The Central Macro Thesis
Our starting point is not the claim that Bitcoin is inevitably going parabolic. The more consequential thesis is broader: the economic transmission mechanism itself may be changing as artificial intelligence moves from a productivity-enhancing software tool toward an autonomous economic actor. That distinction matters because conventional macro analysis is built around humans making decisions, firms employing humans, banks intermediating capital, and money moving through institutions at human speed. An economy increasingly populated by software agents does not necessarily obey those same constraints.
The evidence available today supports part of that thesis, but not all of its more aggressive conclusions. U.S. labor productivity increased 2.2% year over year in the second quarter of 2026, while hours worked increased only 0.2%. Over the same period, output increased 2.5%. The current business cycle has produced annualized productivity growth of roughly 2.1%, compared with 1.5% during the 2007–2019 cycle. Those figures are consistent with an economy becoming more productive, although they do not establish that AI agents are the sole or even dominant cause. r1
At the same time, the distributional signal is unusually important. BLS reports that labor's share of nonfarm business output fell to 52.8% in the second quarter of 2026, the lowest level in its series beginning in 1947. That does not prove that AI is replacing workers at scale, but it demonstrates that the capital-versus-labor allocation mechanism is already moving in a direction that deserves serious macroeconomic attention. r2
I therefore interpret the underlying argument as a regime-change thesis: AI may cause incremental economic activity to accrue increasingly to capital, software, intellectual property, compute, data infrastructure, and automated financial networks rather than appearing first as conventional employment growth. If that happens, the familiar relationship between liquidity, employment, consumption, GDP, and asset prices becomes less reliable.
Why Productivity Is the Critical Variable
For decades, the cleanest macroeconomic framework has been that productivity raises the economy's capacity to produce goods and services, while employment and capital determine how that capacity is deployed. AI complicates the framework because software can replicate cognitive work without requiring a proportional increase in headcount. An agent can research, code, negotiate, schedule, monitor, transact, and coordinate continuously. Its economic contribution can therefore grow much faster than the number of humans formally employed to operate it.
This is not merely theoretical. The Bureau of Labor Statistics has explicitly expanded its research program around AI and productivity. BLS notes that AI is implicitly captured in its measurement through software investment and reports research finding that industries with greater AI exposure are strongly positively associated with labor-productivity growth and are becoming more capital intensive. r3
That creates an important analytical asymmetry. Traditional employment statistics may weaken even while output per worker rises. Corporate profits can expand without equivalent hiring. Software expenditure can increase the productive capacity of a business without producing the same visible employment footprint as a conventional factory expansion. The resulting economy can look weaker through a labor-market lens while simultaneously becoming more productive through a capital-and-technology lens.
We should therefore resist the simplistic conclusion that declining labor intensity automatically means economic deterioration. It can instead represent a redistribution of the production function. The crucial question becomes who owns the productive capital, who controls the software, who supplies the compute, and how the resulting economic surplus is distributed.
The Liquidity Argument Needs to Be Reframed
The conventional crypto macro framework frequently emphasizes monetary liquidity. In that framework, expanding money supply and falling real rates increase the amount of capital seeking risk assets, while tightening liquidity suppresses speculative valuations. There is genuine historical evidence behind this framework, but it becomes incomplete if technology itself changes the velocity and utility of capital.
Imagine an economy in which trillions of dollars of assets remain economically underutilized because collateral cannot move efficiently, ownership cannot be fractionalized cheaply, settlement takes days, counterparties require extensive verification, and financial institutions impose minimum transaction sizes. Tokenization potentially changes that architecture. Instead of asking only whether the monetary authority has created more money, we can ask how much existing wealth can be made more liquid, programmable, collateralizable, and continuously transactable.
This distinction is increasingly recognized by mainstream financial institutions. The Bank for International Settlements said in June 2026 that digital innovation could improve competition and efficiency in payments and financial intermediation, while identifying tokenization as a potentially important component of the next-generation financial system. The BIS simultaneously warned that stablecoins have structural limitations and could create macro-financial risks if adoption becomes widespread. r4
That is a much more useful framework than simply declaring that tokenization will unleash a fixed dollar amount of dormant assets. The magnitude of any future liquidity release remains uncertain. But the mechanism is credible: programmable ownership and collateral can reduce settlement friction, improve capital utilization, and allow assets to participate in financial transactions that previously required institutional intermediaries.
