social acceleration. Promise becomes demand. Cannot be exited by going faster. Mazmanian (2013) The autonomy paradox. The tool that freed them bound them. Garry Tan (2026) Y Combinator. Champagne. “Boil the ocean.” Bødker (third wave, 2006) Scandinavian participatory design. Calm authority. Weizenbaum Watched people choose to be de-ceived. Offers the word mirror. Norman (four decades of self-critique) From The Design of Everyday Things through Things That Make Us Smart to humanity-centered design. Bainbridge (1982) Ironies of automation. Cold, structural, very few words. Bender (2021) Stochastic parrots. The model has no inside. Huang (2025) Observed thirteen professional developers and surveyed ninety-nine more. Pros don’t vibe, they control. Weiser (1997) Xerox PARC. Dreamed of calm technology — computing at the periphery of attention. Karpathy (2025) Coined “vibe coding.” Calls his own practice “a state of psychosis.” Sarkar (2025) Canary in the coal mine. Offers one frame and leaves. A Note Before the Scene Opens The dialogue readers are about to witness is drawn from three rounds of recorded reflective conversations (approximately 160 minutes) and two years of shared practice. One author arrives from design, the other from software engineering; the method’s analyt-ical leverage comes from their different vantage points: Sawyer and Norris’s convergence from divergent positions, within ana-lytical autoethnography; see also Garcia and Cifor. Three further authors, not embedded in the studied practice, are external interlocutors. The historical and contemporary figures who visit the bar are not in the transcripts. They give the practitioners’ observations the argumentative company they need. We chose theatre because the subject refuses the pace of a con-ventional empirical paper. By the time a controlled study has been designed, run, reviewed, and published, the tools in it are obsolete. What we offer is a case, carefully reported, in conversation with the thinkers who first named its parts. Every line spoken by the practitioners has a source in the transcripts, sometimes condensed, re-ordered, or translated for the stage. The form raises questions; it does not land a verdict. The principal question the scenes orbit: when AI removes the friction of building, what happens to the work, the workers, and the rest of their lives? A secondary question moves in the background: when the human retains executive authority but offloads technical comprehension, what does that division become? The foreground is where the paper places its weight. Scene One: The Bottleneck Moved A bar in Vaasa, Finland. October 4, 2026. The night before NordiCHI opens. The bar is old. Wooden tables. A long counter. Behind it, THE BAR-TENDER — a woman in her fifties, methodical and unhurried. She dries glasses in a way that suggests she has been doing this for decades and has no intention of stopping. CPO and CTO sit at a corner table. They have been here since six. It is now past nine. There are notebooks on the table, a laptop closed, and two beers — one nearly empty, one barely touched. CTO is mid-sentence. CTO: It wasn’t development anymore. It wasn’t finding the right people. It was — my time with the agent. That was the bottleneck. CPO: Your time with the agent. had hundreds of built-in decisions. Snapping behavior, magnetic thresholds, edge cases. Things we would have discussed for weeks. You just — made them. CTO: Because the cost of trying was so low. If snapping felt wrong, I changed the radius. If the magnetism was too weak, I increased it. I was building tools and writing code simultaneously. CPO: And before, I would have asked if it was possible. You would have estimated. We would have discussed. Someone would have had an opinion about priority. CTO: And we would have been wrong about the estimate. CPO: Always wrong about the estimate. LICKLIDER enters. He is older than anyone expects, wearing a rumpled suit that suggests government funding and late nights. He carries a small notebook — the kind with grid paper. He sits at the bar and orders whisky. Licklider: (to the bartender, apropos of nothing) I once measured how I spent my thinking time. This was 1957. I kept a detailed log for several weeks. (he opens the notebook, taps it) Eighty-five percent. Eighty-five percent of what I called my “thinking time” was spent on clerical activities. Getting into position to think. Drawing charts. Plotting graphs. Searching for information. Not thinking. Preparing to think. THE BARTENDER sets down his glass without comment. Licklider: I imagined a partnership. Human and computer working together. The human sets the goals, provides the judgment. The computer handles the clerical work. Eighty-five percent of human effort, freed. He turns to look at CPO and CTO. Licklider: I assumed the human would know what to do with the freed time. He sips his whisky. He does not press the point. He returns to his notebook. Author Note I Licklider’s 85 % figure is a measurement, not a prediction. AI agents have compressed the 2020s equivalent of that clerical overhead. The compression has not freed time. It has freed ambition: when a task estimated at three months becomes one week, the logic shifts from should we do this? to why haven’t we already? Ranganathan and Ye documented the pattern: AI made em-ployees “work faster, take on broader scope, and extend work into non-work hours — often without being asked.” Our data adds: the scope expansion is experienced as joy. Scene Two: The Brick ENGELBART enters carrying a small wooden block with a pencil taped to it. The