Market Forces

Before asking what AI and crypto do to the economy, ask what kind of force each is. The kind determines which questions are worth asking.

Frame First: What Kind of Forces?

Six forces. Each starts somewhere on the endogenous–exogenous line. Where it sits decides whether policy and design can shape it — or only react to it.

AI capability progress

Starts exogenous (arrives from outside any single firm). Becomes endogenous once economies invest in R&D, data, and human capital to accelerate it. If endogenous, the growth rate is designable — not just absorbable.

AI adoption

Endogenous. Shaped by regulation, organisational inertia, trust, and infrastructure investment. The gap between frontier capability and economic impact is almost entirely endogenous.

Blockchain and tokenization infrastructure

Exogenous to individual industries (new rails arrive). Endogenous to financial-system design choices. Who controls the rails determines who captures the rents.

Regulatory frameworks

Exogenous shocks to businesses. Endogenous to political economy. Compliance deadlines are starting guns, not obstacles.

Agent feedback loops

Deeply endogenous. Agents create value that funds more agent capability. Self-reinforcing — the core risk and the core opportunity.

The IMF (2026) frames this explicitly: treat AI as a macro-critical transition, not a standard technology shock. That reframe changes every downstream question.

The Questions Worth Asking

Layer 1 — Growth: endogenous or exogenous?

The deepest question. Does this compound from within the system, or arrive and settle?

• Is AI automating AI research? If yes, the loop more ideas → more AI → more ideas mirrors the Romer endogenous-growth mechanism. Potentially explosive — not linear.
• Does tokenization change the growth rate — or only the distribution? $255T of marketable securities exist; only $28.6T is actively used as collateral. Unlocking that gap could be growth-generative — or could redistribute existing rents to new intermediaries.
• What is the compounding rate of cheaper AI? Inference costs for frontier models dropped over 99% in recent years. At what point does cost become irrelevant and adoption become universal?

Layer 2 — Labor: substitution or complementarity?

Discourse gets this wrong by asking too broadly. The sharper version:

• Which task layer is being automated — high-expertise or low-expertise? Opposite outcomes. Automating high-expertise tasks (e.g. legal analysis) lowers barriers and can increase employment at lower wages. Automating low-expertise tasks raises expertise requirements and reduces employment in that occupation while raising wages for those who remain.
• Where is the intelligence/physical boundary in your industry? AI substitutes for intelligence-intensive labor and reallocates workers to physical activities. What share of value in your market is intelligence-based vs presence-based?
• Will agent negotiation compress buyer margins, or expand them? Assistant agents reduce search friction and switching costs — shifting power from suppliers to buyers. Which side are you on?
• Does AI create a new expertise bottleneck? Expertise (not formal education) accounts for roughly one-third of wage variation. If AI commoditises codified knowledge, what tacit know-how stays irreplaceable in your domain?

Layer 3 — Power and concentration: who captures the rents?

• Endogenous risk amplifier or diversifier? Endogenous risk — created by participant interactions inside the system — is behind most financial crises. Multi-agent systems with aligned incentives can create dangerous feedback loops. What prevents your agent architecture from becoming a systemic-risk amplifier?
• Winner-take-most dynamics? Economies of scale in frontier models concentrate market power. Is the market you're building in one where network effects + compute advantages produce a single winner, or one where protocol-level openness preserves competition?
• Does tokenization democratize or recentralize? Governance claims about democratization obscure a contradiction: adapting tokenization to prevailing financial infrastructure may undermine the promise. Who controls the ledger? Who sets the rules?
• What moat survives agent commoditization? As geographic moats give way to protocol moats, which protocol standards will you help define — or be locked into?

Layer 4 — Adoption: why does the gap exist?

38% of companies pilot AI. 11% reach production. The gap is not technical.

