Money, Minds, and Machines: What Two Days in Zürich Told Us About the Convergence

Notes from the inaugural CAIRO Forum, Zürich, 7 to 8 May 2026

My Uber driver was, fittingly, Egyptian. He was also deeply confused. I had asked for an old aircraft hangar at the edge of town, the one where all the robotics spinoffs now reside, and he kept gently explaining that if I wanted Cairo, I needed the real airport. He was not the last person that day to be thrown by the name. CAIRO stands for Cryptography, AI, and Robotics, and it was the first conference to bring all three onto one stage.

There is a logic to holding it here, and it is not an accident. Switzerland built Crypto Valley around Zug before most countries had a word for it. It quietly became a serious place for AI. And now the same small country is leaning into the robotics revolution: three waves, one country. CAIRO exists because those waves are no longer separate. They are merging into one, and most of the people who can see it coming were in that hangar.

Two days. Roughly twenty speakers. Here is what was actually said.

Day One: The Body Learns to Think

The keynote that set the tone

Davide Scaramuzza opened with twenty-two years of robotics in half an hour, and sensibly did not start with humanoids. He started flying a helicopter without GPS in 2009, the same line of work that later helped NASA fly the Ingenuity helicopter on Mars, which carried a camera and a small computer.

Then the part everyone quotes at dinner. Seven years, forty researchers, twenty million simulated crashes, one drone that beat the human world champion at racing. First recorded case of an AI beating a human in the physical world, published in Nature. But his real message was the opposite of the obvious one. The AI was already ten times faster at reacting than a human. Speed was never the gap. Humans win because they make better decisions when things get messy. A human driver errs roughly once every hundred million kilometres. That is the bar, and it is humiliatingly high.

His punchline: physical AI is robotics with one new ingredient, common sense. Pour water into a glass next to a laptop, and a human instinctively does not. Foundation models can now reason about that. Last year, they could not. That is the whole revolution in one spilled drink.

Who gets to be trusted?

Micha Roon of The Hashgraph Group then changed the channel, and the room had to find the remote. His opening fact is the kind that ruins your afternoon: machines already outnumber us on every network, comfortably. We obsess over verifying humans. We have barely begun verifying machines.

You can see a BMW badge on a robot, he pointed out. That tells you precisely nothing about who built it. Authentication still assumes a human with a password and a phone, which collapses the moment one person spawns a thousand agents that all impersonate them. His line, delivered as a verdict: the only currency of trust is cryptography. Not the Bitcoin kind. The signatures-and-hardware-attestation kind. He also noted that the European business wallet becomes mandatory in 2027, which, in business time, he said, is roughly tomorrow.

Restoring movement by reading thought

Julien Camisani of ONWARD Medical drew the pre-lunch slot and used it to silence a room full of crypto people quietly. Around 650,000 people live with spinal cord injury in the US and Europe alone, nine million worldwide, with almost no real alternatives and a lifetime cost that can run into millions per person.

His company stimulates the spinal cord electrically. The newer work wires a brain-computer interface into it: the patient relearns how to think about moving, the system decodes that intention from 64 channels and turns it into more natural movement. There is a video of someone with paraplegia walking. The best part was his refusal to oversell it. Early, he kept saying. The model does not generalise yet, the sessions are short, and the data is precious. In a field full of Neuralink theatre, honesty was the flex.

The afternoon, which belonged to robotics

After lunch, a small piece of theatre. "Kristen Shen" delivered a sweeping talk titled Money, Minds, and Machines: Field Notes From the Convergence Decade. Tidy history of crypto, tidy history of AI, three threads of robotics ending with a humanoid running a half-marathon faster than the human world record in April 2026.

Then the reveal: Kristen Shen does not exist. The whole talk was AI-generated, voiced through ElevenLabs, based on a prompt I wrote. Slightly unnerving, mostly because it was good. Her frame survived not being a person. Three nested problems: productivity, because rich countries are running out of working-age people; coordination, because machines need to transact and crypto is the only technology built for non-human actors; and meaning, the hard one, about a dignified life when machines do the work. And a sharp geopolitical point. The US owns the cognitive layer. China controls the physical layer, selling a humanoid for under $5,000. The EU's chosen role is to regulate, and in choosing to regulate first, it is, by the same logic, choosing not to build. The layers are being claimed by different players, and Europe claimed the rulebook.

