The Calculation Problem in the Age of Tokens

With the recent release of the latest large models, the media is again full of talk about "AGI" and the end of work. But from where I sit—balancing the physical requirements of industrial facilities with the deductive logic of Austrian economics—most people are missing the real story.

We are treating AI like a magic oracle. But in economic terms, an AI output is just a "Token"—a unit of processed information. The real question isn't how "smart" the model is; it’s how we allocate the massive amounts of capital (energy, silicon, and human time) required to produce those tokens.

Ludwig von Mises famously argued that without price signals, central planners cannot calculate. Right now, AI is centrally planned. We have big models "serving" users. But for AI to truly integrate into our economy, it has to move from being a "service" to being an economic actor that can respond to price signals in real-time.

I suspect we are nearing the end of the "Chatbot" era. The next phase won't be a more talkative AI; it will be an Agentic AI that has to justify its own marginal cost.

Comments

Post a Comment

Popular posts from this blog

The Liability Wall: Engineering Intent in the Age of 'Decide-to-Pay'

Image
The financial world just moved past the 'biological bottleneck'. With the release of IMF Note 2026/004 , the transition from ' Click-to-Pay'  to 'Decide-to-Pay'  has been codified into institutional design. For those of us managing the physical and digital infrastructure of the modern corporations, this isn't just a technical upgrade—it is a fundamental shift in the nature of commercial liability and labor allocation. By introducing a three-layer framework— Intent, Authorization, and Settlement —the IMF is creating a 'legal airlock' between the probabilistic reasoning of an AI agent and the deterministic finality of financial markets. The Wave: From Human Latency to Machine Autonomy For decades, Corporate Real Estate (CRE) and Facilities Management (FM) have operated on a 'human-in-the-loop' cadence. A sensor detects a chiller failure; a facility manager reviews the quote; a human signs the purchase order. The IMF’s formalization of 'Deci...
Read more

Hardware Sovereignty: Why Your Office Needs a GPU

Image
In the early 20th century, factories had to be built next to rivers or have their own massive steam engines to function. Then came the central power grid, and we moved toward a 'Utility' model where we outsourced our energy needs. For the last decade, we’ve done the same with AI. We outsourced our 'Intelligence' to the central cloud. But if the Agentic Revolution of the last few weeks has taught us anything, it’s that the centralized model is failing the industrial user. The Latency of Permission When you run an agentic loop—where an AI is making hundreds of micro-decisions a minute to optimize a physical facility—you cannot afford the 'Latency of Permission'. Technical Latency: The milliseconds lost sending data to a central server and back. Economic Latency: The risk of a central provider changing their API terms or 'throttling' your agent during a peak load. In Austrian terms, this is a Property Rights issue. If you don't own the hardware runn...
Read more

Fixed Assets vs. Fluid Intelligence: The FM Lens

Image
In the mainstream tech press, Artificial Intelligence is often described as 'fluid'. We hear about it 'flowing' into every sector, scaling infinitely in the cloud, and being accessible via a simple browser window. But as an  Facilities Management (FM) guru , I see it differently. When you spend your days managing the physical lifecycle of industrial assets, you learn a hard truth: Nothing is fluid without a fixed foundation.   The Physical Reality of "Tokens Every time an LLM generates a response, it isn't 'thinking' in a vacuum. It is consuming a specific amount of electricity, requiring a precise degree of cooling, and occupying a slice of high-bandwidth memory (HBM). In economic terms, AI is not a service; it is a processed output of fixed capital. If we apply the lens of Capital Heterogeneity (a favorite topic of Austrian economist Ludwig Lachmann), we see that the hardware running these models isn't just a generic 'computer'. It is a h...
Read more