What to Do in the Age of AI: Chasing Scarcity in a Rapidly Deflating World

In the age of AI, money continues to flow toward scarcity and unique value. The difference now is that AI compresses the lifespan of any given scarcity. What was defensible and highly profitable two years ago can become commoditized or drastically cheaper within the next cycle. Opportunities shift every couple of years as capabilities improve and diffusion accelerates.

We are currently living through two distinct phases of this transformation.

The Buildout Era (2023–2024)

What to Do in the Age of AI: Chasing Scarcity in a Rapidly Deflating World The first wave centered on building the foundational infrastructure and tools that let people and companies actually use advanced AI. This included frameworks like LangChain for orchestration, coding tools like Cursor that dramatically boosted developer productivity, massive data center buildouts to power training and inference, and specialized “neolabs” or AI-native research environments.

These were the picks and shovels of the AI gold rush. Demand was high because the underlying models were powerful but raw —developers and organizations needed better interfaces, infrastructure, and tooling to make them usable at scale. The winners here built the rails on which everything else would run.

The Rollout Era (2025–2026)

What to Do in the Age of AI: Chasing Scarcity in a Rapidly Deflating World We have now entered the implementation and integration phase. The focus has shifted to embedding AI deeply into existing organizations, workflows, and markets.

This includes:

AI rollups and vertical applications that acquire or rebuild companies around AI capabilities.
Fractional or embedded AI expertise (sometimes called FDE-style consulting) helping traditional firms adopt these tools.
Large-scale AI transformation projects inside enterprises.
Agentic systems — autonomous or semi-autonomous AI agents — that reimagine and automate entire business processes across industries.

Neither the pure buildout companies nor the rollout players will vanish overnight. Many will continue to thrive by evolving. However, the highest returns are migrating toward newer, more complex, and more uniquely necessary applications. The more an offering solves hard, real-world problems that are difficult to replicate or commoditize quickly, the greater its economic upside.

The Coming Deflation After 2027

What to Do in the Age of AI: Chasing Scarcity in a Rapidly Deflating World A major inflection point looms. By 2027 and beyond, nearly everything that can be produced primarily with tokens — software, digital content, many forms of analysis, design, and even certain types of code or media — will face severe deflationary pressure.

Margins in these areas will compress dramatically as AI capabilities improve and costs plummet. Prices will fall because supply becomes abundant. Scarcity and sustained high value will migrate to domains that cannot be fully produced or replicated with tokens alone.

This doesn’t mean software or content businesses disappear; it means the easy wins and high margins become rarer. The durable opportunities will lie in areas where AI augments rather than fully replaces the scarce resource — real-world experimentation, physical constraints, deep domain expertise combined with AI, or entirely new institutional frameworks.

AI + Science: An Anti-Deflationary Opportunity

What to Do in the Age of AI: Chasing Scarcity in a Rapidly Deflating World Scientific discovery stands out as particularly promising. Unlike pure software or process automation (where making something cheaper can shrink the overall addressable market over time), AI-accelerated science tends to expand the frontier.

Each new hypothesis generated by AI, each simulated or real experiment, and each insight can open new avenues of research and application. There is no automatic deflationary spiral because better tools increase the rate of discovery rather than simply reducing the cost of existing outputs.

Big frontier labs (OpenAI, Anthropic, etc.) excel at scaling models and generating predictions, but they generally lack the infrastructure, regulatory access, physical labs, and rapid real-world feedback loops needed for biology, chemistry, physics, materials science, and other experimental domains. This creates space for agile players — startups, academic spinouts, or hybrid teams—that can combine frontier AI with wet labs, specialized equipment, and domain-specific iteration. The winners here will turn AI into a force multiplier for genuine scientific progress, not just faster computation.

AI + Institutions: The Final Boss

Perhaps the largest and most transformative opportunity lies in reimagining social and economic institutions themselves.

Consider the scale: Optimizing sales, marketing, or customer service departments across companies might generate hundreds of billions in efficiency gains. Rebuilding core societal institutions — healthcare systems, regulatory bodies, education, legal systems, and government administration — could unlock trillions annually in productivity and cost savings. Healthcare alone accounts for roughly 17% of U.S. GDP; government and public administration represent an enormous share of global economic activity. These are among the biggest markets on the planet.

Institutions fundamentally represent trust and coordination mechanisms. Today, that trust is often captured by brands, historical reputation, professional cartels, or political machinery rather than pure expertise or outcomes. Courts rely heavily on precedent and reputation; medicine on licensing cartels; legislatures on fundraising and campaigning skills.

AI now makes it technically feasible to build parallel institutions that deliver 2x (or far greater) effectiveness at a fraction of the cost—perhaps 1/10th or even 1/100th the price of legacy systems. Imagine AI-augmented arbitration platforms that are faster, more transparent, and dramatically cheaper than traditional courts. Or AI-powered educational credentials and assessment systems that are more personalized and outcome-focused. Or regulatory review processes that combine AI analysis with human oversight for speed and rigor.

The challenge is that institutions cannot simply be “disrupted” or torn down like a startup competing with an incumbent app. Legitimacy is sticky and path-dependent. The realistic strategy is parallel construction: build superior, AI-native alternatives that are so clearly better, cheaper, and more trustworthy that migration happens organically. Businesses and individuals will vote with their feet when a new arbitration system resolves disputes 100x faster and cheaper while maintaining (or improving) fairness and transparency.

This is not about replacing government or medicine wholesale. It is about creating credible, high-performance alternatives in specific high-friction areas and letting demand pull adoption.

Where to Focus Your Efforts

In the age of AI, the winning strategy is to position yourself at the intersection of powerful models and enduring scarcity:

Build infrastructure and tools while the buildout phase still has momentum, but plan for evolution into applications.
Drive deep integration and agentic automation in existing industries during the rollout.
Pursue frontier science and experimental domains where AI expands possibilities rather than just cheapens them.
Reimagine institutions by constructing parallel, AI-leveraged systems that solve trust and coordination problems more effectively.

The highest-value work will increasingly involve things that are hard to fully tokenize: physical-world execution, novel scientific discovery, complex human judgment in high-stakes domains, and the redesign of coordination mechanisms that societies rely upon.

AI is not just another technology wave — it is a general-purpose capability that systematically reduces the cost of intelligence and prediction. Those who focus on amplifying what remains scarce, necessary, and uniquely valuable in this new reality will capture the greatest rewards. The game is moving fast, but the principles of value creation — scarcity, uniqueness, and solving hard problems — remain constant. The question is no longer whether to engage with AI, but where you will apply it to create something the world cannot easily replicate or deflate away.

---