AI Agents Create a New Class of Economic Participant
The genuinely disruptive step occurs when AI agents become capable of acting economically rather than merely advising humans. A chatbot that tells me which airline ticket to buy has limited macroeconomic significance. An autonomous agent that searches hundreds of airlines, compares prices, books the ticket, verifies identity, executes payment, records the transaction, and reconciles the expense is materially different.
At that point, the economy contains a new category of transaction initiator. Humans remain the ultimate beneficiaries and principals, but software increasingly performs the economic search and execution function.
This introduces three potentially transformative characteristics: speed, scale, and persistence. A human can compare a handful of prices. An agent can theoretically compare thousands. A human negotiates during waking hours. An agent can operate continuously. A human generally performs transactions sequentially. An agent can orchestrate multiple transactions simultaneously.
The macroeconomic consequence is that transaction velocity can rise without a proportional increase in population. That matters because many economic models implicitly treat human time and attention as scarce resources. Autonomous agents partially relax that constraint.
The result could be a form of economic growth in which the number of transactions, decisions, searches, and optimization cycles increases dramatically while human labor grows much more slowly. That is exactly why traditional employment metrics can become increasingly incomplete indicators of technological economic activity.
The Financial System Becomes Part of the AI Stack
Once agents can transact, they require financial infrastructure. They need identities, permissions, wallets, settlement mechanisms, collateral, payment rails, and potentially privacy-preserving systems. This is where the AI and crypto thesis becomes considerably more interesting than a simple argument about Bitcoin's scarcity.
Stablecoins are particularly relevant because they provide a digital representation of fiat currency that can potentially move programmatically. BIS research published in June 2026 describes stablecoins as the dominant medium of exchange within the crypto ecosystem and examines their broader implications for monetary, banking, and fiscal systems. r5
Tokenization is potentially broader still. Tokenized deposits, securities, collateral, invoices, commodities, and other claims can make financial assets machine-readable and programmable. If agents become economically significant, that programmability becomes increasingly valuable because machines require deterministic settlement rules.
The Federal Reserve has itself acknowledged the convergence of AI, stablecoins, tokenization, and payments as a major technological theme. A 2026 speech by Governor Christopher Waller described the accelerating intersection of those technologies and emphasized the need for financial infrastructure capable of operating at the speed of technological change. r6
That does not mean blockchains automatically win. Centralized databases, bank payment systems, and conventional financial infrastructure will continue to compete aggressively. The investment question is therefore not “Will crypto replace finance?” but rather “Which parts of financial infrastructure benefit when economic agents become software rather than people?”
Why Bitcoin Is Different From the Agent Economy
Bitcoin should be separated analytically from the broader collection of cryptoassets. Bitcoin's strongest macroeconomic proposition is not that it will necessarily process every machine transaction. Its more distinctive characteristics are scarcity, portability, settlement independence, collateral potential, and a monetary network whose supply is not controlled by a conventional issuer.
That makes Bitcoin potentially complementary to an agent economy rather than synonymous with it. Agents may transact primarily in stablecoins or tokenized deposits while institutions continue to use Bitcoin as collateral or a strategic reserve asset. Alternatively, Bitcoin could capture a larger share of machine-to-machine settlement if its infrastructure becomes sufficiently fast and inexpensive. At present, the latter proposition remains a technological and adoption question rather than an established economic fact.
The broader crypto ecosystem has a different risk profile. New protocols can capture enormous value if they become important infrastructure, but they also face competition, technological obsolescence, regulatory changes, liquidity shocks, and network substitution. A rapidly expanding agent economy does not imply that every token connected to it will appreciate. Economic utility and token value are separate propositions.
The Investment Implications of Capital Intensity
The most immediate investment consequence of this regime is that capital intensity matters more. Data centers, semiconductors, networking equipment, power generation, cybersecurity, cloud infrastructure, model-serving systems, and specialized software are all potential beneficiaries of increasing computational demand. The market is already treating portions of this infrastructure as strategic assets rather than ordinary technology expenditure.