Adam Feiler followed with Minima, which turned out to be the spine of the whole event. Machines must trust data blindly to make real-time decisions, and today that trust runs through centralised cloud servers built for a world with far fewer machines. He was rude about the incumbents in the most precise way: a handful of pools control most of Bitcoin's hash rate, and Ethereum does the same through its validators. Minima's answer is a Layer 1 that's light enough to run a full validating node on a drone, a car, or, eventually, a chip. No miners, no validators, no fees. His demo said it in one breath: a drone, a sensor, data validated on-chain, an NFT certificate authorising the flight. Trust on the edge, where integrity is intelligence.

Yulia Sandamirskaya of ZHAW then said the quiet thing loudly. European robotics is being overtaken left and right, she said, like being passed on a German motorway. Chinese robots move beautifully, US robots gesture convincingly, and almost none of them are safe to touch. She tried to shake a robot's hand and lost. She was told to keep her distance from a dancing robot for safety reasons, which is a sentence that should worry everyone. Her bet for the missing piece: stop forcing one giant neural network trained on data that does not exist for robots, and copy the brain instead. Event-based cameras that only see change, neuromorphic chips with research numbers like 100,000 times the energy efficiency on some workloads. And build these robots with care homes from day one, not in a lab and then drop them on the elderly like a surprise.

Luka Živković closed the robotics block from Belgrade with the least glamorous and most useful talk of the day. Agremo runs AI on drone and satellite imagery to count plants, distinguish weeds from crops using their spectral signatures, and track plant health. The punchlines are numbers: variable-rate spraying that cuts chemical use by 20 to 70 per cent, and a Hungarian wildlife-damage case where the AI estimate differed from manual sampling by 12 to 21 per cent, a gap between a fair insurance payout and a fight. This is robotics once it stops being a demo and starts feeding people.

The day ended with two builders. Kristóf Floch presented the ETH Robotics Club, which is not a hobby club: six flagship projects, including a fully Swiss-made humanoid, a drone-racing team that already cleared its first qualifier, and the first student club anywhere to land a Tesla partnership. Then Justinas Mišeikis of Sony AI delivered the keynote that made the executives shift in their seats. 67% of senior leaders call physical AI game-changing. Roughly 0% of boards can describe a physical AI strategy. That asymmetry is dangerous, he argued, because physical AI is an expensive marriage of hardware and software with real consequences, not a software patch. Strict short-term ROI thinking kills it, since the necessary homework never shows a direct return. His closing question to every organisation in the room: shaper, or consumer, trying to catch up?

Day Two: The Money Catches Up

Claudio Tessone of the UZH Blockchain Center opened with the governance keynote, and his framing was economic, not technical. Agents cannot coordinate, share reputation, or rely on common institutions, so they cannot exchange value. They need a digital institutional frame: ownership, dispute resolution, governance. Blockchains are the substrate, and DAOs make the governance programmable. His sharpest line was about law, not code: recognising blockchains as a settlement layer and stablecoins as real payment instruments is a huge legal leap, and success looks like people using blockchains without noticing they are.

Oleg Lavrovsky stepped in for Imanol Schlag to present Apertus, the Swiss AI Initiative's open model, and this was sovereignty made concrete. Two open models, 8 and 70 billion parameters, 15 trillion tokens, over 1,800 languages, 40 per cent non-English data, Apache 2.0 licence, a 119-page report on how to build the thing, hydropowered data centre. The motivation was never benchmark glory. It was a legal necessity, autonomy, scientific integrity, and the ability to run your own AI internally, no vendor, no rising subscription, data kept inside.

Then the applied crypto block, and Bill Lee, online from Hong Kong, quietly stole it. DualMint takes the cash flow from deliberately boring machines, laundromats, HVAC units, robotic vertical farms, and turns it into on-chain yield, settling in USDC every month. After years of DePIN promising a glorious machine economy and delivering inflated tokens, his innovation was almost subversive: make the machine economy real by making it dull. No 100x, no narrative, just a non-correlated, audited 13 to 15 per cent yearly yield from machines that actually exist and actually earn. He has effectively built the plumbing that the whole sector kept sketching on whiteboards, a working bridge between physical machines and on-chain capital. Boring, he called it, and meant it as the highest compliment.

