Why Robots Can Backflip but Struggle With Real Work | Enzo Wälchli
Enzo Wälchli of Operator Circle, former ANYbotics Chief Commercial Officer, on why hardware isn't SaaS, field reliability, humanoids and Zürich's deep tech edge.
Discuss this episode and meet other Swiss tech builders in the TechDrive WhatsApp group.
Key insights: executive summary
- A robot can do a backflip, yet pulling keys out of a pocket or putting a wire somewhere is far harder for it. Enzo's rule of thumb is that what is easy for robots is hard for humans, and the other way round.
- Spectacle has a purpose. Impressive demos create excitement and bring in money, and that money pays for the hard problems that create the real value. Treating the demo as proof that those problems are solved would be a mistake.
- Hardware cannot be judged like software. A robot changes how a plant runs, so customers want a pilot, then a longer pilot, then pilots at other sites with different people before they scale. Expecting software-style expansion misreads what the customer is being asked to change.
- Margins need a different yardstick. Software gross margins of 90% and above set the benchmark, while hardware at 50 to 60% is respectable rather than outstanding, and an investor trained on SaaS can mistake it for a broken business.
- The last improvements come from the field. A model trained to 99.9% accuracy in the lab or in simulation is not enough for an industrial setting that needs 99.9999%, and only real robots doing real work produce the data that closes the gap. Older robotics companies, which look slow from outside, may hold an advantage.
- Zürich has moved from banker city to tech city. By Enzo's estimate the city now has about 150,000 people in tech against 100,000 in finance, and robotics founders with no Swiss customers still choose to build here because the network and talent are available.
“Every deep tech hardware company is also a software company in a way, right? Because the value is not the metal box in a robot. The metal box is just the enabler.”
“The interesting thing in robotics is what's easy for robots is hard for humans and vice versa.”
Why can a robot do a backflip but struggle with real work?
Enzo's observation is that what is easy for robots is hard for humans, and the other way round. A backflip is out of reach for most people and easy for a robot, while pulling keys out of a pocket or putting a wire in place is learnable for a person and incredibly hard for a machine.
In his view the order makes sense. Do the easy things first, even if they create little value, because they generate excitement and bring in money. That money then pays for the big, hard problems, and eventually those get solved too.
Spectacle has a purpose
Demos such as backflips attract attention and funding, which helps pay for the useful work that comes later.
A mixed fleet, not one form factor
Two legs, four legs, wheels or flight: Enzo expects every form factor to find a place, with robot teams of different kinds working together.
Why should hardware companies not be judged like software companies?
Everything a software company has to prove, a hardware company has to prove as well: that the product creates value, works and can be implemented. On top of that it has to prove reliability over a long period, and it cannot simply ship a patch when something goes wrong. A fix means changing the machine, so iteration cycles are much harder.
A robot also interrupts the customer's day-to-day processes far more than a new app does, so customers want a longer time to observe it. One month of working does not settle whether it will still work in six months, which is why a pilot turns into a longer pilot and then a pilot at another plant, with different people and conditions, before anyone scales.
Expect longer timelines
Several pilots across several plants before scaling is normal for hardware, not a sign of failure.
Read margins differently
Software margins of around 90 to 95% set the usual benchmark, while 50 to 60% for hardware is respectable rather than outstanding. An investor using a SaaS lens can see it as a problem.
What does it take to reach industrial-grade reliability?
Training a model to 99.9% accuracy in the lab or in simulation is great, but Enzo calls it useless in an industrial setting, where the requirement is 99.9999%. The only way to close that gap is real-world examples, which means shipping robots and collecting data from them doing real work.
Many companies that did that are now five to ten years old and struggle to raise their next round, because they burned a lot of capital relative to their traction. They do have the data, though. Enzo sees them as having completed the first camel hump, and as the ones best placed to reach the reliability that newer entrants still have to earn by deploying.
New robotics companies will move faster and use less capital, because AI now speeds up software, hardware design, testing, failure analysis and the reading of field data. Customers still want to see and test a robot before they rely on it, so deployment time does not disappear.
Can robot software be built without the hardware?