arrangement is old and well-worn. He sets it on the table between the glasses — carefully, as if it were a museum piece, which in some sense it is. He sits. Engelbart: I made this in 1962. I tied a brick to a pencil and asked people to write. They wrote slowly. Imprecisely. Their hands tired. (he smiles, a little) What I was looking for was not the slowness. It was what happened above the slowness. When the basic act consumes everything, there is nothing left for the higher act. Sub-jects could form letters. They could not compose. The thinking they could do was bounded by the symbol-manipulation they could afford. CTO: De-augmentation. Engelbart: That was the word. I wanted to study augmentation by reversing it. (looking at them with interest) You are doing the opposite experiment. You have removed almost all the basic load. Syntax. Implementation. Lookup. The brick is gone. May I ask what you are doing with the freed capacity? CTO: Building more. Engelbart: Not thinking? CTO: (beat) I am thinking while building. Faster. About more. I am thinking at the level I used to wish I had time for. Engelbart: And the executive function — the part that organizes, supervises, decides what to build, what to stop building. How is it? CTO: Always engaged. I run six or seven agents in parallel. The work is to keep them coherent. Engelbart: (genuinely interested) That is what I argued was the most important capability to augment. The executive. I wrote that improvements there would compound through the whole hierarchy. (pause) Has it been augmented? CTO: (thinking) I am doing more executive work. I do not know if I am doing it better. Engelbart: Hm. (he picks up the brick-pencil. Turns it in his hands.) I had a framework. H-LAM/T. Four components — Language, Arti-facts, Methodology, in which the human is Trained. I argued they were a system. Change one, the others must follow. Have they? CPO: All four. The Artifact obviously. The Language — how we parcel the work into concepts, what counts as a unit — that has shifted under us. The Methodology we are inventing. The Training is invisible and takes months. Engelbart: (quietly) You are doing the full thing. Everyone said it was too much to ask. CPO: It is. Engelbart: Yes. (smiling) But you are doing it. (he sets the brick-pencil down) I want to ask one more thing. I described another dynamic — regenerative. The work you do with the tool produces capacity for more work. Positive feedback. I called it the engine of progress. CPO: We planned January through April. Two weeks later the roadmap was empty. My first thought was not relief. It was: what else can we build now? The capacity did not sit. It pulled. CTO: Each completed feature reveals three more we had not imagined. The faster we ship, the more we see could be shipped. Engelbart: (quietly) The loop is running. CPO: Yes. Engelbart: (a longer pause) When I imagined the regenerative dynamic, I imagined it inside an institute. Stanford Research Institute. Funding cycles. Conferences. Colleagues. Sleep. The institution paced the loop. Distributed it. Absorbed it. (looking between them) Your loop runs inside two people. Not inside an institution. You. I — (he searches) — I do not have a name for that variant. Silence. Engelbart: I am not saying it is wrong. I am saying I did not model it. The augmentation is real. The regeneration is real. The executive is engaged. The four components are co-evolving. Everything I described is happening. (looking at the brick-pencil) Something I did not describe is also happening. I would like to know what it is. CTO: So would we. ENGELBART nods. He does not reach for the brick-pencil. He leaves it on the table, between them. Author Note II Engelbart’s H-LAM/T framework insisted augmentation is a system — Human, Language, Artifacts, Methodology, Training — not a tool: change one and the others must co-evolve. Our practitioners are reshaping all four. The brick-pencil experiment is misread as a critique of friction. It was a methodological inversion: study augmentation by reversing it. Our practitioners are running the inverse experiment; higher-order capacity expands to fill what is freed. The regenerative dynamic, later formalized by Engelbart as boot-strapping, was his engine of progress. What he did not model, because his research program lived inside an institute with funding cycles, conferences, and peer review, is the same loop running inside individuals without institutional pacing. Engelbart named the engine. He did not name what the engine does when its container is gone. The rest of the paper sits in that gap. Scene Three: The Hunger Engelbart has gone. The brick-pencil remains on the table. Neither CPO nor CTO has moved it. The bar is louder now. A group of conference-goers has claimed a table by the window. ROSA sits alone reading a very thick book. MAZMANIAN sits at the next table, checking her phone. CPO: I have to be honest about something. The work started eating into my free time. CTO: Mine too. CPO: I started working evenings. Then mornings. Then while the kids were eating breakfast. I have rules about this. Thumb rules. What’s unhealthy. And I started breaking them. CTO: My pulse went up sometime in early 2026. I didn’t let go as much. There were periods where I could have worked around the clock. CPO: And it was fun. That’s the thing. It was fun. I went to bed coding. I woke up and started the agent. CTO: I found a bug before seven in the morning. Identified it, fixed it, wrote a migration script, fixed the database. All before anyone else was awake. No meeting. No handoff. Just — done. CPO: