Humanoids Are a Bridge. The Destination Is Intelligent Automation.

The most important question in robotics isn't who builds the best humanoid. It's who builds the intelligence layer that powers physical work.

Humanoid robots are attracting attention because they make the next phase of automation easy to imagine, but their real importance is not that they look like people. Their importance is that they represent a broader shift towards intelligent automation adoption: systems that can understand, decide, and act in the physical world with more adaptability than traditional automation. The humanoid form factor may become useful in some environments, especially those already built around human movement and human work, but the bigger story is not the body. The bigger story is the intelligence layer that will eventually move across many different types of robots, machines, and industrial processes.

I have been around manufacturing long enough to have seen a few technology waves arrive with enormous promise. When I started working in and around industrial automation, Industry 4.0 and digitisation were the dominant ideas. Every factory was going to become smarter. Every machine would be connected. Every process would generate data. Dashboards, digital twins, predictive maintenance, cloud platforms, and connected production systems were all presented as part of a more intelligent industrial future.

A lot of that progress was real. Manufacturing did become more connected, more measurable, and in many cases more efficient. But it also taught the industry an important lesson: technology does not automatically solve the underlying complexity of production. If a process is poorly understood, inconsistently managed, or full of hidden edge cases, digitising it does not magically make it simple. Sometimes it simply makes the complexity easier to see.

We are now entering a similar moment with AI. The language has changed, but the ambition feels familiar. We are again talking about flexible systems, adaptive automation, intelligent decision-making, and software that can reduce the burden of engineering every detail in advance. This time, the promise is not only that machines will be connected, but that they will be able to understand, reason, and act.

That is a powerful idea, and I believe there is substance behind it. But it also deserves some caution.

In manufacturing, “flexibility” has always been one of the most attractive words in automation. It suggests that a system can handle variation, absorb uncertainty, and continue working even when the real world refuses to behave exactly as expected. The challenge is that flexibility can also become a way of postponing difficult process decisions. Rather than fully defining the edge cases, stabilising the workflow, or addressing the operational reasons why variation exists in the first place, we sometimes hope that a more flexible technology layer will absorb the messiness for us.

I saw a version of this many times with industrial cameras before the current AI era. Vision systems were often treated as a universal answer because they felt flexible. A camera seemed to offer the possibility of detecting almost anything, especially in situations where the correct sensor was difficult to define. In practice, many of those applications could have been solved with simpler, cheaper, and more reliable solutions: a basic sensor, a mechanical guide, a fixture, a light barrier, or a small change to the process. But customers were often willing to pay a premium for the expectation of flexibility, even when that flexibility created more complexity than value.

That experience shapes how I look at humanoid robots today.

Humanoids are compelling because they represent flexibility in its most visible form. They look like they should be able to step into the human-built world and deal with its complexity. They suggest that instead of redesigning a process around automation, we might one day deploy automation into the process as it already exists.

That is why the current interest in humanoids matters. It is also why we need to be careful about what we are really looking at.

Humanoid robots will not succeed simply because they are the best robots. In many manufacturing environments, they are not. Industrial robot arms are faster, more precise, more robust, and easier to justify when the task is clearly defined. Autonomous mobile robots are usually better suited to moving goods. Purpose-built automation remains the obvious choice when a process is repetitive, stable, and economically worth optimising.

So when manufacturers pay attention to humanoids, I do not think the important question is whether humanoids are better than industrial robots. Most of the time, they are solving a different problem. The real question is whether we are entering a new phase of intelligent automation adoption where adaptability becomes valuable enough to justify a very different kind of machine.

Why Humanoids Are Back in the Conversation

Humanoid robots have been a long-standing ambition in robotics, but for decades they lived mostly in research labs, technology demonstrations, and science fiction. They were impressive to watch, but difficult to imagine as reliable industrial tools. The gap between a controlled demo and a real production environment was simply too large.

That is starting to change.

