Physical Intelligence is no longer a concept from the future — it is already up and running inside real factories, moving robotic arms, optimizing production lines, and making decisions in real time.
The world now has nearly 5 million industrial robots in operation, and that number grew almost 10% in a single year.
Pretty impressive, right?
But here is the detail that changes everything: the vast majority of those robots still follow instructions written by humans before the work even begins. In other words, someone had to anticipate every condition, every variation, every possible situation. That model defined industrial automation for six decades and delivered solid results in precision, safety, and productivity.
But the game is changing — and fast 🚀
Physical Intelligence is moving smarts directly into the manufacturing work itself. Robots, vehicles, production systems, and connected infrastructure can now sense changing conditions, reason about context, and act within defined boundaries. It is no longer just following a script. It is adapting the script while the work happens.
During the International Manufacturing Technology Show (IMTS) 2026 in Chicago, Microsoft presented this opportunity around a very practical question: how do manufacturers turn intelligence that works on one specific machine or process into a capability that can be deployed, governed, and improved across the entire enterprise?
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Companies like KUKA, ABB, and Krones are already seeing real results with this approach — and the numbers are worth paying attention to. In this article, you will learn what Physical Intelligence is, how it is transforming manufacturing operations in practice, and where an industry can start this journey without having to turn everything upside down all at once.
What Physical Intelligence is and why it matters right now
Physical Intelligence is, at its core, intelligence operating in the real world through machines, systems, and people. It works within a continuous cycle that repeats and improves with every loop:
- Sense: read machines, environments, and operational signals.
- Reason: interpret that context and determine the next best action.
- Act: execute through machines, systems, and people, using the outcome as feedback for the next cycle.
Think of it this way: a traditional robot knows how to tighten a bolt because it was programmed to do so. A robot with Physical Intelligence knows how to tighten the bolt, notices the part is slightly misaligned, adjusts the pressure and angle in real time, and logs that variation to improve future runs. Connected operations gave people visibility. Industrial AI added the ability to reason about patterns and outcomes. Physical Intelligence closes the loop by acting in the physical world, measuring the result, and refining the next round.
Now, there is an important point here. When AI moves beyond just recommending something on a screen and starts acting in the physical world, both the value and the risk become physical too. That is why operational boundaries, identity and security, observability, validated fallback mechanisms, and human approval need to be designed into the system from the start. This is not an optional detail — it is part of the architecture.
What makes this concept especially relevant right now is the convergence of technologies that used to exist separately. Computing power has gotten cheap enough to run inside industrial machines. Sensors have become small, accurate, and affordable. And machine learning algorithms have evolved to the point where they can process real-world data — vibration, temperature, pressure, images — with a speed and precision that surpass human capability in many scenarios. That combination has created a window of opportunity that the most forward-looking industries are already taking advantage of.
How machine learning is changing industrial automation in practice
For decades, industrial automation was built on a simple logic: define the rules, program the system, let it run. That approach worked extremely well for repetitive, predictable, and controlled tasks. Automotive assembly lines, product packaging, welding of metal structures — all of that was automated with impressive efficiency using deterministic logic. The problem is that the real world is not always predictable. Materials have variations. Parts arrive with imperfections. Environmental conditions change. And every time something goes off-script, the traditional system stops, throws an error, or worse, produces a defect.
That is where Physical Intelligence creates value across the entire industrial lifecycle, from engineering to live operations. On the engineering side, KUKA is making the programming and deployment of industrial robots much simpler. With the iiQWorks.Copilot tool, a user describes a task, generates the code, simulates the workflow, and deploys it. For simple tasks, programming can be up to 80% faster. Meanwhile, Krones is using AI agents and physically accurate digital twins to accelerate filling simulations. Processes that used to take three or four hours can now be completed in five minutes or less.
In live operations, the picture is just as encouraging. ARUM is bringing this transformation to the machining center with the TTMC Brain, which turns a simple conversation into a machining program and reduces CNC programming from 177 steps to just two. That is a massive shift in how the work gets done. ABB, in turn, is applying real-time industrial data for predictive maintenance and continuous optimization. Across its customer base, ABB reports up to 20% more reliability in critical assets and up to 60% less unplanned downtime.
