Envisioning Headless Software: How Physical AI Could Transform the Manufacturing Shop Floor

How Humans Communicate?

From the earliest days of human civilisation, communication was closely connected to our senses and physical presence.

We communicated through our voices. We listened to others. We observed expressions and movements. A look could convey a message, a nod could acknowledge something, and a gesture could communicate an instruction without a single word being spoken.

These were natural ways for humans to communicate and understand one another.

Written communication changed how we preserved and shared knowledge. What was written could be read by someone who was not present when it was created. Knowledge could travel across places and generations.

Today, we can still read works written centuries ago in classical languages such as Tamil, Chinese and Greek. Writing has allowed ideas and knowledge to survive long after their authors are gone.

Then came the computer.

The Computer Introduced a Sixth Mode of Interaction

The arrival of computers fundamentally changed how we interacted with information and systems.

We could now interact with a machine through a new mode: the graphical user interface (GUI).

Keyboards, screens, buttons, menus, forms and dashboards became the primary means through which we interacted with software. We entered information. The computer processed it. The results appeared on a screen. Over time, almost every business process began to follow this pattern.

Need to understand production performance? Open a dashboard.

Need to report a machine breakdown? Enter the details into a system.

Need to know the status of an order? Navigate through an application.

Need to communicate an instruction to another department? Update a system and expect someone to read it.

This pattern extended across enterprise software, from ERP and manufacturing execution systems (MES) to maintenance applications and industrial monitoring dashboards.

Software became increasingly capable, but our interaction with it became centred around screens.

We built applications for almost every function, and each application brought its own interface. The more software we introduced into an organisation, the more time people spent moving between screens and applications.

The computer gave us a powerful new mode of interaction. But somewhere along the way, we began designing our digital world as though every interaction had to pass through it.

Physical AI: Bringing Our Natural Modes of Communication Back to Life

What if we could interact with software, machines and the entities around us in the same natural ways we have communicated with one another for thousands of years?

We could speak to them. They could listen to us. They could observe what is happening around them. They could interpret gestures, recognise situations and respond in ways that make sense in the physical world.

Physical AI in manufacturing is opening up possibilities in this direction—not as a replacement for every existing system, but as a way to connect industrial AI with what people and machines actually do on the plant floor.

Industry analysts have begun to describe this convergence of AI with physical operations. The World Economic Forum’s report Physical AI: Powering the New Age of Industrial Operations outlines how AI that perceives and acts in real environments could change industrial work. I treat that as a directional view, not a forecast the shop floor must meet tomorrow—but it matches the question this article explores: whether software can meet people and machines through interaction that fits the physical world.

Voice can become a way to ask a system for information or give it an instruction. Vision can help a system understand what is happening on the shop floor. Sound and other sensor signals can provide clues about the condition of a machine. Gestures and physical actions can become part of an interaction.

AI systems that work with different forms of input, including language, images, video and sensor data, are often described as multimodal AI. In manufacturing, these capabilities could help connect human instructions, machine data and real-world operating conditions.

Consider a maintenance technician standing beside a machine. Instead of walking to a terminal, opening an application and navigating through several screens, the technician could ask about the machine's failure history and receive a relevant response.

Consider a production supervisor who needs to know why a work centre has stopped. Instead of checking multiple dashboards, the system could analyse machine data, identify a possible cause and communicate the relevant information directly.

These examples are not simply about replacing a keyboard with a voice command. They represent a shift in how software, machines and people can interact with one another and with the physical environment—and how that shift could support headless software on the shop floor.

The underlying software may still use databases, APIs, industrial protocols and complex processing. What changes is how people access its capabilities and how systems communicate with the world around them.

Physical AI has the potential to bring our natural modes of communication back into our interactions with software, machines and the physical world.

The Era of Headless Software

This brings me to an idea I have been thinking about: headless software everywhere, including across the shop floor.

Traditionally, software has been built around an interface that people use to access its capabilities. Even when the underlying architecture is sophisticated, the user often needs to open an application to get something done.

Headless software takes a different approach. Its core capabilities operate independently of a dedicated graphical user interface—they remain available to other programs, workflows and devices while the GUI becomes optional rather than mandatory. Physical AI could become one way to reach those capabilities: through voice, vision or context on the plant floor instead of through another screen.

The idea of headless software is not new. APIs, industrial protocols, event-driven architectures and machine-to-machine communication have allowed systems to exchange information and operate without a person interacting with a screen for years.

Industrial IoT (IIoT) and practical industrial connectivity have also made it possible to connect machines, collect operational data and exchange information across industrial systems—provided the operational questions and data foundations are clear first.