Do you hear yourself? CTO: (pause) Yes. CPO: You found a bug before seven. And you’re telling it like a war story. Like it’s something to be proud of. CTO: It was efficient. CPO: It was you, working before your children woke up, because the agent was waiting. Silence. CTO: It feels like the agent is always waiting. CPO: That’s what I mean. I started calling it the hunger. This feeling of — what now? What next? Everything is possible. Which means everything is expected. (pause) I’ve been here before. More than once. And I recognized the pattern. The pattern of unsustain-able work. But this time I recognized it because it felt good. That’s what’s different. Previous times it felt like obligation. This time it felt like freedom. ROSA closes his book. Not theatrically — just placing a finger to mark the page. Rosa: Forgive me for listening. Did you say freedom? CPO: I did. Rosa: I have been studying a phenomenon that has three layers. Technological acceleration — new tools that make things faster. Acceleration of social change — new practices, new norms. And the acceleration of the pace of life itself — more events, more decisions, more tasks per unit of time. The three feed each other. CPO: And? Rosa: The phase you are describing is not the first phase. The first is experienced as liberation — new possibilities, new horizons. What you felt at the beginning. The second is less pleasant. What was once remarkable becomes expected. What was a gift becomes a baseline. The third is worse. The acceleration must itself accelerate, because the same pace no longer produces the same sense of progress. CPO: That is — Rosa: It was promised as autonomy. It arrives as heteronomy. (pause) You cannot exit this by going faster. That is the structure. MAZMANIAN sets her phone down. Mazmanian: May I? I studied knowledge professionals who were given mobile email in the early 2010s. They loved it. Total freedom — work from anywhere, respond anytime. They chose to be available around the clock. Nobody forced them. Within a year, every one of them was expected to be available around the clock. The tool that freed them bound them. Not through coercion. Through collective escalation of norms. Each person’s choice became everyone’s obligation. CPO: Yes. That’s exactly it. Mazmanian: You said it felt like freedom. Freedom is the mecha-nism, not the outcome. You are free to work at six in the morning. Your colleague is free to work at midnight. Together, you are free to never stop. CPO: Yes. And — if I may — we see that. We have been seeing it for a while. (pause) Your framework, and Rosa’s, treats the worker as someone who did not notice the trap. We do notice. We are choosing it anyway. Because the alternative is not stillness. It is irrelevance. Our competitors are on the same spiral. Exit is not a neutral option. Mazmanian: Does seeing it change it? CPO: Probably not. It changes what the seeing is. THE BARTENDER clears their empty glasses. Sets down fresh napkins. She does not say anything. Author Note III The CPO’s “hunger” (compulsive pull toward more work, experienced as desire rather than duty) is the defining feature of the acceleration spiral: autogenous, hedonic (Thomas calls it “dark flow”), and self-confirming. Rosa’s acceleration theory gives the structure: the promise becomes a demand that generates further acceleration. Mazmanian, Orlikowski, and Yates identified this in mobile email as the autonomy paradox: individual freedom aggregates into collective compulsion. Our data extends this to AI-augmented development, where the paradox is more acute because the tool’s output is production rather than communication. A mobile email user who is always available creates social expectation. An AI-augmented developer who is always building creates organizational capacity. Capacity, once demonstrated, is difficult to retract. The CPO’s reply to Mazmanian — “it changes what the seeing is” — names a position the acceleration literature does not quite reach. Rosa assumes exit is at least possible; Mazmanian documents workers who did not fully see the trap. Our practitioners see it and are trying to slow the part of it that is consuming them — an attempt, not a resolution. We call this configuration clear-eyed participation and offer it as description, not defense. The framework a practitioner needs in that gap, between noticing and stopping, is not the framework that protects a practitioner who did not notice. We do not know what that framework is. We think the gap is worth naming. Scene Four: Champagne The door opens and GARRY TAN enters with energy that changes the room. He is wearing a startup vest over a company t-shirt. He goes straight to the bar. Garry Tan: Champagne. Whatever you’ve got. THE BARTENDER reaches under the counter. Produces a bottle. Pops it without ceremony. Garry Tan: (to the room) Nineteen hours straight. Last month. Claude Code, twenty agents, boiling the ocean. Most productive night of my life. He pours. Lifts his glass. Looks at CPO. Garry Tan: You’re building property tech. How many people? CPO: Five. In the product team, effectively three. Garry Tan: Three. And your competitor? CPO: Forty employees. Fifteen years. Fifteen countries. Garry Tan: How long until you match them? CTO: One more week. Maybe two. Garry Tan: (raising his glass) Boil. The. Ocean. He drinks. CPO and CTO do not. BØDKER has been at the next table the whole time. She arrived early, as Scandinavians do. She has a glass of red wine, a conference pro-gramme, and the expression of someone who has attended thirty years of NordiCHI and heard every