These examples tackle two major obstacles to scaling automation: the engineering effort needed to deploy it and the reliability needed to keep it running. Intelligent systems can optimize energy consumption in real time, adjusting the intensity of motors, conveyors, and cooling systems based on actual demand at that moment — not on an average estimate made during the planning phase.
What Physical Intelligence changes for operations leaders
In conversations with manufacturers, it becomes clear that the biggest challenge today is increasingly about orchestration. A single successful machine or work cell is just the first step. Leaders need to connect that capability to how the company operates, makes decisions, and grows. And Physical Intelligence transforms three fundamental models within an organization.
From automating sites to coordinating networks
The operating model shifts from automating a single site to coordinating a network. Lines, plants, fleets, and field environments can operate as a single connected system, sharing information and adjusting in an integrated way.
From manual planning to agent-assisted operations
The decision model moves from manual planning to operations supported by intelligent agents. People define the intent, the policy, the approval, and the intervention boundaries, while agents interpret context and coordinate work within those limits. The human operator shifts from executor to supervisor, focusing on strategic decisions.
From disconnected tools to a governed environment
The technology model evolves from isolated tools to a single governed environment. Data, controls, platforms, and systems from different vendors need to work together, both in the cloud and at the edge. And here is a key takeaway: repeatability matters more than any single deployment. A successful machine only becomes an enterprise-wide capability when its data, integration, governance, and operating model can be reused across other lines, plants, and fleets. Otherwise, the organization ends up scaling exceptions instead of capability.
AI governance in manufacturing is part of the architecture
One point that many people underestimate is governance. Identity, policy, monitoring, security, and accountability need to travel along with the capability from one machine or site to the next. Operational, engineering, and business context should feed the same decisions, regardless of whether the system runs in the cloud or at the edge.
For global manufacturers, that architecture also has to account for data sovereignty, data residency, and operational technology security across different sites and regions. Leaders need clear visibility into what the system can do, why it acted a certain way, and at which points a person needs to approve, step in, or cancel an action. A consistent trust model makes it possible to deploy, manage, observe, and improve capabilities across mixed environments without having to rebuild controls at every new location.
Collaborative robots and the new relationship between humans and machines
One of the most visible transformations brought by Physical Intelligence on the factory floor is the rise of so-called cobots — collaborative robots designed to work side by side with humans, rather than in spaces separated by safety cages. This completely changes the logic of traditional automation. Before, the robot was an isolated island of efficiency. Now, it is a work partner that adapts to human presence, detects nearby movement, slows down when someone approaches, and can handle tasks that require a combination of mechanical strength and fine dexterity.
What machine learning adds to this equation is the ability for continuous improvement without constant human intervention. A modern cobot can analyze its own operational data, identify patterns of inefficiency — such as unnecessary movements or time variations during certain steps — and propose or even implement adjustments autonomously within the limits set by the operator. This means the machine gets smarter over time without anyone needing to manually reprogram it.
Where to start without reinventing the entire factory
One of the biggest concerns industrial managers have when it comes to Physical Intelligence and industrial automation is the cost and complexity of implementation. The good news is that manufacturers do not need to automate an entire plant to get started. The most practical starting point is to pick a single operational problem where the outcome matters and the boundaries are clear.
Choose a workflow with measurable value, such as quality inspection, maintenance triage, machine programming, or process optimization. Then, map out the data and operational context needed to make a solid decision. Define what the system can recommend, what it can execute on its own, and at which points a person needs to approve or step in. And finally, design the deployment so that the successful pattern can be reused on another line, site, or fleet.
A very common path is to start with predictive maintenance. Vibration, temperature, and energy consumption sensors can be installed on critical equipment without interrupting production. The collected data feeds machine learning models that identify patterns associated with imminent failures, enabling scheduled interventions before the problem actually happens. This type of implementation delivers a relatively fast financial return — fewer unplanned stops mean less loss from missed production and from emergency corrective maintenance, which always costs more than preventive work.
Another possible entry point is automated quality control using computer vision. High-resolution cameras combined with machine learning algorithms can inspect products at line speed, catching defects that escape human inspection and generating valuable data about the root causes of problems. At the end of the day, Physical Intelligence will be measured in operations: faster engineering, safer work, more reliable assets, and systems that improve under clear human authority. Start with a relevant workflow, define your boundaries, and build it so that success can be repeated. It does not have to be an overnight revolution. It can — and maybe should — be a well-planned evolution. 🤖
