Massimiliano Moruzzi, Founder and CEO of Xaba – Interview Series

Massimiliano Moruzzi, founder and CEO of Xaba, is a longtime industrial automation and AI executive with deep expertise spanning robotics, manufacturing systems, CNC machinery, and AI-driven industrial control. Before founding Xaba in 2022, he held leadership roles at Augmenta, where he led research and development efforts focused on AI-powered automation, and previously served in senior engineering and software R&D positions at Ingersoll Machine Tools and IMTA. Across more than two decades in industrial technology, Moruzzi has focused on bridging the gap between advanced robotics and practical manufacturing deployment, with a particular emphasis on enabling machines to operate more intelligently, adaptively, and autonomously.

Xaba is a Toronto-based industrial AI company developing what it describes as “synthetic brains” for industrial robots and factory systems. The company’s platform combines generative AI, reinforcement learning, robotics control, and industrial automation to allow robots, CNC machines, and PLC-controlled systems to self-program and adapt in real time without manual coding. Its flagship technologies, including xCognition and PLCfy, are designed to automate robotics programming, optimize manufacturing workflows, and accelerate deployment across industries such as aerospace, automotive, and advanced manufacturing. Xaba positions its technology as a way to modernize factory automation by replacing rigid, manually programmed systems with AI-driven cognitive control capable of learning from operational data and dynamically adjusting to changing production environments.

What first sparked the idea for Xaba, and when did you realize industrial robots needed a fundamentally different approach—essentially a synthetic brain rather than more lines of code?

The spark came from observing how most industrial robots fail at the most basic level of variability. These machines are mechanically precise, but cognitively fragile. Small changes in part tolerances, process parameters, or material behavior can throw off an entire operation.

The industry’s response has been consistent: write more code, add expensive rigid fixtures to eliminate variability, layer on more rules, rely on human oversight, and keep recalibrating the system.

That’s when the realization hit me: this is not a software problem — it’s a missing brain.

Today’s industrial robots and controllers blindly execute instructions without understanding whether the outcome is actually good or bad. They don’t reason about the physical world around them.

Robots are not failing because they lack instructions; they are failing because they lack understanding. Humans don’t rely on thousands of lines of code to tighten a bolt or apply adhesive. We adapt instinctively based on force, movement, and physical feedback.

It became clear that industrial robots need a synthetic reasoning system grounded in physics, not just another layer of programming.

How did your experience at Augmenta AI and earlier roles shape your perspective going into Xaba, and what specific gaps or insights pushed you to build this company?

At Augmenta AI, we were deeply focused on AI-driven decision-making, optimization, and autonomy. What became obvious is that most AI systems were operating in an abstract manner, meaning they optimized data representations rather than interacting with physical reality.

In earlier roles, I’d seen automation projects stall or fail not because the robots weren’t capable, but because the engineering overhead was unsustainable. The gap was clear: there was no intelligence layer that could connect high-level intent with real-world physics. Xaba exists to bridge that gap, giving machines the ability to reason about force, motion, constraints, and outcomes in the same way skilled humans do.

Xaba is building the world’s first physics-based GenAI system for industrial robots. How does this approach differ from traditional robot programming and from today’s mainstream AI models?

Traditional robot programming depends on predefined paths, process parameters, forces, and sequences of actions. It assumes the environment behaves the same way every time, like a CAD model.

Mainstream AI models take a different approach, but they’re still largely statistical. They’re good at prediction and imitation, but they don’t truly understand physical cause and effect.

Xaba’s Physics-AI introduces a third paradigm. Instead of relying primarily on visual data or static instructions, we use time-series data from sensors such as force, temperature, acceleration, voltage, acoustics, and vibration to understand the underlying physics of a process.

This gives the system an understanding of how actions impact outcomes. Instead of simply following instructions, the machine can adapt in real time when conditions change.

We’re moving industrial robots from rigid automation to systems that can reason physically about the work they’re performing.

How does synthetic reasoning improve quality, repeatability, and real-time adaptability on the factory floor?

Synthetic reasoning enables robots to adapt during the task. If the resistance changes, the robot compensates accordingly. If material behavior shifts, it adapts the motion. This leads to higher quality because the robot responds to reality, not assumptions.

Repeatability improves because the system isn’t replaying fragile trajectories; it’s re-solving the task each time based on physical intent. And adaptability becomes native, not an exception that requires reprogramming.

Why do you believe the next major breakthrough in AI will happen in physical systems, rather than purely digital ones?

Because the real world runs on physics, not correlations. Most of today’s AI is built around pattern recognition and prediction.

The biggest AI breakthroughs so far have happened in digital environments where recognizing patterns is often enough. But physical systems like welding, machining, and assembly work differently. They depend on causal relationships between force, energy, temperature, motion, and material behavior. In these environments, small variations can break a process, and errors have real consequences.

This is why the next breakthrough requires a shift from data-driven prediction to physics-based reasoning.