Major automotive manufacturers are now testing humanoids in real production environments. BMW has piloted Figure humanoid robots at its Spartanburg plant, including work related to sheet-metal handling in the production process. Mercedes-Benz has also announced work with Apptronik to explore Apollo humanoid robots in manufacturing logistics, including bringing parts to production lines and inspecting components. [1][2]

These examples should not be overinterpreted. They do not mean humanoids are ready to replace industrial automation, and they do not mean factories will soon be full of general-purpose robotic workers. What they do show is that manufacturers are beginning to explore a different category of automation. This category is less about optimising a single task and more about adapting across many tasks in environments that already exist.

That distinction matters because traditional automation works best when the environment is engineered around the machine. Fixtures, conveyors, safety systems, work cells, tooling, and process flows are designed so that the robot can repeat a specific task with high reliability. This is an extremely powerful model, and it will remain central to manufacturing.

But it also has limits. Not every process is stable enough. Not every factory can be rebuilt. Not every use case justifies a dedicated automation project. Humanoids suggest a different possibility. Instead of asking manufacturers to redesign the environment around the robot, they point towards robots that can operate in environments already designed for people.

That is where the opportunity begins.

The Uncomfortable Truth: Humanoids Are Often Worse Robots

The current hype around humanoids can make it sound as though the human body is the ideal design for manufacturing. It is not.

If the job is to weld a car body, pick components from a known position, move material from A to B, or palletise boxes at high speed, there are usually better machine designs available. A humanoid brings a level of complexity that many industrial robots avoid entirely. It must balance, perceive, navigate, manipulate, recover from uncertainty, and operate safely around people. Every one of those capabilities adds cost, risk, and engineering effort.

This is why I do not think the future of humanoids should be framed as a replacement story. Industrial robots are not going away. Fixed automation is not going away. Purpose-built machines are not going away. In the places where speed, precision, repeatability, and throughput are the main drivers, specialised automation will continue to win.

The more interesting space is where adaptability creates more value than efficiency.

Many industrial environments are not perfectly optimised. Tasks change. Products vary. Workstations evolve. Labour availability shifts. Infrastructure ages. The surrounding process may not be stable enough to justify a dedicated automation solution, even if the task itself looks simple from the outside.

That is where humanoids become interesting. They may not be the best machine for any single task, but if they can be redeployed across many tasks, the business case starts to change. This is especially relevant for manufacturers exploring robots-as-a-service models, where the value of a robot is not only measured by the efficiency of one process, but by how effectively it can be reused over time.

In that sense, the economic promise of humanoids is not maximum performance. It is optionality.

The Factory Was Built for Humans

Manufacturing environments are full of human assumptions.

Tools are placed at human height. Workstations are designed around human reach. Doors, handles, carts, bins, stairways, inspection points, and maintenance areas all reflect the fact that people have been the default unit of physical work for more than a century.

Many processes remain manual not because manufacturers prefer manual work, but because automating them would require redesigning too much of the surrounding environment. A fixed robot cell might solve the task, but only after changes to layout, safety systems, tooling, fixtures, software integration, and process flow. For high-volume, stable production, that investment can make sense. For variable work, it often does not.

This is one of the strongest practical arguments for humanoids. They offer a possible path to automation without rebuilding everything around automation.

A humanoid can, in theory, move through the same spaces as a person, use similar tools, interact with existing workflows, and perform tasks in areas where fixed automation would be too expensive, too disruptive, or too inflexible.

That does not make humanoids universally better. It makes them strategically different. Their value is that they are compatible with a world built for people.

The Generalist Trap

A lot of attention in humanoid robotics is now focused on fully autonomous humanoids that can reason, understand the world, and perform tasks they have never seen before. This is understandable. It is an exciting vision, and it fits naturally with the current momentum around large AI models and world models.

But from a manufacturing perspective, it also risks missing something important.

Manufacturing has spent more than a hundred years doing almost the opposite. It has taken broad human capability and focused it into narrower roles, clearer responsibilities, repeatable processes, and specialised skills. The same is true of much of our education and training system. We do not generally train people to do everything. We train them to become useful in particular contexts, with particular tools, constraints, and responsibilities.

