Shaping the flow of ideas at the frontier of cognitive systems.
—
Ninety-five percent of financial transactions worldwide still run on COBOL, a programming language designed in 1959. Modern AI models can write replacement code in seconds. The capability exists. The deployment hasn’t happened. This gap — between what AI can technically do and what organizations have actually reorganized to use — is the force that every economic forecast about AI is currently ignoring.
The AI doom scenario assumes labor displacement will happen faster than new roles appear, triggering demand contraction and economic collapse. The boom scenario assumes productivity gains will compound into sustained growth as AI reorganizes entire industries. Both narratives share a hidden assumption: that organizational change happens on the same timescale as model improvement. It doesn’t. When Gemini’s reasoning scores double in three months while the average enterprise procurement cycle runs eighteen, the distance between capability and deployment isn’t a minor implementation detail. It’s the shape of the entire transition.
—
Call this the Capability-Dissipation Gap. On one axis, AI capability — the reasoning depth, task endurance, and cost efficiency that models can achieve. This curve is steep and accelerating. Costs are falling by orders of magnitude; reasoning benchmarks are climbing vertically. On the other axis, organizational dissipation — the rate at which companies, agencies, and institutions actually redesign workflows, retrain staff, and redeploy resources around new AI capabilities. This curve is slow, bound by forces no amount of model performance can compress.
Four friction forces govern dissipation speed. Regulatory inertia means financial services need regulatory approval before deploying AI in transaction flows; healthcare needs HIPAA clearance, FDA approval, and institutional review boards before patient-facing systems change. These are measured in years, not quarters. Organizational inertia means workflows and roles are designed around current capabilities — redesigning them requires change management and process reengineering that can’t be shortcut by releasing a better model. Institutional trust means deploying AI at scale with appropriate guardrails and audit trails takes time; trust is built incrementally, not purchased with a benchmark score. Integration complexity means the chain from “the model can do this” to “the economy has reorganized around the model doing this” includes deployment, adoption, and integration layers, each introducing delay.
The doom scenario is internally consistent if displacement outpaces adaptation. But it requires displacement speed that these friction forces historically haven’t permitted. Technologies have always created more jobs than they destroyed, but always on generational timelines — the timeline dissipation friction enforces. The boom scenario is internally consistent if productivity compounds into sustained demand. But it requires adoption speed that organizational reality won’t deliver. When the average Fortune 500 company still runs core systems on infrastructure designed two decades ago, the idea that AI reorganizes entire industries in three-year cycles is a fantasy about capability, not a forecast about dissipation.
—
The gap is not symmetric. Markets are currently pricing AI as simultaneously too weak to justify infrastructure spending and too strong for any incumbent to survive. IBM’s stock posted its worst single-day loss in twenty-five years after a blog post described a fictional 2028 scenario in which AI replaced COBOL systems. The memo wasn’t real. The panic was. This incoherence — AI as underperformer and existential threat in the same breath — reveals how poorly the current discourse maps to structural reality.
What the gap reveals is where professional value accumulates during the transition. When capability races ahead of dissipation, the people who can bridge the distance — who can take what AI can technically do and make it organizationally usable — hold leverage. This is not a story about learning to prompt. It’s a story about understanding which of your skills live in the capability-ahead zone, where models are already better than you, and which live in the dissipation-lag zone, where organizational friction means human judgment still commands a premium. The Shopify CEO built a personal evaluation framework to map exactly this: not “what can AI do?” but “where has AI reached the ceiling, and where does integration complexity still require me?”
—
Technological change doesn’t move at the speed of the technology. It moves at the speed of the slowest necessary system. AI capability is no longer that system. Organizational dissipation is. The gap between them is not a delay to be optimized away. It is the structure of the transition itself.
—
Source: “The economist who formalized the AI doom scenario says its own conditions are too extreme to hold” — Nate’s Newsletter, February 26, 2026.
Text summarized and optimized using Anthropic’s models and reviewed by a human.