The Price of Intelligence

In the late nineteenth century, industrial firms generated their own power. Factories were organised around steam engines. Production lines were designed around where power could be physically generated and distributed. Owning power was considered part of running a serious business because without it, nothing else could happen.

Then electricity arrived.

Gradually, power moved out of the factory and into the grid. Organisations stopped asking how much power they could generate and started asking how much power they needed. What had once been a strategic capability became a utility. The firms that adapted most successfully were not necessarily those that generated the most electricity, but those that learned how to redesign their economics around a world where power could simply be accessed.

More than a century later, something similar happened with computing. Organisations built data centres, purchased servers, hired infrastructure teams and managed technology as an owned asset. Cloud computing changed that equation. Computing became something companies could consume rather than possess. Capital expenditure shifted toward operating expenditure. Infrastructure became elastic. Scale became available on demand.

Artificial intelligence appears to be the next step in that progression. But unlike electricity or computing, what is becoming a utility is intelligence itself. And that changes the economics of organisations in ways that are easy to underestimate.

The cloud era was ultimately about the economics of computation. The AI era is increasingly becoming about the economics of intelligence.

Those are not the same thing.

Three Economic Models

Across this series, we have explored three broad organisational responses to AI.

Some organisations will build intelligence into their operating fabric. They will train models, develop proprietary agents and continuously refine them using their own data, workflows and institutional knowledge.

Others will participate in an agent economy, accessing intelligence through specialised providers in much the same way organisations accessed computing through the cloud.

A third group will remain predominantly human-centred, using AI selectively while continuing to organise work around people rather than systems.

These are often presented as technology choices. They are more accurately understood as economic choices.

Because each creates a different relationship between cost, margins, speed, risk and competitive advantage.

The Economics of Ownership

The agent-first organisations will make substantial investments in building intelligence that understands the unique context of their business.

For most of corporate history, organisations paid for labour while retaining most of the learning produced by that labour. The economics of AI introduces a new possibility.

In a highly agentic organisation, every customer interaction, operational decision and workflow execution contributes to a growing body of machine-readable intelligence. The learning no longer resides exclusively in people. Increasingly, it resides in systems.

When organisations build and train their own intelligence systems, the learning compounds internally. Every interaction improves future interactions. Every decision strengthens organisational capability. Intelligence begins to behave less like labour and more like accumulated capital.

This helps explain why some organisations are willing to make substantial investments despite uncertain short-term returns. They are attempting to convert knowledge into something that compounds.

Historically, firms accumulated advantage through factories, distribution networks, brands, software platforms and data. AI introduces the possibility that organisational intelligence itself becomes an asset class.

These organisations are hence effectively betting that proprietary intelligence will become as strategically important as proprietary technology was in the previous era.

The return comes from reducing dependence on external intelligence while building capabilities that become increasingly specific to the organisation itself.

For a small number of organisations, that trade-off will be attractive. For most, it will not.

The Agent Economy

Most organisations are likely to occupy the middle ground. Just as most firms never built hyperscale cloud infrastructure, most will not train frontier intelligence systems. They will access intelligence through an increasingly sophisticated marketplace of agents.

At first glance, the economics appear compelling. Capabilities arrive quickly. Capital requirements are lower. Providers absorb much of the complexity associated with model development. Improvements arrive continuously as underlying systems evolve.

Yet the economics of the agent economy are more complicated than they first appear.

Every rented agent requires translation. Organisational processes must be interpreted. Data must be connected. New products must be explained. Policy changes must be reflected. Local knowledge must be continuously maintained.

This work does not happen once. It happens continuously.

As organisations deploy larger numbers of agents, they may discover that they are not primarily paying for intelligence. They are paying for contextualisation.

The challenge becomes particularly visible when agents are replaced, upgraded or moved across providers. Context that was carefully built around one system must often be recreated, revalidated and retested for another.

The economics of the agent economy therefore extend well beyond licence fees and subscriptions. They include the ongoing cost of ensuring that intelligence understands the organisation it serves.

Over time, this may become one of the largest categories of AI expenditure for firms operating in the middle archetype.

The Human-Centred Organisation

The third archetype follows a different economic logic.

