The Deployment Is Not the Interesting Part

At the Davos Tech Summit this week, Pudu Robotics staged what it called "Robot City," a demonstration of commercial service robots deployed across retail, hospitality, and public spaces. The press release language is predictable: real-world value creation, physical AI, everyday life. What is not predictable, and what the announcement quietly exposes, is a coordination problem that neither Pudu nor the summit organizers appear to be treating as a problem at all. When you place algorithmically-mediated agents into organizations staffed by human workers who have no structural understanding of how those agents make decisions, you have not solved a labor problem. You have created a new one.

Physical AI Is Still Algorithmic Mediation

The framing around service robots tends to emphasize the physical dimension, the tray delivery, the floor navigation, the customer-facing interaction. But a Pudu robot operating in a hotel lobby is coordinating the same way a content recommendation system coordinates: it is allocating tasks, sequencing interactions, and prioritizing actions through algorithmic logic that is not visible to the human workers operating alongside it. This is precisely what Kellogg, Valentine, and Christin (2020) identified as the defining feature of algorithmic work arrangements - the algorithm becomes the de facto manager, and human workers must adapt to its outputs without reliable access to its inputs. The hospitality worker who now shares a shift with a Pudu robot is in a structurally identical position to the gig worker trying to infer surge pricing logic. The environment is different. The coordination problem is the same.

The Awareness-Capability Gap in Physical Space

Here is what concerns me about the Davos demonstration. The robot performs. Humans observe. Stakeholders applaud. But nobody at Robot City is asking what competence the surrounding workforce needs to develop in order to actually coordinate with these systems rather than simply coexist with them. Algorithmic literacy research has documented this gap repeatedly: workers develop awareness that an algorithm governs their environment, but awareness does not translate into improved coordination outcomes (Gagrain, Naab, and Grub, 2024). Physical deployment makes this worse, not better, because the robot's presence is legible in a way its decision logic is not. A worker can see the robot. The worker cannot see why the robot prioritized table four over table seven. The visibility of the agent creates an illusion of transparency about the system.

Routine Expertise Will Not Transfer Here

Organizational responses to service robot deployment almost universally follow a training template built around procedural knowledge: here is how the robot moves, here is how to call it, here is what to do when it stops. This is precisely what Hatano and Inagaki (1986) identified as routine expertise - competence that is stable when conditions are stable and brittle when conditions shift. A robot firmware update, a room reconfiguration, a change in service protocols: any of these will break the procedural knowledge a worker acquired during onboarding. What the worker actually needs is schema-level understanding of how the robot allocates decisions, so that they can adapt when the topography shifts. Gentner's (1983) structure-mapping framework predicts that workers trained on structural features of the coordination system will outperform workers trained on platform-specific procedures when the environment changes. The Davos demonstration offers no evidence that Pudu or its clients are investing in that kind of preparation.

The Organizational Theory Problem Underneath the PR

Schor et al. (2020) argued that platform dependence creates precarity not primarily through income volatility but through the worker's fundamental inability to model the system governing their outcomes. That argument was developed in the context of gig work, but it applies directly to the hospitality or retail worker whose job is now partly defined by their coordination with an autonomous agent they cannot interrogate. Pudu Robotics is commercially successful. The robots work. The deployment numbers are real. None of that resolves the question of whether the organizations adopting these systems are building the internal competence structures that would make human-robot coordination sustainable rather than fragile. Right now, the evidence from the Davos framing suggests the answer is no. The demonstration was designed to show that robots can operate in human environments. The harder demonstration - whether humans can develop adaptive expertise within robot-mediated environments - was not on the agenda.

What This Means for Coordination Theory

The ALC framework I am developing treats platform coordination as a mechanism where competencies must develop endogenously through participation. Physical AI deployment extends that argument into embodied organizational space. The power-law variance we observe in digital platform outcomes - where identical access produces dramatically different results - should be expected in human-robot work environments for the same structural reasons. Initial differences in schema-level understanding of the robot's coordination logic will be amplified by the system itself. Some workers will adapt. Others will develop increasingly rigid workarounds. The Davos demonstration is a useful event not because it shows AI entering everyday life, but because it marks the moment when the coordination problem I study stopped being confined to screens.