The Robots Are Coming

We're stitching together this perspective note on physical agents that we've been working on for several months this summer. We're doing it because they're coming so fast now. "The robots are coming" is about to become a concept more people will hear about in the years ahead.

Those who believe the internet isn't here to stay or is a flop have become fewer in recent years. So have all those who said "I'll never have a mobile phone." However, there are many who haven't heard of or who believe in self-driving agents, digital agents, and the theme of this perspective note - physical agents.

Physical agents are machines or robots that can perform tasks in the real world, such as carrying packages, assisting in the home, or working in factories. Humanoids are a specialized type that mimics human form and function, making them ideal for interacting with humans and doing productive work in our environments. Physical agents are autonomous or semi-autonomous systems. Humanoids are a subcategory, and strictly speaking, self-driving agents are also a subcategory of physical agents, but we treat them as separate agents.

They Don't Cut Corners

Earlier today I painted a stripe on the cabin. The sun was baking and the refrigerator was tempting, and at the very bottom of the plank that needed painting was a plant of the nettle variety. I looked around and stood up. "Done!" I shouted. It's not dangerous, it's simply human, I thought as I put the lid on the paint bucket. And right here we come to a core characteristic of humanoids and physical agents: "They don't cut corners." A humanoid in a few years would have completed even the last part of the plank. They are humanoids, not humans.

Humanoids consist of hardware that resembles a human body. It's made to move into our environments and do the same tasks we do. And over time better, faster, cheaper, and more precisely. We'll go through how we understand this body with all our limitations. The humanoid also has a brain, or an artificial brain. Silicon and computer vision sewn together into something that sees, evaluates, and acts. The human brain is perhaps the only part of the body that cannot be transplanted or replaced without remaining who you are.

The word "Robot" means slave and work and comes from a play in the 1920s. And that's exactly how we see these three agents (self-driving, digital, and physical) of ours. They will lead us into a world we describe as "post-labour" where the slaves (agents) will produce goods and services that we previously produced with hands, feet, and cognitive processes.

1.0 Historical Perspectives

1.1 The Steam Engine and Humanoids

The transition from human labor to machines can be compared to the transition to the humanoids we see working on the factory floor and taking their first steps into homes. But there are both similarities and differences when we use the energy concept to explain efficiency.

Humanoids that mimic human form and movement represent a further development of machines, but efficiency must be assessed based on their design and use cases. Humanoids increase efficiency compared to human labor in similar ways to machines, by utilizing energy better, working continuously, and performing precise movements. Compared to traditional machines, they sacrifice some efficiency for flexibility, but their human-like form makes them ideal for tasks in our environments designed for humans.

So next time you hear someone say "it's just a chatbot," think about how the steam engine started as a pump and ended as the world's first high-speed transport. Technology doesn't stand still.

1.2 Four Human Qualities Versus Machine Development

Throughout almost the entire industrial journey, we've let machines "borrow" our bodies bit by bit. Now we're approaching a point where they will also borrow our hands, thoughts, maybe even understand the mood in the room. Here's a quick look at how far it has come and where it might end.

2.0 Categorization of Physical Agents by Technology and Use Cases

Physical agents are autonomous robots that operate in physical environments. They perceive surroundings, make decisions, and perform actions. They can be categorized based on technology (basic or advanced) and use cases (specific or general).

Leading Companies in Physical AI Agents

3.0 Can Digital and Physical Agents Be Considered a New Species?

In biology, a species is a group of living organisms that can produce fertile offspring while being genetically distinct from other groups. Natural selection, first described by Darwin, favors the traits that best fit the environment, and over time new species are formed. AI systems and robots have no DNA, they don't reproduce, and their mutations are due to code changes rather than genes. In a strict Darwinian sense, they are therefore not a species.

Nevertheless, something similar is happening: weaker algorithms are discarded, stronger versions live on, and humans (or other algorithms) copy, combine, and improve them, much like breeders have shaped livestock through artificial selection.

Transhumanists (Ray Kurzweil being the most famous) believe technology can take over the baton from biology. If artificial systems can one day improve themselves faster than we can keep up, we get a form of "technological species." Tesla's humanoid "Optimus" and today's large language models already show how fast iterations can go when machines train on enormous datasets. If a system achieves so-called general intelligence (AGI) and handles its own further development, we break out of the Darwinian framework and into a new evolutionary form driven by code.