Hima Abed of Bybit Europe gave the exchange-side view: the financial paradigm shift is real; crypto is the driver; fintech and crypto are collapsing into one space; and prediction markets on exchanges may be the next quietly enormous frontier.

That set up the panel, Paying Tomorrow: The Future of Money, moderated by Svetlana Sailer with Hima Abed of Bybit, Nicola Plain of Zühlke Group, and Nicole Kälin von Euw of sumixx. The shared conclusion was refreshingly unromantic: there is no single correct model yet, and tomorrow's winners will be whoever combines regulation, liquidity, UX, AI, tokenisation, payments, and institutional trust. Not one pillar. All of them, simultaneously, which is exactly why it is hard and why nobody has done it yet.

The afternoon turned to applied AI in health, and Genorare was the standout. Quy Vo-Reinhard's premise is that genetic data is the one thing you cannot un-expose; once it is out, it describes you forever. So their model never sees it. Using fully homomorphic encryption, the AI analyses the data while it remains encrypted, with federated learning across hospitals and doctors signing off on the system's capabilities. This is the cleanest answer of the forum to a question everyone keeps dodging: how do you let AI learn from the most sensitive data on earth without ever actually handing it the data? Currently, it is only 14 to 24 times more expensive than normal methods, which hardware will compensate for. Real pilots in Vietnam and France, where the data cannot legally leave, so the model travels to the data instead.

Son Nguyen continued from Ho Chi Minh City with dHealth Intelligence: the AI agent as the brain, the "blockchain of things" as the nervous system, a chronic-care agent that carries its own wallet and fights the insurance paperwork so the patient does not. Marc Ballandies closed with the Swiss road to DePIN and the best analogy of the two days. The Gotthard tunnel was never just a tunnel. Financing it seeded the banks, engineering it built ETH's reputation, and electrifying it built the industrial giants. One infrastructure vision created a whole set of institutions. DePIN, he argued, can do the same, and the current die-off of badly-designed projects is healthy, not tragic. Switzerland always commits late, then commits with force.

What Was Actually New

Robotics crossed from impressive to capable. Not better motion, machines have had that for a while, but common sense from foundation models, the first verified case of AI beating a human in the physical world, and the honest admission from the people closest to it that the real frontier is working safely next to you, not dancing more convincingly.

AI stopped being a black box you ship your data to. Apertus showed you can run a serious model yourself, fully open and documented, with no vendor and no data leaving the building. Genorare showed you can let a model learn from the most sensitive data on earth without ever actually handing it over. Different problems, same shift: from renting intelligence on someone else's terms to owning it on your own.

Crypto got boring on purpose. The frontier conversations were not about price. They were about who a robot is, whether its data can be trusted, and how a drone authorises its own flight. Bill Lee made the machine economy real by refusing to make it exciting.

Why the Sum Is Larger Than the Parts

Each field alone hits a wall. A robot can act, but cannot prove who it is or pay for anything. An AI can decide, but cannot be trusted blindly when the decision involves physical action. A blockchain can verify and settle, but it alone has nothing to settle.

Put them together, and the walls fall over. Robotics gives AI a body and the blockchain something real to settle. AI gives the robot judgment, and the blockchain something to coordinate. Crypto gives the robot an identity, a wallet, and a way to trust another machine without a referee. Adam Feiler's drone is the entire forum in one object: a robot that perceives with AI and authorises its own flight with a cryptographic certificate. None of the three layers does anything on its own. Together, they are an economic actor that does not need us in the loop.

That was the real argument under the CAIRO acronym. The convergence is not a conference theme someone picked. It is a structure still being built, and the only job for the people in that hangar is not to make it harder to finish than it has to be.

The forum closed with a soft launch of the CAIRO Club: no token yet, regulators willing, but a community, a soulbound NFT membership badge, and a promise of a robot-kitchen popup where the machines have to prove themselves by actually making the food. Anyone who was in the hangar is invited.