But the second-order effects are more important. If AI raises productivity while suppressing the amount of labor required for a given level of output, margins can expand. If margins expand faster than wages, capital income gains a larger share of economic growth. If that capital is then reinvested into additional compute and automation, the process can become self-reinforcing.
That is the positive feedback loop behind the agentic-economy thesis: better models increase agent capability; better agents increase software utilization; greater utilization raises demand for compute and infrastructure; larger infrastructure investment improves capability; and greater capability creates additional economic applications.
There is, however, an equally important negative feedback mechanism. If AI infrastructure becomes overbuilt relative to monetizable demand, capital returns can collapse even while technological capability improves. The dot-com era demonstrated that a transformative technology can be real while many individual investments remain poor investments. Infrastructure can be economically necessary and still become temporarily overpriced.
What the Labor Data Actually Tell Us
We should be precise about what is happening rather than retrofitting every economic statistic to the AI narrative. The latest BLS figures show productivity rising 2.2% year over year in the second quarter of 2026, while hours worked rose only 0.2%. That is meaningful evidence of improved output efficiency. Yet the data do not establish that AI agents caused the increase. Productivity also reflects capital deepening, organizational changes, industry composition, energy, software, cyclical factors, and measurement revisions. r7
Likewise, the decline in labor share deserves attention but should not be interpreted as a one-variable AI statistic. Labor share can move because of changes in compensation, productivity, corporate margins, industry composition, and relative prices. The fact that it reached a historical low in the second quarter of 2026 makes the development important; it does not, by itself, identify the cause. r8
The important analytical conclusion is therefore conditional: if AI-driven capital intensity continues increasing while output grows faster than hours worked, conventional employment-based interpretations of economic strength will become progressively less informative.
The Measurement Problem Could Become a Policy Problem
Economic statistics were not designed for an economy in which software agents perform increasingly autonomous cognitive tasks. GDP measures market production, not the full quantity of useful activity. Productivity measures output relative to inputs, but technological quality improvements can be difficult to capture. Household welfare can rise through lower prices, better services, longer healthy lifespans, or time savings without creating equivalent increases in measured GDP.
The statistical agencies recognize this challenge. BEA announced in 2026 that it was developing experimental statistics intended to measure the size of the American AI economy, including activity associated with data-center construction, algorithm development, energy consumption, and related production. r9
This is strategically important. Measurement determines what policymakers believe is happening. If productivity is underestimated, policymakers may interpret strong technological disinflation as weak demand. If labor displacement is underestimated, social-policy responses may lag. If intangible investment is poorly measured, capital formation may appear weaker than it really is.
The economic dashboard of the next decade may therefore need to track compute utilization, software capital, AI adoption, inference demand, autonomous-agent activity, data-center capacity, tokenized settlement volumes, and machine-mediated transactions alongside traditional employment, inflation, GDP, and money aggregates.
The Key Macro Tension: Productivity Versus Distribution
There is an important distinction between an economy becoming richer and households becoming richer. AI can raise aggregate productivity while distributing the resulting gains unevenly. The BLS labor-share data make that distinction particularly relevant. r10
If productivity gains accrue primarily to owners of capital and intellectual property, aggregate profits can rise while wage growth becomes less central to the growth mechanism. That could change consumption patterns, taxation, fiscal policy, asset allocation, and political economy. It also means that investors cannot treat GDP growth as a sufficient proxy for corporate earnings or household welfare.
In the most extreme version of the thesis, the economy becomes increasingly capital-led: fewer workers generate more output using vastly larger quantities of software and compute. That would not necessarily produce recessionary conditions. It could instead create an unusual combination of strong productivity, weak labor demand, high capital expenditure, concentrated profits, and persistent political pressure over distribution.
The Three Economic Regimes We Should Watch
| Regime | Primary Growth Engine | Liquidity Mechanism | Key Risk |
|---|---|---|---|
| Traditional industrial-digital economy | Labor, physical capital, software | Banks, fiscal policy, monetary policy | Recession and credit contraction |
| AI productivity economy | Compute, software, intellectual property, capital | Corporate investment and financial markets | Overinvestment and unequal distribution |
| Agentic financial economy | Autonomous software and machine-to-machine commerce | Programmable money, tokenization, digital settlement | Cybersecurity, systemic automation, regulatory fragmentation |
Our base analytical framework should not assume that the third regime has already arrived. Instead, we should monitor whether the second regime is becoming measurable and whether the third begins appearing in transaction data. That distinction prevents us from confusing a compelling technological narrative with demonstrated macroeconomic causality.