That narrowing is not a failure of imagination. It is one of the reasons modern manufacturing works.

A skilled operator, technician, welder, machinist, quality inspector, or maintenance engineer is valuable not because they can do every possible task, but because they can perform a meaningful set of tasks reliably within a specific environment. Their value comes from focused competence, context, judgement, and experience.

This is why the idea of making machines more generalist than humans deserves careful scrutiny. Many of the devices we use in automation today are good at one task, or a small number of tasks, precisely because that focus makes them reliable. Now we are asking new robotic systems to become general-purpose enough to deal with unfamiliar tasks, changing environments, uncertain instructions, and physical edge cases that even people often need training to handle.

There may be a future where robots can do that well, but it is probably not the most practical starting point for manufacturing.

A more realistic path sits somewhere in the middle. Not a fixed machine that can only perform one narrow motion, and not a fully autonomous humanoid expected to reason through any unseen situation like a general worker. The useful middle ground is a machine that is highly capable at a selected set of valuable tasks, within known constraints such as location, available resources, safety requirements, tooling, process variation, and business value.

That is also how intelligent automation adoption is likely to happen in practice.

Companies will not adopt humanoids because they can theoretically do anything. They will adopt them when they can reliably do something valuable, then something else, then a growing set of related tasks that make sense in the same environment. The value will come from building useful capability over time, not from expecting general intelligence on day one.

This applies to people as well as machines. A factory becomes more adaptable either by training people to hold multiple useful skills, by deploying machines that can perform multiple valuable skills, or by combining both in a way that makes the whole system more resilient.

That is the happy medium manufacturers should be looking for.

Acceptance May Matter More Than We Think

There is another reason humanoids are attracting attention, and it is not purely technical.

People respond differently to robots that look and move like people. Humans naturally personify humanoid machines. We instinctively interpret where they are looking, what they might do next, and how they are interacting with the environment. That can make them easier to understand than other forms of automation, even when the underlying technology is more complex.

This matters because technology adoption in manufacturing is never only an engineering problem. It is also a human and organisational problem.

Operators, managers, safety teams, unions, and works councils all influence whether a new technology is accepted. A robot that feels understandable may face less resistance than a machine that feels unfamiliar or opaque, even if the unfamiliar machine is technically more efficient.

At the same time, the humanoid form factor may create new concerns. If a robot looks like a worker and performs worker-like tasks, people may perceive it as a more direct threat to jobs than a conventional machine. The symbolism is stronger, and the emotional reaction may be stronger as well.

That is why acceptance cannot be treated as an afterthought. For humanoids to succeed commercially, manufacturers will need to communicate clearly what these systems are for, where they add value, how they are governed, and how they fit into the human workforce.

Mechanical performance will matter, but trust, perception, and communication may matter just as much.

The Real Story Is Software

The most important question in humanoid robotics is not who builds the best mechanical body. It is who builds the best software layer for physical work.

Recent advances in AI have changed what robots can potentially do. Large language models and vision-language-action models are beginning to connect perception, instruction-following, reasoning, and action. Research systems such as Google DeepMind’s RT-2 have shown how web-scale vision-language learning can be connected to robot control, while NVIDIA’s GR00T work points towards foundation models for humanoid reasoning and skills. [3][4]

This is significant, but it should not be confused with readiness for full autonomy in manufacturing. Factories are demanding environments. They require reliability, safety, repeatability, traceability, and integration with existing systems. A robot that succeeds nine times out of ten may be impressive in a demo, but unacceptable in production. Industrial automation is not judged by whether it works once. It is judged by whether it works predictably, safely, and continuously.

That is why the future of Physical AI may not be one giant model that understands everything. A more realistic path is a system of specialised capabilities coordinated by a reasoning layer.

A humanoid does not need to understand the world in an abstract, human-like sense. It needs to perform useful industrial actions. It needs to load a machine, move a tote, inspect a part, open a door, handle a tool, fetch material, support an operator, or recover from a small process variation.

Each of these capabilities can become a skill. The reasoning layer decides which skill to use, when to use it, and how to respond when the situation changes.