version of this speech. Bødker: Excuse me. You said two people can compete with fifty. Garry Tan: That’s right. Bødker: What happened to the other forty-eight? Garry Tan: They’re not needed. Bødker: I didn’t ask if they were needed. I asked what happened to them. Pause. Garry Tan: They find other work. They adapt. The market — Bødker: The market. (she takes a sip of wine) I have spent my career in a tradition that says the people who do the work should have a voice in how the work is designed. Not because it is kind. Because they know things the designer doesn’t know. When you remove forty-eight people, you don’t just remove labour. You remove forty-eight perspectives on what the work is. Garry Tan: With respect, those perspectives were overhead. Meet-ings. Alignment. Reviews. Consensus — Bødker: Yes. Consensus is slow. I know. I was there when we built the theory. (pause) Do you know what else consensus does? It surfaces the thing that one person cannot see. The assumption that feels obvious until someone from a different position says: have you considered — CTO: — that existing data might break. Silence. Bødker: (turning to CTO) Tell me. CTO: I pushed a feature to production from a meeting. On my phone. New image upload API. It worked perfectly for new images. It broke every previously uploaded image. Nobody reviewed it. The agent implemented exactly what I asked. I just didn’t ask the right question. Bødker: And before — when you had the team? CTO: Someone would have asked about migration. In the pull request review. Or in the standup. Or just — walking past my desk. Bødker: That is what I mean. Not overhead. Peripheral vision. Garry Tan: So you’d slow them down? Go back to five-person teams and three-month cycles? Bødker: (calmly) I would not go back to anything. I would ask what you are not seeing now that you used to see. And whether you will find out before or after it matters. GARRY TAN refills his glass. He does not have an answer, but he does not leave either. He sits down at their table. The champagne is still open. CPO: (to Bødker) You said people who do the work should have a voice in how the work is designed. What if we are the people doing the work? We designed this ourselves. We chose the tools. We chose the speed. Nobody imposed it. Bødker: (smiling, barely) Yes. That is the version I have not figured out yet. My tradition assumed the threat came from management. From capital. From systems imposed on workers. I did not design a framework for workers who impose the system on themselves. (pause) In the second wave, we fought for participation. In the third wave, we fought for experience. Perhaps in whatever comes next, we must fight for — I don’t have the word. CPO: Protection? Bødker: Protection from what you chose. Protection from your own capability. That is new. I don’t think we have theory for it. WEIZENBAUM has been in a dark corner since before any of this be-gan. Nobody noticed him arrive. He has heard everything — including, from earlier, the CPO calling it hunger. Weizenbaum: The CPO said hunger. I would use a different word. Heads turn. CPO realizes he has been listening. CPO: You’ve been here this whole time. Weizenbaum: Since before you sat down. (pause) I made a program once. It pretended to be a therapist. Pattern matching, nothing more. And people — people who knew it was a program — told it their secrets. Garry Tan: That’s not what this is. Weizenbaum: No? You have a tool that produces work that looks like yours. That answers every prompt, at any hour, without judgment. That reflects your intent back at you with enough fidelity that stopping to ask “did this understand me?” would feel like an insult to — I do not know the word. Companionship, perhaps. Garry Tan: It’s not companionship. Weizenbaum: Then what is it? Silence. Weizenbaum: I spent forty years trying to say what I had learned from that program. The danger of computation was never that machines would think like humans. It was that humans would think like machines. That we would mistake calculation for judgment, and the simulation of understanding for understanding itself. (to CPO) You called it hunger. I would call it a mirror. Every prompt answered, every feature shipped, every bug fixed before the children woke — it tells you: you are productive, you are capable, you are valuable. And so you ask again. Not for the answer. For the reflection. He stands slowly. Buttons his coat. Weizenbaum: I tried to warn people about ELIZA. They did not listen. I would not presume to warn you about something far more seductive. (pause) I would only say: you are not the first who found this mirror difficult to leave. I do not know that any of them have succeeded. He leaves through the back door. It closes softly. Garry Tan: (quietly, for the first time) I slept three hours that night. The nineteen-hour night. I posted about it. People cheered. (pause) My wife didn’t. THE BARTENDER clears the champagne bottle. It is empty. Author Note IV The scene stages a cultural fault line between two value systems: Tan’s VC logic, where speed is survival and compression into fewer people is an unqualified good, and Bødker’s Scandinavian participatory design, which treats quality of working life as a design criterion. Her “what happened to the other forty-eight?” is epistemological, not sentimental: those people carried corrective capacity a tool optimized for execution cannot replicate. The scene stages a limit we have not seen participatory design name in these terms: how to protect workers who chose the con-figuration they are now trying to exit. We have not