AI tokenomics

This video is also published on YouTube. It was produced using Google NotebookLM and is based on my exploratory system dynamics model of AI tokenomics.

Please note: The model draws on published research and my professional experience. It is intended for conceptual exploration rather than prediction. No proprietary, operational, or empirical data were used to calibrate or validate the model; the relationships, scenarios, and numerical outputs shown are illustrative.

AI Tokenomics Needs a System Dynamics Model. Here’s Why the Spreadsheet Is Not Enough.

The conversation about AI token economics has settled into a familiar shape: route intelligently, match the model to the task, monitor the Jevons paradox, and capture the savings.

All useful advice.

But notice what is actually being described underneath it.

It is not simply a ledger. It is a system.

Chatbots become agents. Agents become coordinated teams of agents. Falling token prices do not necessarily reduce total spending. They can pull more work into AI because tasks that were previously too expensive suddenly become economically viable.

Freed capacity is expected to move into higher-value work. That value can fund additional AI deployment, which may free even more capacity.

Each of these mechanisms contains a feedback loop. Several also operate through significant delays.

That is the limitation of reading AI economics primarily through the P&L.

A cost line is a snapshot. What leaders are managing is a set of interacting conditions that accumulate, change at different rates, and feed back on one another over time.

Static ROI can provide a useful financial baseline. But it cannot adequately represent how adoption, cost, capability, governance, organizational capacity, and risk evolve together.

It is not necessarily the wrong calculation.

It is an incomplete representation of the system producing the return.

This is precisely the class of problem system dynamics was designed to examine:

Stocks and flows.

Reinforcing and balancing feedback loops.

Nonlinear relationships.

Delays between a decision and its consequences.

When leaders say they cannot isolate AI’s contribution from other factors, the problem is not always missing data.

Often, it is interdependence.

AI-enabled outcomes emerge from changes in workflows, workforce behavior, model capability, process design, governance, demand, and investment. A linear model will struggle to assign a clean contribution to one factor when the result is being produced by the interaction among all of them.

The answer is not simply a better spreadsheet.

It is a model of the system.

That model could include:

• Stocks that accumulate: deployed AI workflows, organizational AI capability, liberated capacity, value successfully recaptured, governance maturity, technical debt, and security exposure.

• Flows that change those stocks: workflow deployment and retirement, adoption, token consumption, capacity liberation, reinvestment, governance improvement, and the accumulation or mitigation of risk.

• External drivers and decision rules: token prices, model capability, routing policies, budget limits, governance thresholds, and infrastructure constraints.

• Feedback loops that connect them: the Jevons loop, agent proliferation, the capacity-reinvestment loop, and the reinforcing cycle that quietly underwrites many AI ROI claims.

A credible model must also represent the balancing forces.

Budgets become constrained. Integration capacity reaches its limits. Governance reviews create delays. The marginal value of additional use cases may decline. Security exposure can trigger tighter controls. Employees and workflows can absorb only so much change at once.

These balancing loops determine whether expanding AI use produces greater value, greater consumption, greater risk—or some combination of all three.

Model these structures, and AI ROI stops being only a number defended after implementation.

It becomes a system that can be simulated before commitment.

Leaders can test how a routing policy, a decline in token prices, faster agent adoption, a governance bottleneck, or a steeper reinvestment rate might ripple through cost, capacity, risk, and return across near-, medium-, and longer-term horizons.

The cost side is the easier half.

The system is the whole picture.

The next question is not simply how much AI will cost.

It is which feedback structures will determine whether falling costs become greater value, greater consumption, or both.

Diagram to follow—mapping the stocks, flows, drivers, and feedback loops shaping AI token economics.

Who is already modeling this system rather than budgeting against a snapshot?