That is a much more practical vision for manufacturing than waiting for artificial general intelligence.

From Humanoid Robots to Intelligent Automation Platforms

This is where the conversation becomes bigger than humanoids.

Humanoids are one embodiment of Physical AI, but they are not the only one. Physical AI means software that can understand, decide, and act in the physical world. That intelligence may be deployed through a humanoid robot, an industrial robot arm, an autonomous mobile robot, a mobile manipulator, a drone, or a machine that does not yet exist.

The long-term opportunity is not to build one perfect robot body. The opportunity is to build platforms that allow intelligence to move across different robot forms, environments, and tasks.

This is why humanoids are so important, but also why they should not be mistaken for the destination. They are a visible entry point into a much larger transformation. They attract attention because they are familiar, dramatic, and easy to understand. But the deeper shift is towards software-defined automation: automation that can be configured, adapted, and improved through software rather than rebuilt from scratch for every new use case.

For manufacturers, that is the strategic point. The future may not be defined by whether a robot has two arms, two legs, or a human-like face. It may be defined by whether physical work can be programmed, adapted, deployed, and orchestrated with the flexibility we increasingly expect from software.

That is why intelligent automation adoption matters more than the humanoid itself. The lasting value will come from making physical work easier to deploy, easier to adapt, easier to govern, and easier to scale across real industrial environments.

Why Europe Has a Role to Play

This shift is especially relevant for Europe.

The DACH region is already one of the strongest industrial automation markets in the world. Germany remains a leading robotics nation, with very high robot density and a major share of Europe’s industrial robot base. [5]

At the same time, European manufacturers are facing a structural labour challenge. The German Chamber of Commerce and Industry reported in its 2025/2026 skilled labour report that 36 percent of surveyed companies were at least partially unable to fill vacancies due to a lack of suitable personnel. [6]

That combination creates pressure to automate, but Europe is unlikely to adopt Physical AI in exactly the same way as every other region. European manufacturers care deeply about quality, safety, reliability, data ownership, and governance. With the EU AI Act creating a risk-based framework for trustworthy AI, and the EU Data Act strengthening access and rules around industrial data, AI in manufacturing will increasingly need to be explainable, controllable, and compatible with European expectations around data sovereignty. [7][8]

This may become a strength rather than a limitation.

Europe does not need to win by building the flashiest humanoid demo. It can win by building trustworthy, high-quality systems for orchestrating physical work in real industrial environments.

The companies that matter most may not be the ones with the most viral robot videos. They may be the ones that make intelligent automation safe, reliable, sovereign, and genuinely useful on the factory floor.

What Happens by 2030?

By 2030, the conversation around humanoids may look very different.

Today, the form factor is the story. It captures attention because it is visible, familiar, and emotionally powerful. But as the technology matures, manufacturers will care less about whether a robot looks human and more about what work it can reliably perform.

The first wave of interest will be driven by the form factor. The second wave will be driven by useful applications. The third wave will be driven by orchestration: how different robots, skills, AI systems, and human operators work together as part of one production environment.

Some of today’s hype will fade. Limitations will become clearer. There will likely be disappointing pilots, overpromised deployments, and use cases where humanoids simply do not make sense.

But that does not mean humanoids will fail. It means the market will become more honest.

Humanoids will not replace every worker, they will not replace industrial robots, and they will not make every factory fully autonomous. Their real contribution may be different: they may accelerate intelligent automation adoption by making the idea of adaptable physical work easier to understand, test, and deploy.

The Humanoid Is a Bridge

I do not think humanoid robots are the destination. I think they are a bridge.

They are a bridge between human-built environments and software-defined automation. They matter because factories were built for people, because people can understand them, and because they create a visible and practical entry point into Physical AI.

The destination is not a robot that looks like us. The destination is intelligent automation that can be deployed into the physical world safely, reliably, and flexibly.

By 2030, manufacturers may not be buying “humanoids” in the way we talk about them today. They may be buying adaptable physical work, delivered through whatever machine makes the most sense for the task, the environment, and the business case.