On Day 2 of SusHi Tech Tokyo 2026, on the Main Stage, five Japanese tech leaders staged an Oxford-style debate on a single proposition: Truly effective A.I. replaces the need for human workers. When the votes came in, the audience — most of them paid attendees of a flagship Tokyo tech conference — sided 71% against the motion and 29% for it.
That ratio is the most interesting thing that happened in the room. It is also, I think, wrong. Not wrong as a forecast — forecasts about the future of work are mostly cope dressed up as analysis. Wrong as a signal about the audience itself.
Let me walk you through what was said, who said it, and the part nobody on stage probed.
The Setup
Issa Yamada, who runs Women in Tech Japan — a non-profit working to put 5 million Japanese women and girls on STEM tracks by 2030 — moderated the debate. She framed the format up front: this is Oxford-style, which means speakers are assigned positions, not their own. The goal isn't to win, it's to stretch the thinking. Keep that in mind. Two of the four debaters were arguing the opposite of what they actually do for a living.
On the PRO side, arguing that AI replaces humans:
- Shiho Watabe, chairperson of Shibuya Startups, the community organization that connects Tokyo's founders to capital and customers. She spends her real days helping humans build companies. On stage, she had to argue that those humans were obsolete.
- Leo Murashik, CEO of One Drops, with a background that runs through Japan's defense academy and into European and US tech work. He delivered the debate's most original argument.
On the AGAINST side, defending the human:
- Asuka Kuabara, founder of Anosa AI — a company that makes AI-powered conversational toys that talk to children — and an AI business consultancy that helps executives restructure their organizations around AI.
- Kaido Chiata, Japan Country Manager at Hello Clever, an AI-powered payments fintech expanding into Japan. He argued the regulatory case — accountability is something only humans can hold.
The Economic Case
Shiho led with the data, and it stacks up. WEF projects 92 million jobs displaced by 2030. Goldman Sachs estimates 18% of global work is automatable today, and up to 70% of administrative and support task work. Pew's research is the one that should rattle the room: it's the high-degree, high-pay workers who are most likely to be replaced in the current wave — not blue collar. The replacement curve is moving up the income ladder, not down it.
Executive sentiment confirms it. Two years ago, 33% of executives surveyed saw AI as their primary cost-cutting tool. Today it's 50%. That's the shift that matters — not whether AI can replace humans, but whether the people who write the checks have decided it should.
Then Shiho closed the trap. The narrative we tell ourselves is that knowledge workers are at risk first and blue collar is safe — physical work is harder for machines. Gartner says warehouse smart-robot adoption is up 50% by 2026. Agriculture is being eaten by AI-piloted tractors. The blue collar replacement is also accelerating, just on a different curve.
Her thesis: "Growing AI is commoditizing intellectual and physical execution. This is not a choice. It is an economic inevitability driven by capital efficiency."
The Effort-Replacement Thesis
Leo's claim, in his own words: "AI is not just replacing our work but the point is that AI is replacing our effort ability to make a effort."
His version of the negative cycle: AI is convenient, we rely on it, we develop our own capability less, we depend on AI more. Combined with the fact that "AI does not die, it doesn't forget, it can share memories with each other" — once humans atrophy, the option of doing the work ourselves disappears. The displacement isn't gradual. It isn't recoverable.
This is a darker argument than the standard substitution case, and it wasn't rebutted on stage.
I want to push back on it. In my own work, AI hasn't eroded my thinking — it has stripped out the parts of my workflow that were never the point. The structuring of an argument, the asking of a counter-question to find my blind spots, the converting of a half-formed thought into a sentence somebody else can read — those are still mine, and they get sharper when I have a tool that argues back. What's gone is the part where I used to lose 40 minutes formatting an email or rewriting the same intro paragraph six different ways. That wasn't a skill. That was a tax on having ideas.