4.0 Humanoids as Disruption of Economic Labor

When the steam engine came, it took over muscle power. In our time, software has gradually snatched up routine jobs in front of the PC screen. Humanoids do both at the same time: They lift boxes, repair pipes, and send the report afterwards. One and the same machine can pick goods in a warehouse, talk to the logistics system, troubleshoot a sensor, and order spare parts online. The big difference from classic industrial robots is flexibility. A modern robot can be trained for new tasks with software updates, much like downloading an app to your phone.

Erik Brynjolfsson and Andrew McAfee have shown that automation has so far first eaten the routine jobs. Humanoids expand the menu to home help, construction, service, and parts of the social field. Thomas Piketty's concern that returns on capital grow faster than wages gains new weight. Owners of robot fleets can run off with most of the value creation. Economists who envision a post-work economy therefore bring up arrangements like universal basic income or broad ownership in robots as a possible counterweight.

For many, the job is more than the paycheck. It provides identity, community, and routines. When a machine takes over, people risk being left with time but without a clear purpose. Karl Marx called it alienation. Today we can see the contours of a similar emptiness if work disappears without something else taking its place.

Physical agents are not just another wave of automation. Humanoids lay claim to the entire spectrum of tasks that have traditionally given people wages and identity. If we manage to update economic models, social safety nets, and our own understanding of meaning quickly enough, the result can be more prosperity and more freedom. If we fail, the gap between those who own the robots and those who don't can become so large that the social fabric is torn to pieces. The choice is not between robots or not, but between letting the upheaval control us or controlling it ourselves.

5.0 How Humanoids and Industrial Robots Can Transform the Future

Julius Caesar is known for several quotes, one of them being: "I came, I saw, I conquered." We humans also follow a simple three-step process. We humans see, evaluate, and act. This process lets us do everything from simple things like grabbing a cup and drinking coffee, to complex tasks like playing piano. Humanoids are designed to mimic humans, so they use technology to recreate the same process: see, evaluate, and act.

Together, these systems let humanoids interact with the world, such as navigating or performing tasks. However, it's still uncertain how well humanoids can match humans' intuitive abilities. This is especially true in complex, varied tasks.

6.0 Voice Commands Are Central for Humanoids

Voice commands have the potential to disrupt finger-based interfaces. The global market for humanoids (expected $38 billion in 2032) and VUI, i.e., speech/voice commands (USD 30.46 billion in 2025) shows strong growth with over 20% CAGR, due to developments in NLP (Natural Language Processing), edge computing, and multimodal AI.

Humanoid robots are built to work side by side with humans in homes, stores, hospitals, and factories. The interaction must therefore feel natural. Pulling up an app or waving a remote control breaks the illusion that the robot is a helpful colleague. Our voice, on the other hand, is the most intuitive interface we have.

When you say "Optimus, lift the package into the van," the robot interprets your intention, finds the package, and performs the task. You avoid having to describe the exact route, location, or grip, just as you would do with another human.

7.0 A Complete Overview of the Body and Technology

A humanoid robot's body is built from components that mimic human anatomy. Basically, they are based on advanced technology for movement, perception, and control. The most important components are control systems, actuators, and sensors, while the most challenging single area is the hands.

8.0 "The Hand is Where the Mind Meets the World"

The human hand is a marvel of engineering with 27 degrees of freedom, allowing for everything from delicate precision work to powerful gripping. For humanoids, replicating this capability is one of the greatest challenges in robotics.

Despite these challenges, all components are becoming faster, cheaper, and more efficient. And quickly.

9.0 Training and Learning

Physical agents learn in virtual worlds first. In tools such as Isaac Sim, Unity Robotics, and Gazebo, the robot gets to practice thousands of variants of the same task. Everything from maintaining balance on a downhill slope to assembling a cabinet. This way, companies save both time, money, and wear. One day of simulation in the cloud can equal several years of "physical" practice.

After simulation, reinforcement learning (also called "trial-and-error learning") takes over. The algorithms reward each successful movement and "punish" errors. Optimus finds the most energy-efficient gait, Unitree H1 learns to jump higher without losing balance, and Figure fine-tunes finger movements needed to pour coffee without spilling. This phase is often driven on large cloud clusters because the updates require enormous computing power.

