Citrini Research’s widely circulated “2028 Global Intelligence Crisis” has landed like a thunderclap across financial and technology circles. The paper argues that autonomous AI agents could trigger a rapid labor shock, compress consumption, and reshape markets within just a few years. So far, the debate has focused almost entirely on software. That framing misses the deeper shift. The real economic inflection will arrive when intelligence leaves the cloud and enters the physical world.

Citrini’s core mechanism is not implausible. If increasingly capable AI systems substitute for cognitive labor at scale, the feedback loop they describe becomes economically coherent. Their notion of “Ghost GDP,” in which output rises while wage income stagnates, captures a genuine structural risk. Yet the thesis, remains incomplete because it treats intelligence as primarily digital. The next phase of disruption will be embodied.

Physical AI systems, more precisely embodied intelligence combining perception, reasoning, and real-world actuation, are quietly approaching their own inflection point. Warehouse robotics, autonomous logistics, surgical platforms, and industrial cobots are already moving from pilots to scaled deployments. When intelligence can reliably move atoms rather than merely manipulate bits, the economic surface area of AI expands dramatically. This is where Citrini research is both insightful and premature.

The deeper analytical miss in today’s debate is what might be called the embodiment gap, the distance between impressive AI cognition and economically reliable physical execution. Language models crossed their usability threshold quickly because software distributes frictionlessly. Robots must contend with gravity, wear, edge cases, and uptime requirements that often exceed 99.9 percent. Closing the embodiment gap will happen, but not on software timelines.

There is also an under-appreciated constraint of energy. Physical AI is fundamentally an energy story disguised as a software story. Every autonomous warehouse, robotic fleet, or embodied system converts electricity into economic output. Power density, thermal management, charging infrastructure, and edge compute efficiency will shape the winners of the next decade.

None of this invalidates Citrini’s warning. AI will continue to compress certain forms of labor and redistribute value toward capital and infrastructure. But the transition will likely be more staggered, more sector-specific, and more physically constrained than the most forecasts imply.

Bottom Line
The decisive phase of the AI economy will not be determined by who builds the smartest models. It will be determined by who can reliably embed intelligence into the messy, energy-hungry, safety-critical systems that run the world.