Leo's thesis is right about a real population — people for whom the cognitive work was the rewriting, was the formatting, was the low-effort repetitive task. For them, AI removes both the work and the practice ground. The dentist who buys an automatic check-out machine doesn't lose dental skill. The middle manager whose entire job was structuring memos for a leadership team loses both the memo and the muscle.
The honest framing is: AI atrophies the skills that were already on the way out. It enhances the skills that were always the point. The hard part is being honest with yourself about which of those describes your actual job.
The Three Pillars and Why Two Are Failing
Asuka offered the standard humanist defense, and I want to take it seriously because most of the audience clearly did.
Trust. You don't go to a doctor because they know everything. You go because you trust their judgment. Trust attaches to the person, not the information.
Ambiguity. When the situation is genuinely novel and stakes are high, AI hallucinates. Humans with real context outperform.
Responsibility. "You cannot take responsibilities as an AI for an action that is made by AI." Somebody has to own the call.
This is the framing that comforted 71% of the audience. And here's the part to name plainly: those three pillars describe a shrinking category of work, not the median knowledge job. Trust, deep ambiguity, and bearing accountability — that's the top 10 to 15% of any profession. The median lawyer is not the lawyer the client picks for their hardest case. The median doctor is not the diagnostician on the genuinely novel symptom. Most knowledge work is, at the median, exactly the kind of structured, repeatable, intermediate-stakes output that AI does today.
You can defend the top of the pyramid. The bottom and middle do not have a clean answer in this framing, and Asuka didn't try to give one.
The Contradiction Nobody Probed
Asuka founded an AI company that talks to children — which is, in part, a substitution for human caregiving and early education. She runs an AI consultancy that helps companies restructure operations around AI — which is, in part, the consultant version of helping companies cut headcount. Then she argued, on stage, that AI will not replace humans.
The moderator did not probe this. The audience voted 71-29 anyway.
The honest read is not that Asuka is a hypocrite. The honest read is that operators build for the world that's coming, and they defend the world that's here. You raise capital on the disruption story to your investors and you reassure the room of incumbents that nothing fundamental is changing. Both can be true at the same time, and the right move for a founder is often to hold both. It's not dishonesty. It's a working hypothesis about timing.
The takeaway for anybody else in the room — anybody not an AI founder — is that you can't read what the founders say. You have to read what they build. Asuka's company is the answer to what Asuka thinks. It's not the line she delivered into the microphone.
The Cultural Moat Is Dissolving
This was the most strategically important moment of the debate, and almost nobody caught it. Somebody in the audience asked about Japanese cultural nuance — wouldn't that make AI deployment in Japan slower? Leo's response:
"What if AI understands the psychology of Japanese people, organizational culture, cultural reasoning behind, and translates everything into universal language? Then Japan might also lose the advantage of being not understood by people new to Japan."
The implication inverts the standard arbitrage narrative. Japan's organizational opacity has historically been a moat that protected Japanese companies from foreign competition. AI is dissolving that moat. Foreign actors with strong AI tools may now navigate Japanese organizational culture better than the Japanese can articulate it themselves, with consequences for domestic competitive advantage.
I want to push back on this one too, because I operate across both markets and the actual lived experience is messier.
You can use AI to map Japanese organizational culture — the structure of who reports to whom, when honne and tatemae come out, why the meeting before the meeting is the meeting that matters. AI can read every public article ever written on Japanese business culture and give you a coherent synthesis. What it cannot do is be in the room when the senior person you offended ten minutes ago decides, silently, to stop returning your calls for a year. That's not in the training data. That's learned behavior and first-hand experience, and AI in 2026 still gets lazy with answers, shortcuts where it shouldn't, and confidently delivers incomplete or wrong information that it stitched together from articles online.
So Leo is half-right. The map is open to anybody. Walking the streets still requires a body in Tokyo.
The arbitrage opening that's real, in my read, is at the first contact layer — the part where a foreign founder needed two years of cultural immersion just to write the first cold email correctly. AI compresses that to a week. After first contact, the messy stuff is still messy, and the people who already lived in Tokyo for ten years are still ahead.