Physics-AI enables this shift. By using time-series sensor data to extract the governing equations of a process, AI can move from guessing outcomes to understanding how the system behaves. This allows machines to adapt in real time, even under variability.

• Digital AI → largely built around correlation, prediction, and content generation.
• Physics-AI → Enables machines to reason, adapt, and respond to real-world conditions in real time.

The next wave of AI will not be defined by better LLMs or Imitation Games, but by machines that can understand and control reality.

What makes today’s automation infrastructure outdated, and what does it take to fix it at an industry-wide scale?

Today’s infrastructure is built on the assumption that variability is the enemy. Everything is rigid, over-engineered, and expensive to maintain. It doesn’t scale well because every new product or process variation requires massive human intervention.

Fixing this requires a shift from programming to cognition. You need a universal intelligence layer that can sit on top of existing hardware and make it adaptive. That’s how you modernize automation without ripping out decades of investment.

Many manufacturers struggle with tasks that still require thousands of lines of code and weeks of calibration. How does Xaba eliminate this bottleneck?

Manufacturers hit that bottleneck because today’s systems are code-driven and imitation-based, not understanding-driven. They rely on thousands of lines of logic or on AI models trained on pixels and videos, which we often call an imitation game. These approaches capture patterns, but they don’t understand the underlying process.

Xaba takes a fundamentally different path.

We use time-series sensor data, force, temperature, current, and vibration to build a new class of foundational models grounded in physics. Instead of learning correlations, our Physics-AI extracts the governing equations of the process. This gives the system a true causal understanding of how actions affect outcomes.

From there, the system generates physically valid actions in real time. The robot doesn’t replay examples or follow predefined code; it reasons about the process before acting and adapts continuously under variability.

In practice, that means no thousands of lines of code, no reliance on pixel-based imitation, and no constant recalibration when conditions change. Instead, you get a system that understands the physics and controls it. That’s how we move from programming and imitation to true physical reasoning and autonomous control.

Robots learning from demonstration is a bold shift. What technical milestones made this possible, and what constraints still exist today?

Robots learning from demonstration is an important step, but it’s still largely an imitation-based approach. These systems map observations (such as pixels or trajectories) to actions without understanding the underlying physics of the task.

From a Physics-AI perspective, the real milestone is moving from imitation to causal understanding.

What made this possible is:

• Advances in perception (vision-language models, multimodal data)
• Large-scale datasets of human and robot behavior
• Improved policies that can map observations to actions

But these systems are still fundamentally correlation-driven. They can replicate what they’ve seen, but they struggle when:

• Materials behave differently
• Process parameters change
• Geometry or tolerances vary
• Real-world physics deviates from training data

That’s where the limitations become clear.

At Xaba, we take a different approach. Instead of learning what to do from demonstrations, we learn why it works.

Using time-series sensor data, Xaba extracts the governing physics equations of the process. This creates a foundational Physics-AI model that understands how the system behaves under different conditions.

The real breakthrough comes from a machine’s ability to reason about forces, energy, and material behavior, adapt in real time, and generate physically valid actions.

How does Xaba’s system adapt to unpredictable real-world conditions—material variations, tool wear, or subtle environmental changes?

Because the system continuously reasons about force, motion, and outcomes, it can detect when reality deviates from expectations and adjust in real-time. Tool wear becomes a variable, not a failure. Material variation becomes part of the reasoning loop.

This is fundamentally different from threshold-based error handling — it’s continuous adaptation.

Looking ahead five years, how do you see physics-based GenAI evolving, and what does a fully autonomous factory enabled by synthetic reasoning look like?

From my perspective, the next five years will mark the transition from automation to true cognitive manufacturing.

Physics-based GenAI will evolve from optimizing individual tasks to building foundational models for entire industrial systems. Instead of training on pixels or past trajectories, these systems will continuously learn from force, temperature, energy, and dynamics, enabling causal understanding of every operation.

The shift is profound:

• From programming → self-generating control strategies
• From static models → continuously learning systems
• From correlation → physics-grounded reasoning

A fully autonomous factory enabled by synthetic reasoning will look fundamentally different. Machines will self-program based on desired outcomes, adapt in real time to variability in materials and geometry, and inherently control quality rather than inspect it after the fact. Knowledge will not be siloed — it will propagate across machines, lines, and even factories, improving performance continuously.

But the most important transformation is human. With a true synthetic brain for manufacturing, the relationship between humans and machines becomes bidirectional. Humans will not just program machines, but will learn from them, just as machines learn from human intent and experience.

Automation stops being a job function and becomes a platform for career growth, continuous learning, and discovery. Engineers, operators, and technicians will collaborate with systems that explain, adapt, and elevate their understanding of physical processes.

In that world, there are no weeks of calibration or thousands of lines of code. The factory operates as a coordinated, physics-aware system that amplifies human capability and insight.

Ultimately, we move from factories that execute instructions to factories that understand, reason, and co-evolve with humans. That’s the future we are building at Xaba.

Thank you for the great interview, readers who wish to learn more should visit Xaba.