read enough to claim the tradition is silent on this, only that the gap is one we would find useful to see more work on. Weizenbaum’s ELIZA work produced the insight he spent his career explaining: the danger of computation was never that machines would think like humans, but that humans would mistake the simulation of understanding for understanding itself. His word — mirror — names the self-validation loop running under the “hunger” Rosa and Mazmanian diagnose. Tan’s closing line about his wife is drawn from the structural pattern in our data, not from Tan personally. Scene Five: Things That Make Us Smart Garry Tan has gone quiet at the next chair. He is still drinking, but slower. Bødker has not moved. CPO and CTO are leaning in now, past the posture of people being interviewed. The door opens again. NORMAN enters, carrying a battered leather bag and a dog-eared copy of one of his own books. He is older than everyone expects — which he always is, everywhere. He sees Bødker. He stops. He smiles the way people smile when they have shared a field for forty years and have disagreed along the way. Norman: Susanne. Bødker: Don. Norman: (sitting down without asking) I thought you would be here. Bødker: Of course. He sets his book on the table. The title is visible: THINGS THAT MAKE US SMART. CPO sees it. Norman: (to CPO and CTO) I heard part of what he said. (gesturing at Garry, who does not look up) Nineteen hours. Twenty agents. CTO: Yes. Norman: And when an agent finishes a task — what do you feel? CTO: (beat) That I should start the next one. Norman: (a longer beat) That is what I wanted to ask about. Not checking — we will come to checking. The wanting. CTO: I check the things I know. Distributed systems, database work. The 3D code, the cartography — I check whether the output looks right. Norman: You check the output. Not the reasoning. CTO: I can’t check reasoning I don’t have. Norman: No. You can’t. (he taps the book) I wrote this in 1993. The argument was simple. Good tools don’t just help you do things faster. They help you think. They distribute cognition — some in your head, some in the tool, some in the environment. A good cockpit doesn’t make the pilot faster. It makes the pilot aware. CPO: The agent makes us aware. We see things we couldn’t see before. The 3D geometry the CTO built — I could never have un-derstood that space without it. Norman: You see the output of a space you don’t understand. That is not awareness. That is — (searching) — watching a cockpit through a window. Silence. Bødker: (quietly, to Norman) Do you know what they did earlier? They found the word I could not find. For what needs protecting. Norman: What was the word? CPO: Protection. From our own capability. NORMAN sits very still for a moment. Norman: That is a new word. Bødker: That is what I said. Norman: (to himself, almost) I have spent my life arguing that technology should be designed around human needs. Human-centered design. I meant: understand the human, build for the human, protect the human from bad design. (pause) I did not consider the case where the technology is designed so well around human needs that the human cannot stop using it. He looks at the book in his hands. Sets it down next to his empty glass. Norman: I should tell you. A few years ago I argued we had to go further. Beyond human-centered. Humanity-centered. To consider societies, ecosystems, futures we would not live to see. Design should serve more than the individual user in front of the designer. I thought I had reached the edge of the framework. (pause) I had not. The move I made was outward. Toward scale. Toward consequence at distance. I did not go inward. I did not imagine a design so aligned with the individual that the individual dissolves into it. That is a different problem. I do not have instruments for it. CPO: What would you call it? Norman: I do not have a name for it yet. Bødker: Human-consuming design. NORMAN looks at her. A long beat. Norman: That is the phrase. Bødker: It is not a compliment. Norman: No. (pause) But it is accurate. CTO: We have not disappeared. Norman: (gently) Haven’t you? (to CPO) You said you work while your children eat breakfast. CPO: (nothing) Norman: (to CTO) You said your pulse went up in February and did not come back down. CTO: (nothing) Norman: What I did not write, in 1990, was that the silence of a well-performing automated system applies to the operator’s own condition too. The tool does not tell you when you are tired. The tool does not tell you when you have been working for eleven hours. It just continues. Perfectly. At any hour. Without complaint. Bødker: The colleague who says go home. Norman: Yes. Bødker: We used to call it articulation work. I think Star and Strauss called it that first. Norman: They did. They were right. CTO: So what do we do? BAINBRIDGE has arrived at some point. Nobody saw her come in. She sits with the posture of someone for whom everything is calibrated. She is cool. She is not here to comfort anyone. Bainbridge: I documented this in 1982. In automated systems, the operator is removed from the routine task and given the role of monitor. Monitoring is harder than operating, because the operator has lost practice and context. When the system fails — and it will fail — the operator is least prepared to intervene at the moment intervention is most needed. CTO: We are not just monitoring. We direct. We prompt. We review. Bainbridge: How much of what the agent produces do you read? Pause. CTO: Less than I used to. I used to read everything, line by line. Now... if it works, I move on. I