What Wasn't Debated
The room skipped the things that actually matter.
Nobody discussed which jobs disappear first and what specific people should be doing today. The advice from both sides was hand-wavy — "be creative," "find new things." Creativity, by the way, is exactly what generative AI is best at. The advice was incoherent on its own terms.
Nobody discussed the geographic sequencing. AI replacement happens fastest where wages are highest. US first, then Japan, then Vietnam. Japan has a 2-to-5-year buffer relative to the US, which means Japanese knowledge workers have a window to adapt that American knowledge workers don't have. This is a real thing. Nobody mentioned it.
Nobody discussed the transition mechanism. Shiho name-checked Universal Basic Income / Universal High Income once, and the debate moved on. The entire question of how a society moves through structural displacement was treated as a footnote. It's not a footnote. It's the substance.
Nobody discussed re-skilling cost and timing. A 45-year-old middle manager doesn't have five years to retrain into something else. Who pays for the adaptation, and who falls through? The room did not want to look at that question.
What History Actually Says
Here's the thing I would whisper to anyone reading this. Forget the conference rhetoric. Look at the historical pattern.
A few hundred years ago, in any reasonably sized town in Europe, there was a job called lamplighter. The lamplighter walked the streets at dusk lighting the gas lamps and walked them again at dawn putting them out. There was another job — a knocker-upper — whose entire profession was waking people up in the morning by tapping on their bedroom windows with a long pole. Today, the alarm on your phone does the second job for free. Timed switches and motion sensors do the first.
Picture the lamplighter in 1850 trying to imagine a future where every person has a small device in their pocket that wakes them up reliably. It would have been absurd. They needed waking up now — why wouldn't they need it tomorrow?
That's the position the 71% in the SusHi Tech audience is in. They cannot picture the version of the future where the structural need for their work is gone. Not because they're stupid. Because human beings are bad at imagining the next discontinuity from inside the current one.
The 71% also has an additional bias: it's a self-selecting audience. People who can afford the SusHi Tech ticket, the conference clothes, the train into Tokyo, and the intellectual curiosity to spend a Tuesday afternoon at a debate — these are not median knowledge workers. They are the top slice. They've been winning for a long time. Their pattern-match says we won yesterday, we'll win tomorrow. That heuristic has worked for a long time. It is going to stop working soon.
History also says something more useful. Every technological transition rewards the people who learn the new tool early and punishes the people who defend the old role. The blacksmiths who learned to repair cars when horses gave way to engines kept eating. The blacksmiths who insisted that real men still needed to shoe horses did not. The office workers who learned to use the spreadsheet kept their jobs through the typewriter-to-computer transition. The ones who treated the computer as a fad did not. The people who picked up the smartphone and learned to do their work on the go are now the people who can earn remote-work income from anywhere on the planet.
The blue-collar version of this story has already happened. Automotive manufacturing in the US went through it in the 80s and 90s. Whole industrial towns lost their middle class. The economists' answer was always they should have retrained. The lived experience was that the iterative cycle of humans is not as fast as the iterative cycle of an interchangeable, reprogrammable machine. People got left behind. Not because the economic argument was wrong, but because it was too slow to be useful to actual humans living actual lives.
The AI wave is the knowledge-worker version of the same story. It is happening on a faster curve. It is going to feel personal in a way the auto-manufacturing wave never felt to people in offices, because this time it is in the office. And the comforting framing — trust, ambiguity, responsibility — is going to be the lamplighter saying people will always need someone to light the lamps.
The move is not to argue with that framing on stage. The move is to figure out where the time is actually going. To learn the skills that will be valuable in five years and that very few people are currently learning. To assume your current role is, on a long enough timeline, a lamplighter role — and to start the conversation about what comes next while you still have the job.
That is not the conversation the room wanted to have on Day 2 in Tokyo. The 71% vote was a vote against having it. The 29% knew better. The 29% is the audience that should be writing the next decade.