The now-famous dystopian note from Citrini Research has already sparked many reactions. Its 2028 timeline is intentionally provocative, and probably unlikely if read literally. Still, it is difficult to ignore the internal coherence of the spiral it describes.

For my part, I like to think that analogies—historical as well as physical—help us make sense of complex and multifactorial phenomena. Neither predicts the future, but both improve our ability to ask the right questions.

Below are a few thoughts that I draw from reading the Citrini note, as well as the many insightful discussions it has triggered.

When Rational Decisions Aggregate

At enterprise level, the logic is straightforward: automate where possible, reduce cognitive labor costs, protect margins, move faster. At system level, things can invert.

If income diffusion persistently lags behind productivity diffusion, aggregate demand can weaken. If demand weakens, growth assumptions become fragile. If assumptions become fragile, financing tightens. And tighter financing pushes everyone to optimize harder.

Here is the paradox: each decision is rational in isolation, but their aggregate can still become unstable.

None of this contradicts the possibility that AI may generate large productivity gains and welfare improvements. The question is whether distribution and adjustment mechanisms will keep pace with these gains.

Friction as a Stabilizer

The second point is friction. In many tech narratives, friction is framed as a bug. Remove friction, increase speed, win.

In complex economies, friction is often a stabilizer. Delays, governance constraints, integration complexity, organizational inertia, and social negotiation all look inefficient from a local perspective. But these same elements prevent synchronized overshoot.

Yves Caseau made this point in his comment on the Citrini note, referring to the "viscosity" of enterprise software and labor substitution, as factors that may dampen the upcoming transition. So friction is not a defect in transitions of this magnitude, but rather one of the conditions for stability.

A useful analogy here comes to mind from fluid dynamics: in a low-viscosity system, flows accelerate quickly and disturbances propagate rapidly. When multiple flows synchronize, nonlinear behavior becomes more likely. High-viscosity systems evolve more slowly but tend to remain more stable.

Economic friction plays a similar role. Institutional delay, integration costs, governance constraints, and social bargaining act like viscosity. They slow adjustment and reduce surge risk. And this is where the present moment is unusual. We are not reducing friction in one layer only. The AI industry is reducing it across many layers at once: software integration, where each successful integration reduces friction for the next one and AI coding tools are also a self-improving industry; capital deployment, where circular financing and infrastructure scaling strengthen the case for further deployment; AI model distribution, with general availability across multiple hyperscalers and model distillation accelerating performance catch-up; AI talent mobility, with frequent moves in high-demand jobs fostering technology diffusion; and deployment cycles, where shorter cycles tighten feedback loops and speed up improvement.

Few historical transitions combined such breadth with such speed. When positive loops accelerate in a low-viscosity environment, nonlinear behavior becomes more likely.

Enclosing the Cognitive Commons

If friction governs the speed of transition, enclosure dynamics govern how value flows during the transition.

In her own commentary on the Citrini note, Anastasia Stasenko points to something I believe is essential: privatization of a commons.

That idea sends us back to a foundational episode in political economy: the Enclosure movement in 16th-century England, which refers to the gradual transformation of common lands into privately owned property. For centuries, many rural communities had shared rights over common fields and pastures, using them collectively for cultivation and subsistence. From the early modern period onward, these shared lands were progressively enclosed—physically through hedges and fences, and legally through acts of Parliament—and reorganized into private estates.

Many economic historians attribute significant long-term effects to the enclosure movement. By concentrating land ownership and reorganizing production, enclosures are often seen as having increased agricultural productivity while simultaneously accelerating rural displacement and the emergence of a wage-dependent labor force—developments that helped lay some of the social and economic foundations of the Industrial Revolution.

The idea of a modern "enclosure of the information commons" has been studied since the early 2000s, with several authors pointing out a scenario in which ideas and knowledge—once treated as shared resources—become progressively appropriated and organized into private economic assets. Twenty years after, the Internet and LLMs have made it happen at machine scale.

So the analogy between England's sixteenth-century enclosures and the generative AI transition is not literal, but it is structurally powerful: today, a large part of the training substrate of modern AI comes from a broad cognitive commons—scientific work, books, code, archives, public discourse, and cultural output accumulated across generations. This layer is aggregated, operationalized, and monetized at industrial scale.

The enclosure here is not primarily about restricting access to knowledge, since knowledge remains widely accessible in the digital age—actually more than ever. But it is about concentrating the ability to operationalize knowledge at industrial scale, while collapsing its unit economics—see Shuwei Fang's note on the brutal economics of liquid content.

So the question might be this one: what happens when value extracted from a shared cognitive base is captured faster than it is redistributed?

Where to Migrate?

Because there is also an important difference with Britain's sixteenth-century enclosures.

In earlier transitions, displaced workers could eventually migrate to adjacent sectors as new production systems expanded. Slow, painful, unequal, but possible.

In the current transition, migration channels may be narrower. Automation is increasingly targeting transversal cognitive tasks—precisely the tasks that historically absorbed transitions in service economies.

If substitution reaches broad coordination layers, reallocation may become slower and less obvious than standard historical analogies suggest.

This is why simple "technology always creates new jobs" narratives feel insufficient in this case.

Absorbing Accelerated Flows

Another structural lens concerns how accelerated economic flows are absorbed.

High-velocity systems require mechanisms that slow and distribute adjustment. Historically, much of this absorption occurred through social and institutional channels: wages, social transfers, administrative layers, negotiation cycles, regulatory pacing.

These mechanisms spread productivity gains over time and across actors. The speed and magnitude of today's transition—spanning across a few years only, at global scale—somehow weaken these social absorption mechanisms.

At the same time, a growing share of accelerated capital flows is absorbed by compute infrastructure itself: chips, power supply, data centers, cooling systems, hardware replacement cycles.

Compute is primarily a productive input, not a pure loss. But unlike many digital investments, compute ultimately converts large volumes of capital expenditure into physical energy consumption, which itself does not distribute much wealth, but rather weighs over the economy through energy costs and power grid challenges.

The Intertemporal Closure Problem

Ultimately, the Citrini note describes a Minsky-type paradox that becomes quite clear when one considers the economic equation from an intertemporal perspective. The fragility of the system certainly doesn't come from technological underperformance. But it comes from the possibility that financing assumptions and macro-distribution dynamics stop aligning over time.

AI is financed today against expectations of tomorrow's returns. But those returns depend on tomorrow's solvency. If value concentration and labor substitution outpace income diffusion, the economy that is supposed to finance AI adoption may weaken.

This is an intertemporal closure problem, not a capability problem.

Conclusion: The Stability Question

At minimum, this transition cannot be governed by market dynamics alone, nor by industrial policy slogans alone. If part of a global cognitive commons is being enclosed, the return channel to society cannot remain an afterthought. The missing layer is institutional design for value circulation.

So yes, Citrini's note may be too compressed as a timeline. But it raises valid points, especially in the light of analogies drawn from history or the physics of complex systems.

The big question ahead is whether economic institutions will evolve fast enough to maintain a viable circulation of income and demand as productivity accelerates. A low-friction economy can be extraordinarily dynamic—but without sufficient mechanisms for redistribution, negotiation, and adjustment, it can also become fragile.

The challenge is therefore not to slow AI progress, but to ensure that productivity gains translate into broadly shared purchasing power. Without that circulation, the very economy expected to finance AI's expansion may weaken over time.

And if that happens, the instability described by Citrini would not be a technological failure—but an economic one.