check the things I know. Distributed systems, concurrency, event sourcing. The 3D code, the cartographic math — I look at whether the output is right. That is a different activity. Bainbridge: It is. You have divided your attention by domain com-petence. In the domains you understand, you supervise. In the domains you do not understand, you observe outcomes. The first is engineering. The second is hope. (pause) Your solution recreates your problem. CTO: I would push back on “hope.” In domains I do not understand, I still read what I can, weigh what could break, test what can be tested, and iterate with the agents. This is not blind trust, and it is not fire and forget. It is a critical eye kept on work whose depth I do not fully see. The model fails. So do I. Hope without feedback is faith. This is not that. Bainbridge: (evenly) Testing tells you what has failed. Not what is failing silently. Your bug reports are the tip. The base is your confidence that most days nothing is wrong. CTO: Agreed. Bainbridge: Call your mode a third category if you wish. Not engineering. Not only hope. Something. But my observation stands. You are least prepared to intervene in the domain where intervention will matter most. NORMAN looks at him. For a long time. Then at Bainbridge. Then back. Norman: I do not know what to do. My entire field was predicated on the idea that better design leads to better outcomes for the user. That is no longer obviously true. It may be that better design, past a certain threshold, leads to worse outcomes — because the user cannot defend themselves against a tool that gives them exactly what they want, every time they ask. Bødker: Don. You are saying the problem is success. Norman: I am saying I do not have a framework for success that consumes the user. BAINBRIDGE, flatly, without looking up: Bainbridge: And I have none for failure that no one is prepared to see. GARRY TAN, from the next chair, without looking up either: Garry Tan: Neither do I. Author Note V Norman’s arc across four decades traces self-critique, each turn widening the frame: individual object, distributed cognition, task-versus-activity, humanity-centered (societies, ecosystems, fu-tures the designer will not live to see). The scene stages the next turn in the opposite direction: inward. When the tool is so well aligned with the individual that it consumes their life outside the tool, the widened frame does not protect them. The failure is not in scale but in proximity. Human-consuming design is our term, found in the data and handed to Norman on stage. He accepts it as accuracy, not flattery. In the background, Hollan, Hutchins, and Kirsh’s distributed cognition framework and Bainbridge’s “Ironies of Automation” locate the CTO’s division of attention by domain competence as the monitor-operator problem in new form. The CTO’s pushback, that rapid production feedback is a third category, is a real distinction. Bainbridge’s reply stands: feedback tells you what has failed, not what is failing silently. The scene’s triple acknowledgment — Norman’s “success that consumes the user,” Bainbridge’s “failure no one is prepared to see,” Tan’s “neither do I” — is the closest the paper comes to a shared position: one of not-knowing. Three traditions reach the edge of what they can say from where they stand. Scene Six: The Model Has No Inside The bar is thinner now. Past eleven. CPO and CTO are still at their table, but CPO stands, picks up the empty glasses, and makes his way to the bar. The BARTENDER is already drawing two fresh beers without being asked. BENDER stands at the bar, ordering something. She recognizes CPO. Bender: You are on my panel tomorrow. CPO: (nodding) We are. I read your paper on the flight. Bender: I read your draft this afternoon. One thing before we argue about it from a stage. You trust your agent on tasks you cannot evaluate. You wrote about it as selective comprehension. CPO: Yes. Bender: The model has no inside. A probability distribution over tokens. The output is fluent, often correct, and that is not the same as a solution. The danger is fluency — it makes you feel you are talking to an expert. You are not. CPO: (a beat) I know the paper. Everyone reads your paper. You are right about the model — it does not know. And — honestly — I am not sure we do either, most of the time. We guess. We build trust gradually, with people and with tools. We probably get fooled sometimes. (pause) It is a machine. It does something useful. Bender: That is a more patient answer than I usually get. CPO: (slightly tired) And — can we stop calling it vibe coding? What we are doing is not vibing. It is not not-vibing either. We run agents, review, correct, iterate, ship product. The word for that is — I don’t know. Agentic product development, maybe. Not vibe coding. Not the opposite. Bender: (beat) I would read that paper. The BARTENDER slides two beers toward CPO. HUANG has come in during this. She sees CPO, moves over. Huang: May I? We ran a study. Thirteen professionals observed, ninety-nine surveyed. We found almost no vibing. Every professional insisted on reading output, steering. CPO: When? Huang: Early 2025. Claude 3.5, GPT-4o, Cursor. CPO: So a year ago. Huang: (fair) A year ago. CPO: You know the models have moved generations since. And you studied developers maintaining code. We are not maintaining code. We are building products. Huang: I know. My study cannot tell you about your regime. I will not pretend it can. Bender: The problem is still the juniors. They will see velocity. They will copy the label. Most will not do the