Source: Unite.AI
Brian Hartzband, President of US Operations, GMEX Robotics, leads the company’s U.S. expansion, with a focus on building its commercial footprint, developing strategic partnerships, and bringing its physical AI and robotics platforms into real-world deployments. Prior to joining GMEX Robotics, Hartzband spent nearly four years at TEKsystems across technical recruiting and account management roles serving government and higher education clients. He previously co-founded and helped lead Handcrafted 4 Home, where he drove U.S. market expansion and partnerships with major retailers and e-commerce platforms. Earlier in his career, Hartzband worked in wealth management at Merrill Lynch and UBS, where he advised high- and ultra-high-net-worth clients and helped build a substantial international client portfolio.
GMEX Robotics is a Nasdaq-listed physical AI and robotics company developing intelligent machines for commercial environments. Formerly known as Fitell Corporation, the company follows what it describes as a “Terminal + Brain” approach, combining purpose-built robotic hardware with adaptable AI capabilities rather than relying on a fully vertically integrated technology stack. Its first major commercialization focus is 2F Culinary AI, a commercial kitchen robotics platform that includes the Bon Vivant 3.0 and Max systems, which use sensors, AI-driven controls, and programmable workflows to automate food preparation. GMEX has also identified transportation and logistics, industrial automation, and resource exploration as longer-term expansion areas. In March 2026, the company announced an AU$4.2 million agreement covering at least 50 intelligent kitchen robotics systems for an Australian hospitality group, followed by its first deployment order in May.
You recently joined GMEX Robotics as President of U.S. Operations after a career spanning financial services, entrepreneurship, technology recruiting, and enterprise relationships. What convinced you that physical AI and robotics represented the right next chapter, and what are your priorities as you build GMEX’s presence in the U.S.?
I saw the value in AI and robots, particularly around social intelligence, which is happening right now. I wanted to be contributing to it. What excites me about GMEX is the Terminal + Brain model. We’re not trying to do everything ourselves. We build hardware that’s reliable and practical, and we connect it to best-in-class AI. That’s a completely different approach than how most robotics companies operate. Right now I’m talking to investors about what we’re actually building and establishing key partnerships in the U.S. This is the right moment for this.
GMEX describes its strategy as a “Terminal + Brain” ecosystem in which robotics hardware connects the physical world with an AI intelligence layer. Can you explain how this architecture works and what you believe differentiates it from more vertically integrated approaches to robotics?
Most robotics companies try to build everything themselves; the hardware, the software, the AI. That’s expensive and slow. We do it differently. The Terminal is our hardware. The Brain is our AI orchestration layer, which we build with partnerships like 247meta. The advantage is flexibility. If better AI becomes available, we integrate it without rebuilding the robot. We’re not locked into one bet. We focus on what we’re best at; reliable, practical hardware, and we partner with best-in-class AI instead of trying to beat everyone on the tech side alone. That means faster iteration, lower capital, and we move at the speed of AI innovation, not mechanical engineering timelines.
GMEX has entered into an agreement to acquire an initial stake in MediaMeta.ai, whose technology focuses on social intelligence and human behavioral modeling. What does “social intelligence” mean in the context of a robot, and what new capabilities could it unlock that today’s multimodal AI models cannot reliably provide?
Social intelligence means robots that think about people, not just tasks. Today’s AI is great at vision, language, and reasoning. But it’s weak at reading a person, their emotional state, what they actually need, how to work with them in a real environment. We signed a definitive agreement with MediaMeta because their focus on human behavioral modeling addresses exactly that gap. That’s where MediaMeta’s longer-term roadmap is headed, developing social-intelligence AI and behavioral modeling as a contextual understanding layer that could eventually complement our Terminal + Brain ecosystem. Subject to development, integration and testing, these capabilities could help robots interpret context, communication patterns and human intent which is the difference between a robot that completes a task and one that’s genuinely useful.
Where do today’s most advanced robots still struggle when interpreting people? Are the biggest limitations around recognizing emotions and intent, understanding social context, adapting to cultural differences, or something more fundamental?
All of the above. But the real problem is that AI development has optimized for task completion can the robot pick up the object, navigate the room? That’s measurable. Reading people isn’t. So it gets less attention. Robots can recognize that someone looks upset, but they’re weak at understanding why they’re upset or what would actually help. Even small mistakes erode trust. A robot that gets 90% of social cues right sounds good in a test environment, but in the real world, that remaining 10% matters. People decide they don’t want to work with that machine. We need to flip the priority. That’s what we’re building at GMEX.
A robot operating in a warehouse can often function within relatively structured rules, while a robot in a hospital, hotel, school, or home must deal with far more unpredictable human behavior. How does the AI stack need to change as robots move into these more socially complex environments?