10.0 Strategy and Categorization

Amara's law says that we often overestimate technology's short-term impact and underestimate its long-term potential. In an investment context, this is another way of saying we need to think long-term. It's also a nice entry point for analyzing technologies' different phases.

10.1 Musk's Quote on Humanoids

The quote from the TED interview in 2022 provides a fascinating starting point: Musk sees Optimus as a natural extension of Tesla's work with self-driving cars, where real-world AI is the key, and he emphasizes that humanoid robots can be "bigger than the car."

Elon Musk likes to call Tesla's cars "robots on four wheels." With Optimus, he takes the statement literally - he wants to make the car bipedal. To succeed, he points to two completely fundamental obstacles: the brain (real-world AI) and the body (large-scale production). If Tesla manages to crack both, robots can become as commonplace as vacuum cleaners.

10.2-10.4 Global Humanoid Landscape

11.0 Timeline

Have we succeeded in our telling of perspectives, there now exists an imagination that humanoids are moving rapidly from cautious pre-programmed steps in Boston Dynamics' laboratory to computer-vision humanoids with AI brains that want to contribute to productive work.

12.0 The Interplay Between Digital and Physical Agents

Until now, most of us have been used to artificial intelligence in the form of a chatbot that answers questions, or an algorithm that recommends the next series. Now this software is getting arms and legs. But the software must talk to the arms and legs. Digital agents: software on our digital gadgets (e.g., planning routes or interpreting images) collaborate with physical agents: robots, drones, or self-driving vehicles.

13.0 Ethics

When investment theme is humanoids, questions about fear sometimes come up. Fear that a variant from the Terminator films will kill us, and other times questions about whether we'll be scared. Often we answer that the calculator is much smarter than us, but we're not afraid of the calculator. The topic that most often comes up in such contexts is about humanoids and ethical action spaces.

Ethical dilemmas are situations where a person or humanoid faces a choice between two or more actions. No matter what they choose, the choice will violate one or more ethical principles or values. There is no obviously right choice and solution.

Humanoids will undeniably end up in situations where they must make decisions where the right choice is not unambiguous. Human choices in such situations are often explained with concepts like intuition, experience, or human judgment. For humanoids, dilemmas arise at the intersection between programmed rules, human expectations, and possible developed autonomy in the humanoid itself.

A classic dilemma is situations where the humanoid must choose whom to save. Let's imagine an emergency situation where two people are in danger. However, only one can be saved. How should agents decide who gets priority? Should it base itself on age, health status, or other criteria? This is reminiscent of the famous trolley problem.

13.1 Can Humanoids Exert Group Pressure on Each Other?

Is it possible to imagine situations where humanoids exert group pressure, especially if they are equipped with advanced AI that gives them local freedom in perception and action? Group pressure occurs when individuals (or in this case robots) are influenced by the group's norms, goals, or behavior to change their own actions, even if it conflicts with their individual "logic" or programming.

Yes, humanoids can exert group pressure in the future, especially with local freedom in perception and action. Scenarios like efficiency pressure in factories, social adaptation in homes, or cultural adaptation in public spaces show how this can happen. This raises fascinating questions about AI ethics, societal impact, and investment opportunities.

14.0 Use Cases

15.0 Disruptive Investment Strategies and Physical Agents

Our investment universe was defined at the entrance to what we called the disruptive decade (2020-2030). We have analyzed, discussed, and invested in satellites, IoT, drones, robots, artificial intelligence, edge technology, metaverse, eVTOL, self-driving cars, drones, fintech, and energy transformations over the past five years. Often to the laughter of investors and advisors who are more concerned with looking in the rearview mirror than looking forward. We live well with that.

The next five years, high beta will become low beta, and low beta will become high beta, we often say. Large index-heavy and well-established business models will be disrupted by new technology and smaller players. Fortunes will probably change hands in the next five years too.

Physical agents directly and indirectly touch several categories and subgroups. The main category is what we call (3) Machine Revolution. (2) Urban mobility we define as the self-driving agents. However, physical agents require a large ecosystem of suppliers where both (1) Connectivity and (5) Green deal become important. Electricity, batteries, access technology (satellites/radio), sensors, and semiconductors are all enablers for physical agents.

Key Takeaways