Enzo describes two schools of thought. One holds that great software needs your own hardware, an idea he links to Steve Jobs. His view is that this still applies in robotics for now: until hardware is commoditised, meaning very reliable, cheap and abundantly available, you need it to train and learn your software, and if the hardware changes the software has to be retrained.
Some players bet that commoditisation will come faster and raise a lot of money on that bet, because software-only teams can move much faster. For now, he says, a software-only company needs very strong partnerships with hardware makers, or has to build the hardware itself. Every deep tech hardware company is also a software company, because the software is where the value is created.
Why is Zürich becoming a deep tech hub?
Enzo points to a shift in the city's workforce. Twenty years ago Zürich had roughly 100,000 people in finance and about 50,000 in tech. Today finance is still around 100,000, but tech is about 150,000. Big tech arrived first, with Google, which by his account has around 6,000 employees in Zürich, its biggest centre outside the US.
The timing matters too. Switzerland built its venture reputation on life science and biotech and missed the platform, mobile and web waves. Deep tech combines hardware, electronics, mechanical and precision engineering with the AI competence that big tech brought in, and Enzo thinks almost no ecosystem in Europe has a better basis for it. Researchers from abroad with no link to the Swiss universities and no Swiss customers now found robotics companies here, because the network and talent let them grow faster.
Are humanoid robots the future?
Enzo thinks humanoids are probably the future, and sees general purpose as the bigger question. Two arms that can do something reliably and quickly fit into a lot of processes, but he says we are far from that today. Whether they need legs depends on the use case. A robot with every human degree of freedom that is also reliable and cheap is probably still years away, while two arms on a torso, on a movable or fixed platform, may offer benefits over classic industrial arms because they work in an environment built for people.
What does an operator-led VC do differently?
Operator Circle is an operator-led venture fund. Its operators are people who have scaled companies themselves in leading roles, and who now invest time, energy and money into the ecosystem. Enzo says the idea is common in the US, where many venture partners built and sold companies, but rare in Switzerland, where you can count such investors on one hand and most come from fresh graduate routes, wealth management or private equity.
His argument is about empathy. Scaling a company is mostly not nice: the leader is always facing a problem, either growing too slowly or growing too fast, and investors who have not been in that position find it harder to understand. Operator Circle invests at Series A and Series B, but wants to meet founders earlier because it can already help them build the right company for that stage.
Who is Enzo Wälchli?
Enzo Wälchli is the Founder of Operator Circle, an operator-led venture fund investing in deep tech, and the former Chief Commercial Officer of ANYbotics, where he joined as one of the first commercial hires and helped build the commercial organisation. Earlier he worked in sales and marketing at Hilti and in strategy at Sulzer.
Episode show notes
Enzo Wälchli started his career in strategy at Sulzer, writing market-entry plans from a 23rd-floor corner office without meeting a customer. He moved to Hilti to sell drilling machines on construction sites, then joined ANYbotics as one of its first commercial hires and helped build its commercial organisation. Today he runs Operator Circle, a venture fund led by people who have scaled companies themselves. This is the tenth drive of TechDrive Zürich.
What stood out was the gap between the demo and the job. A robot can do a backflip and still struggle to pull keys out of a pocket, and Enzo explains why that matters for pilots, margins, reliability and the way investors read hardware companies. We also cover humanoids, why Zürich has become a deep tech city, and who he wants to hear from.
More conversations from the Swiss tech ecosystem
- Nobody Has Left a Robot Working Overnight | Yulia Sandamirskaya, ZHAWYulia Sandamirskaya, ZHAW Zurich University of Applied Sciences
- Physical AI Is Ready, Buildings Are Not | Kateryna Portmann, ANYboticsKateryna Portmann, ANYbotics
- Switzerland's AI Movement Is About to Start | Swiss {ai} WeeksKatka Letzing and Anna Hakobyan, Swiss {ai} Weeks
Follow the Swiss tech ecosystem, one drive at a time
New in-car interviews with founders, investors and researchers building Zurich's deep tech, AI and robotics scene. Want to discuss this episode, ask follow-up questions and meet other builders?