What Would Validate the Agentic Thesis
I would look for evidence in several areas simultaneously. First, productivity would need to remain elevated without equivalent growth in hours worked. Second, corporate profit margins would need to demonstrate sustained gains associated with software and automation rather than merely cyclical pricing power. Third, inference and agent usage would need to translate into measurable commercial activity rather than remaining primarily experimental. Fourth, machine-initiated payments would need to become material relative to human-initiated transactions. Fifth, tokenized assets would need to develop meaningful secondary-market liquidity and collateral functionality.
The strongest confirmation would not be a cryptocurrency price chart. It would be evidence that autonomous agents are generating measurable economic output and using programmable financial infrastructure to settle transactions at scale.
What Would Falsify or Weaken the Thesis
The thesis also has clear failure conditions. If AI capabilities improve but monetization remains persistently weak, the productivity narrative could fail to translate into corporate earnings. If compute spending grows faster than economic returns, capital expenditure could become a drag on investment performance. If autonomous agents remain unreliable in real-world environments, human oversight could remain too expensive for large-scale delegation.
For crypto specifically, the thesis weakens if agents overwhelmingly use conventional banking rails and centralized payment systems. It weakens further if tokenization produces limited real economic adoption, if transaction costs remain high, or if regulatory and compliance requirements eliminate much of the anticipated automation advantage.
Bitcoin therefore should not be treated as a mechanical beneficiary of AI. The causal chain has multiple links, and every link must work: AI capability must generate economic activity; economic activity must create demand for autonomous transactions; autonomous transactions must create demand for digital financial infrastructure; and that infrastructure must translate into durable value capture.
Our Strategic Interpretation
My interpretation is that the most important development is not a forecast for Bitcoin's next price move. It is the possibility that the economy is transitioning from a human-speed coordination system toward a software-mediated coordination system. That transition would alter how capital circulates, how firms organize labor, how assets function as collateral, and how financial transactions are executed.
The latest productivity data provide a credible early signal that output efficiency is improving. BLS reports 2.2% year-over-year labor-productivity growth in the second quarter of 2026 and a 0.8% increase in total factor productivity during 2025. The agency is simultaneously building dedicated AI measurement frameworks, while BEA is developing experimental estimates of the AI economy. These institutional responses are themselves evidence that the measurement problem has become economically significant. r11
The investment implication is consequently broader than “buy Bitcoin because AI agents are coming.” We should think in layers. The first layer is productive infrastructure: compute, semiconductors, networking, energy, and data centers. The second is software: models, agents, enterprise systems, cybersecurity, and memory. The third is financial infrastructure: stablecoins, tokenization, programmable settlement, custody, collateral, and potentially decentralized networks. The fourth is scarce monetary assets, where Bitcoin occupies a distinct position but competes with other stores of value and forms of collateral.
That framework also explains why the macro debate cannot be reduced to whether the Federal Reserve is tightening or easing. Monetary policy remains crucial, but technological productivity can alter the sensitivity of profits, capital spending, and asset prices to interest rates. A company whose marginal cost of cognitive labor falls dramatically may tolerate financing conditions that would have been restrictive for a labor-intensive business.
The central question for the next several years is therefore not simply whether liquidity is increasing. It is whether the economy is discovering new mechanisms for turning existing capital, software, data, and collateral into productive economic activity at dramatically higher velocity.
If autonomous agents become genuine economic participants, that transformation could be profound. But the responsible macroeconomic conclusion is conditional rather than prophetic: the technology is advancing, the productivity evidence is increasingly visible, tokenization is receiving serious institutional attention, and the financial architecture is adapting. The magnitude of the eventual economic and asset-price consequences remains an empirical question.
Our job as investors is to watch that transition through measurable variables rather than narratives. Productivity, margins, capital intensity, inference demand, employment, labor share, compute utilization, tokenized settlement, stablecoin activity, and machine-mediated transactions will tell us considerably more than any single forecast about where the new regime is actually taking the economy.