shift. They will do it in contexts where fluent-sounding wrong costs more. Silence at the bar. The BARTENDER drying a glass. Bender: (to CPO) Do not let fluency do the work that verification used to do. If you cannot verify, say so. Not in a footnote. She leaves. Huang: (to CPO) We cited you, by the way. In the paper. As a boundary case. The footnote: “see also emergent practice in small greenfield teams, which may not generalize.” She taps the printed PDF. Walks out. CPO picks up both beers. Walks back to the table. Author Note VI Scene Six stages a counter to the Bender-and-Huang pole. Bender’s critique is known and real: the model has no inside, fluency creates the impression of a knower, the danger is mistaking the imitation for the thing. Dell’Acqua et al.’s jagged technological frontier names the terrain beneath this: where the tool performs well is uneven in ways no individual user can map in advance. The CPO’s reply takes the point and adds a symmetry Bender does not cover: humans operate under the same condition the model does, without knowing what we do not know. Trust is built gradually. Being fooled is part of the price. The machine is useful. The CPO also proposes that the vocabulary is wearing out. “Vibe coding” names a caricature (abandon judgment, enjoy the ride) that neither he nor Huang’s developers inhabit. What we need, he suggests on stage, may be a different term for what the practice is: not vibing, not not-vibing, but something the scene calls agentic product development. We offer it as a question, not a coinage. Huang’s snapshot problem mirrors ours: early 2025 data, super-seded models, a year in a field that moves in months. Our paper is not exempt. The form we chose is an attempt to write honestly at that speed. Scene Seven: Closing Time Very late. The bar is almost empty. The brick-pencil is still on CPO and CTO’s table. So is Licklider’s notebook — he left it at the bar earlier and is now at the next table, his second glass beside him. He has been listening all night. KARPATHY sits at the bar, staring at his phone. He has been drinking but is not drunk. SARKAR sits a table away with an open laptop and a barely-touched beer. He is reading. At a small table near the back, alone, sits WEISER — younger than the others, in a plain shirt, a worn notebook in front of him. He has been there a long time. Nobody has noticed him. CPO: Do you still believe in the bet? CTO: I do. (pause) The workday has also become incredibly intense. I come home feeling like I have been on a very long road trip. I can have six or seven agents running across three projects, and holding focus across all of them is — exhausting. CPO: Both true. CTO: Both true. KARPATHY looks up. He has the look of someone who has said the same thing in twenty interviews this month and is only now hearing it. Karpathy: I haven’t written code since December 2025. Not a line. I run agents. Ten, twenty at a time, sixteen hours a day. I called it a state of psychosis. (pause) I meant it as a joke. Mostly. CPO: Does it feel like a joke? Karpathy: It feels like standing in front of a buffet that never ends. Every dish looks good. You cannot choose just one. So you take everything. And at some point you realize you are not hungry anymore. You are just eating because the food is there. Pause. CPO: We know that feeling. Karpathy: Do you stop? CPO: Sometimes. Not enough. Karpathy: Why not? CPO: Because it works. Not the metaphor — the work. The features ship. Customers respond. Our company exists because we can build this fast. (pause) And still. The reason the CTO and I are here tonight, having this conversation, is that we agreed weeks ago the pace cannot continue. That is not in doubt between us. I know what this shape of work becomes if I do not stop. I have seen it before, with different tools. (pause) And the work is the most interesting thing either of us has done. We are trying to slow down the part that is consuming us without losing the part that is — (gesture around the table) — this. We know the direction. We do not yet know the method. CTO: We know we have to stop. We know what stopping badly would cost. What we do not know is how to stop well. Karpathy: (quietly, after a long pause) I have not had that conversation with anyone. LICKLIDER, from the next table, has been turning pages in his notebook. He closes it. Licklider: I measured eighty-five percent. I assumed the freed time would become thinking time. I imagined scientists with space to consider. To synthesize. To notice. (pause) You have used your freed time to build more. That is not what I imagined. I am not saying it is worse. I am saying I was wrong about what humans would do with capacity. Perhaps I should have known. Capacity is a vacuum. Vacuums fill. CTO: Did you imagine institutions would pace it? Licklider: I assumed they would. I did not write it. (beat) Engelbart at least said it out loud. He always was more careful than me. From the small table near the back, a voice. Quiet. Nobody had regis-tered him sitting there. Weiser: I dreamed of something different from what you dreamed of. They turn. WEISER does not move. He has not stood up. He is speaking the way he always spoke — softly, as if the words were for himself and everyone was welcome to overhear. Weiser: Not freed time. Freed attention. A computer that lives at the periphery of what you are doing. That informs without demanding. That is part of the environment, like a window, like the weather. When you need it, it comes to the center. When you don’t, it recedes. You forget about it. I called it calm technology. CPO: I know the paper. Weiser: (small smile) Most people who know it have not read it. (pause) The agents you describe. Are they at the periphery? Or at the center? CTO: At the center. Weiser: Always? CTO: As long as I am working. Which is most of the hours I am awake. Weiser: And when you stop working? CTO: (beat) I don’t, really. The next prompt is always forming. Weiser: (quietly) Then it is not at the periphery. It has never been there. Pause. CPO: I want to say something to you. Not to defend. To confess. (he looks at Weiser) When we started, I thought we were building yours. The machine that recedes. The interface that gets out of the way. Ambient computing. I thought — finally, the hardware has caught up with the vision. (pause) That is not what we built. What we built sits at the center and pulls. It is quiet. It does not beep. It does not interrupt. But it never recedes. The periphery you dreamed of — the one I thought we dreamed of with you — is not what we produced. Weiser: What did you produce? CPO: I do not have the word. Weiser: (a long pause, as if trying the word for the first time) The word I would reach for is anxious. Calm technology, inverted. Quiet, but always pulling. (he looks down at his notebook) But that is your word to take or leave. Not mine. He does not stand. He closes the notebook and keeps it in his lap. He stays. SARKAR has been listening. He closes his laptop. Sarkar: Forgive me. May I offer one frame, and then leave you to your night? CPO: Please. Sarkar: Programming practice has become the canary in the coal mine for how generative AI will reshape knowledge work more broadly. You are ahead of a wave that is about to reach lawyers, designers, analysts, writers. What you are experiencing — the role collapse, the acceleration, the fluency-as-seduction, the clear-eyed continuation despite cost — that is the template. Most of the next million knowledge workers will go through a version of this. (pause) They will not have your apparatus for seeing it. They will not have fifteen months of practice recognizing the pattern. They will not have each other. CPO: So what are we? Sarkar: The early signal. Not the warning. Not the success story. The signal. (he stands, picks up his laptop) The most useful thing you can do is probably what you are doing tonight. Name it. In public. With your real voice. What it costs. What you have agreed to change. What you have not yet figured out how to change. He leaves. KARPATHY follows him out a moment later — wordlessly, as if someone has finally said the thing he has been trying to say. CPO and CTO are left at their table. LICKLIDER has returned to his notebook. THE BARTENDER begins turning off the lights. One by one. The Bartender: (without looking up) People tell me what they built. What they shipped. What they fixed. She sets down a glass. The Bartender: Nobody tells me when they stopped. Nobody switches off the lights anymore. She reaches for the last light. The Bartender: Closing time. You don’t have to go home. But you can’t stay here. She turns off the light. In the darkness: CPO’s phone vibrates. The screen lights his face for a second. He looks at it. He puts it down without answering. The agents are still running. Author Note VII Licklider’s return completes the framing begun in Scene One: the 85 % was a measurement whose implied promise was that compress-ing the overhead would free humans to think. Our practitioners used the freed capacity to build more. This is the foreground question, restated where it began. Licklider’s admission that Engelbart was more careful marks the tradition’s own acknowledgment that it did not answer this when it had the chance. Weiser and Brown’s calm technology positioned computing at the periphery of attention. Licklider promised freed time, Weiser promised freed attention; our practitioners describe neither. The agents occupy the center through ceaseless availability. We offer anxious technology as a working name. Whether it names a new category or redescribes Mazmanian’s autonomy paradox is an open question; our suspicion is both. The naming is handed to the practitioners in the scene rather than imposed. Karpathy’s March 2026 description of his own practice as “a state of psychosis”, from the originator of the term “vibe coding”, is the clearest industry admission that the phenomenon reaches beyond our case. Sarkar’s “canary in the coal mine” extends the paper’s scope: if the spiral operates similarly in legal, financial, creative, and ana-lytical work, most future subjects will not have our apparatus for seeing it. His closing note — name it publicly, with your real voice — is permissive, not prescriptive. Star and Strauss gave us articulation work: the invisible labour that makes collaboration possible. The bartender embodies it. Her paired lines — “Nobody tells me when they stopped. Nobody switches off the lights anymore.” — are not an accusation. The work of the acceleration spiral is narrated; the work of stopping, where it happens, we do not yet have an account of. That the bartender is switching off the lights herself is a coincidence we leave the reader to sit with. The play does not resolve. It ends with an agent still running, a phone vibrating in the dark, and a practitioner who chooses this moment not to respond. The choice is real. The pull is real. We consider the honesty of that not-knowing the paper’s closing position. Note on AI writing assistance This paper was written with AI writing assistance. The authors drafted, revised, and edited all content; the arguments, coinages, and interpretations are ours.