What is changing is the possibility of combining these capabilities with AI systems that can interpret context, understand requests and interact with the physical environment.

AI agents in manufacturing could, for example, retrieve information from different applications, coordinate workflows and present the relevant information to a person without requiring that person to navigate through each application individually.

A user may no longer need to know which application contains the information. A machine may not need a human to enter every event into a system. A workflow may not need someone to monitor a dashboard before the next action begins.

The software capability remains. The need for a dedicated interface in every interaction begins to diminish.

This is where I see Physical AI accelerating the adoption of headless software—not by removing accountability or engineering discipline, but by offering interaction paths that fit how work actually happens.

A Shop Floor Where Machines and People Communicate Directly

Imagine a shop floor where communication can happen in every direction.

A machine communicates its operating condition to another system. An AI system analyses the information and identifies a developing problem. The relevant person receives an explanation through voice or another appropriate channel. The person responds, and an authorised workflow is initiated.

This involves three fundamental forms of interaction:

• Machine to Machine (M2M): Machines and systems exchange information and coordinate activities.
• Human to Machine (H2M): People communicate requests, instructions and decisions to machines or software.
• Machine to Human (M2H): Machines and software communicate information, warnings, recommendations and results to people.

Together, these forms of human-machine interaction and machine-to-machine communication could help create a more connected and responsive factory—one that can detect, explain and act without every step starting at a terminal.

On the robotics side, programmes such as NIST’s Physical AI and Data Generation for Robotics focus on the data and models robots need to work safely in real environments. That matters for manufacturing when headless software must coordinate sensors, automated equipment and human oversight—not when robots alone are the goal.

Consider a material shortage that threatens to stop a work centre.

The production system identifies the risk. It checks available material information and the production schedule. The planner receives a notification explaining the potential impact. An approved workflow can then initiate the next step, whether that involves checking stock, notifying purchasing or revising the production plan.

The interaction does not have to begin with someone opening an application and searching for the information.

Of course, not every decision should be automated. Actions affecting production, safety, quality or equipment must operate within clearly defined permissions and safeguards. People should remain involved wherever judgement and accountability are required—the same principle I argue in knowing the factory before adding AI.

The objective is not to eliminate every screen or remove people from the process. Some tasks will always benefit from a visual interface. Complex planning, engineering, configuration and detailed analysis may continue to require one.

The opportunity is to stop making the screen the mandatory destination for every interaction.

Beyond the Shop Floor

Although manufacturing makes this vision particularly interesting, the idea is not limited to factories.

Across industries, software could increasingly operate as a capability accessed through the most appropriate form of interaction for the situation.

A field service engineer could ask for equipment history while working on site. A logistics system could communicate an exception to the person responsible for resolving it. Connected equipment could exchange information with other systems without requiring an operator to intervene.

The interface would depend on the task, the environment and the person involved.

Sometimes it would be a screen. Sometimes it would be a voice interaction. Sometimes it would be a machine signal or an automated exchange between systems.

In many situations, the user might not even be aware of which software components are involved.

This could change how we approach enterprise software and manufacturing digital transformation. Instead of expecting people to move between applications, we could make software capabilities available wherever and whenever they are needed—while still sequencing investments from how the factory works today, not from a generic technology checklist.

The same principle could apply to industrial automation, building management, logistics, energy systems and other environments where people, software and physical assets need to work together.

The Future I Envision

For decades, we have invested in making software easier to use through better screens, simpler navigation and more intuitive interfaces.

That work will continue to matter. But I believe we are approaching another shift.

The next evolution may not be about making every software interface better. It may be about identifying where an interface is needed at all.

Our earliest forms of communication were natural, direct and closely connected to the physical world. Computers introduced a new mode of interaction that transformed how we worked with information. Physical AI is creating opportunities to reconnect software with our natural modes of communication and with the machines and environments in which we operate.

Headless software is not about removing user interfaces altogether. It is about freeing software capabilities from the assumption that every interaction requires a dedicated screen or application—and letting Physical AI in manufacturing provide natural access where that genuinely helps.

For manufacturing, this could mean moving towards a shop floor where machines exchange information directly, AI helps interpret operating conditions, and people interact with systems in ways that fit their work.

For other industries, the same idea could change how software becomes part of everyday operations.

This is what leads me to envision headless software across the shop floor and beyond.

If machines can communicate with machines, people can communicate naturally with software, and software can communicate directly with people and the physical world, how much of our future interaction with software will still need to begin with opening an application?