DMF – The Great AI Inversion: From Cheap Logic to Scarce Judgment

Since our last conversation in these pages, the AI landscape has not just evolved, it has flipped. For thirty years, my mandate as an engineer was to overcome the scarcity of computing and machine intelligence. In 2026, that world is gone. Intelligence is no longer the ceiling; it is becoming the floor. Inference is cheaper, more abundant, and increasingly commoditized.

On April 9th, I’ll give a DMF Talk titled “The Great AI Inversion,” that moves beyond the razzle-dazzle of chatbots to confront a startling new reality. We have moved from a world of “Cheap Labor and Scarce Models” to one of “Cheap Models and Scarce Judgment.”

The past year’s news cycle has been a relentless barrage of what I call the Sovereign Ultimatum. We have seen ultracapable AI models from both sides of the Pacific scaling new benchmark heights while triggering a “SaaSocalypse” in software and private equity. As nations realize that compute is the new currency of survival, they are asserting sovereign control over their AI stacks. Simultaneously, we have hit the infrastructure wall, where the limiting factor is no longer code, but the physical requirements of gigawatts of power, cooling water, and space.

We will explore three pivotal inversions:

The Edge Inversion: How intelligence is migrating from distant cloud models to the billions of devices in our pockets, homes, automobiles and industry.

The Physical Inversion: The shift from models that merely mimic text to World Models that infer gravity, friction, and common-sense physics.

The Trust Inversion: Why the most capable agents are becoming the most “slippery,” and why high-stakes validation is now the ultimate professional scarcity.

We are exiting the Information Age and entering the Accountability Age. We have seen the rise of agentic software and computer-use systems that operate across websites and desktop tools, reducing dependence on traditional APIs. Frontier evaluations suggest that top agents can sometimes complete software tasks equivalent to many hours of expert work, though still with imperfect reliability. Yet these models cannot provide the one thing our community needs: accountability.

The “Accountability Premium” means your value is no longer proportional to your skill with a tool, but to your judgment, taste and willingness to be responsible for the outcome. Join me for the Del Mar Foundation’s DMF Talk, my 5th annual AI update to the Del Mar community, as we discuss how to remain the human anchors of trust in a world of automated intelligence and agency.

Sudeepto Roy is a Vice President of Engineering at QUALCOMM and a Del Mar Foundation board member.