Here, AI functions primarily as an accelerator rather than a workforce. Agents summarise information, support analysis and increase productivity, while humans remain responsible for the majority of consequential decisions and interactions.

The economics are more familiar. Labour remains the dominant cost structure. Knowledge remains deeply embedded in people. Organisational capability develops through experience, experimentation and collaboration.

This approach may appear less efficient than highly agentic alternatives when viewed through a narrow operational lens. But efficiency has never been the only source of economic value.

Throughout history, many transformative innovations emerged from environments that prioritised exploration, unusual combinations of knowledge and the freedom to pursue ideas before their value could be measured.

Human-centred organisations may ultimately compete through adaptability, originality and discovery rather than through pure operational leverage.

Their economic advantage lies in preserving optionality.

When Intelligence Becomes Metered

The most underappreciated aspect of AI economics today may be tokenisation. Cloud computing charged organisations for storage and computation. AI increasingly charges organisations for cognition.

For most of modern economic history, intelligence behaved largely as a fixed cost. Organisations hired capable people, trained them and absorbed the cost through salaries, management structures and incentives. Once employed, thinking itself was not separately measured.

AI changes that relationship.

Every prompt, workflow, recommendation and agent interaction carries a measurable cost. The economics become particularly interesting when intelligence begins operating continuously across the enterprise.

A customer-service agent interacts with customers. A sales agent accesses product information. A compliance agent checks policy. An operations agent updates workflows. Agents exchange information with databases, systems and increasingly with one another.

Thousands of small interactions begin accumulating throughout the day, each consuming tokens.

Leaders will increasingly need to ask which activities justify this level of sophisticated and costly reasoning, where simpler intelligence is sufficient and whether the value created justifies the cost of intelligence consumed.

The Cost Nobody Is Measuring

A significant portion of modern organisational life is devoted to moving information from one place to another, reconciling competing interpretations and determining who should make which decision.

The systems described throughout this series, shared context, behavioural assessment, interoperability standards and agent-to-agent coordination, systematically change those economics.

If intelligence becomes abundant, analysis itself becomes easier to access. The economic challenge shifts toward alignment.

Recommendations can be generated rapidly. Options can be evaluated continuously. Simulations can be run at scale.

The value emerges when those insights become coordinated action.

The organisations that benefit most from AI may therefore be those that redesign information and intelligence coordination itself, reducing the organisational friction required to turn knowledge into action.

The Strange Economics of Intelligence Production

Every infrastructure transition creates concentration. Railroads concentrated transportation. Electric grids concentrated power generation. Cloud computing concentrated infrastructure.

AI appears likely to create its own centres of gravity because training frontier models requires extraordinary concentrations of capital, data, talent and computing infrastructure.

Yet the economics of intelligence production differ from those of most previous infrastructure businesses.

Railroads could remain productive for decades. Power infrastructure generated value over long periods. Even cloud infrastructure often remained economically useful for years.

Frontier intelligence operates on a compressed cycle. Models require continuous retraining.

As a result, providers incur substantial costs long before intelligence is consumed.

Training investments must be made regardless of how many tokens are ultimately used. Revenue arrives later through usage.

This creates a different economic structure from traditional software and even from cloud computing.

The market may ultimately reward providers that continuously reduce the cost of intelligence creation just as much as those that improve intelligence itself.

For organisations consuming AI, this matters because the economics of providers eventually shape the economics of customers.

The cost of intelligence is unlikely to be determined solely by technical capability. It will also be determined by how efficiently intelligence can be produced, maintained and delivered at scale.

The Economics Beneath the Future of Work

Across this series, we have explored how AI reorganises work, knowledge, motivation, hiring, assessment and standardisation.

For two centuries, organisations have largely been built around the economics of labour.

The AI era introduces the economics of intelligence.

The winners of the AI era may be those that understand where intelligence should be owned, where it should be rented, how much contextualisation is required to make it useful, and whether the value being created exceeds the intelligence being consumed.

Because the future of work may ultimately depend less on access to intelligence and more on how organisations redesign themselves around it.

(This essay is part of Founding Fuel’s five-part series on the Future of Work and Agentic AI, which explores how AI is reshaping organisations, decision-making, hiring, motivation, interoperability and the economics of intelligence. Explore the full collection here.)