Social intelligence. In a warehouse, you can encode the rules. The environment is predictable. You move a robot into a space with people, everything changes. People improvise, they change their minds, they have bad days. The AI needs to read that in real time and adapt. It’s not rules-based anymore. It’s contextual. The robot has to understand the moment, the person, the environment and decide what makes sense. That requires fast connectivity and a completely different architecture underneath. It also needs to know when to ask for help instead of guessing wrong.
Human behavioral modeling inevitably raises questions around privacy, consent, bias, and misinterpreting human intentions. How should developers build social intelligence into robots without creating machines that become overly intrusive or make potentially consequential assumptions about people?
At the end of the day, building social intelligence into robots is about being helpful without being weird. Robots shouldn’t psychoanalyze people or guess what someone “meant” that’s how you get privacy issues and bad assumptions. Stick to what’s observable: how someone moves, how close they are, gestures. Give people control with obvious privacy cues and simple opt-in interactions. Keep the data footprint small, process locally when you can, and audit behavior so it treats everyone the same. MediaMeta is formalizing a Data Protection and Responsible AI framework designed to cover data minimization, transparency, bias review, human oversight, and traceability. That’s the kind of thoughtful approach we want integrated into our systems. If the robot isn’t sure what someone wants, it should slow down, give space, or pause. That’s how you build social awareness without crossing lines.
How do you actually evaluate whether a robot has become better at understanding humans? What types of benchmarks, simulations, or real-world testing are needed to measure social intelligence rather than simply whether a robot completed a task successfully?
You can’t just look at whether it finished a task. You look at micro-behaviors: does the robot give comfortable spacing, yield when someone steps into its path, move with the right timing, hesitate when intent is unclear? In simulation, you stress-test with dense crowds, sudden stops, ambiguous gestures. In the real world, you watch whether the robot feels predictable and comfortable to people around it, and whether it behaves consistently across different groups. That’s the real benchmark not task completion, but whether humans feel the robot “gets” how to move and coordinate in shared space.
GMEX has already begun commercializing its Bon Vivant cooking robotics platform. What have these early real-world deployments taught the company about the gap between demonstrating an AI-powered robot and operating one reliably in a commercial environment?
There’s a real gap between proving a robot can perform a task and proving it can do that task reliably, day after day, in a commercial environment. In a demo, everything is controlled. Real kitchens are high-heat, high-volume, unpredictable equipment gets bumped, ingredients vary, humans constantly change the rhythm. Bon Vivant is stationary, so the challenge isn’t navigation; it’s repeatability, durability, error recovery, and staying integrated with existing workflows. We learned that reliability comes from handling the “boring” stuff: tolerating heat and grease, staying calibrated, recovering gracefully when something is off-script, keeping pace without slowing humans down. Commercial robotics isn’t about proving the robot can cook. It’s proving it can cook consistently, safely, without becoming a burden in a real business.
GMEX is also developing a multi-agent AI platform designed to coordinate different models and workflows. How do you see digital AI agents and physical robots eventually converging, and could the same intelligence layer ultimately coordinate both software agents and fleets of machines?
Digital AI agents and physical robots will increasingly converge around a shared coordination layer. Software agents can interpret requests and manage digital workflows. Robots perform tasks in the physical world. A shared intelligence layer connects the two by translating a business objective into tasks, selecting the right models and agents, and coordinating execution. The key distinction is coordination versus control the shared layer decides what needs to happen and who should do it, while robot-specific systems retain responsibility for navigation, motion control and safety. In a warehouse, for example, software agents process an order and check inventory, while the coordination layer requests a suitable robot to move the goods. That feedback loops back and informs the next step. Human oversight and safety mechanisms stay essential. That’s where we’re focused with 247meta a staged path through interface integration, simulation and controlled pilots to support coordinated workflows across enterprise software and physical machines.
If social intelligence becomes a foundational layer of physical AI, what do you believe it will change about the robotics industry? Could understanding human behavior ultimately prove as important to widespread robot adoption as advances in dexterity, computer vision, and reasoning?
Social intelligence will reshape robotics fundamentally. The industry has focused on hardware capability, vision systems, performance all important but none of that guarantees a robot operates smoothly around people in a kitchen or on a production floor. Social intelligence is what makes a robot predictable, easy to work around, aligned with how humans naturally move and coordinate. As robots shift from controlled demos into real commercial settings, that part of the system becomes just as important as the hardware and the control stack. Understanding human behavior will matter just as much as advances in perception or reliability. It’s the difference between a robot that can do the job and one that fits into the flow of real operations.
Thank you for the great interview, readers who wish to learn more about this robotics company should visit GMEX Robotics.