• What is the binding constraint in your target market? Candidates: regulatory uncertainty, compliance burden, organisational inertia, trust failures, managerial risk aversion, inadequate infrastructure. Which dominates yours?
• Lagging or leading jurisdiction? Compliance deadlines (EU AI Act, MiCA) hit August 2026. Jurisdictions with regulatory clarity attract capital. Are you building for compliance-ready markets first?
• Do you reduce the organisational friction cost of adopting agents? If the blocker is org inertia, an architecture that lets teams adopt agents incrementally — without re-architecting — solves the actual problem.
• When does the experience-revelation effect trigger mass adoption? Many AI innovations are only appreciated after using them. What is the minimum viable experience that creates irreversible adoption?

Layer 5 — Sovereignty: data footprint and flow

This is where the framing "data footprint and data flow determine capability and sovereignty" is most predictively powerful.

• Who owns the training data in your vertical? In winner-take-most markets, proprietary datasets are the primary moat. Who controls ground-truth data in trusted commerce, DePIN, and health coordination?
• Does on-chain data create a new class of verifiable economic identity? Blockchain-native proof-of-work, reputation, and transaction history may replace credential systems. Who defines the standard for on-chain trust signals in your market?
• Open web of agents or walled gardens? The architecture of agentic communication — open protocols vs siloed APIs — decides whether generative AI democratizes economic access or concentrates it.
• What happens to economic sovereignty when agents negotiate for you? If agents optimise for your stated preferences, who validates they represent your interests vs the platform's?

Layer 6 — The questions discourse avoids

Four that current commentary tends to understate:

1. Is AI an endogenous growth engine or an endogenous risk amplifier — and can it be both at once? Most discourse treats these as separate scenarios. The more dangerous reality is that they co-evolve: the same feedback loops that drive explosive productivity growth drive explosive systemic risk.
2. Does tokenization shift the locus of endogenous risk from banks to protocols? Traditional endogenous financial risk concentrates in leveraged institutions. Tokenization spreads risk across more actors — but may make it harder to detect and contain because feedback loops run at machine speed.
3. Where is the small-economy leverage point? Advanced economies capture disproportionate AI gains via digital infrastructure, AI-ready labor, and institutions. Small advanced economies with strong institutions and narrow specialisations face a specific question: where is the leverage point — and what would you tokenize or agentify first?
4. Does trusted execution become more or less valuable as agents proliferate? If agents verify and transact autonomously, trust infrastructure could become the new bottleneck — or become commoditized. The answer depends on whether verification stays endogenous to human judgment or goes fully machine-executable.

Prediction-Worthy Questions

Each can run through a prediction checklist. Numeric priors as a starting point — calibrate against your own evidence.

Question	Domain	Suggested initial p
AI agent task horizon exceeds 1 full workday by end 2027	AI capability	0.70
Tokenized RWA market exceeds $100B by end 2027	DePIN / Finance	0.55
Major jurisdictions adopt AI regulatory framework by end 2027	Regulation	0.35
Agentic walled gardens dominate over open protocols by 2028	Market structure	0.45
AI drives measurable wage bifurcation (expertise premium +20%) in OECD by 2028	Labor	0.60
Trusted execution environments become a distinct compliance requirement by 2027	Governance	0.50
A major financial crisis with endogenous AI-amplified dynamics occurs before 2030	Systemic risk	0.30

The Meta-Question

Are AI and crypto making economic forces more endogenous — system-generated, self-reinforcing, designable — or more exogenous: faster, less predictable, outside any actor's control?

If more endogenous: the strategic priority is designing the feedback loops deliberately — choosing which compounding cycles to be inside.

If more exogenous: the strategic priority is resilience, optionality, and positioning to absorb shocks faster than competitors.

The honest answer: both, at different layers, at different speeds. The skill is knowing which layer you are operating in at any given moment.

Context

• Economy — the spine
• Autonomous Agents — most readers start here
• Intelligent Hyperlinks — the three pipes (information, value, intent) tokenization and agents jointly run
• Countries — which societies have the culture and infrastructure to thrive in a programmable economy
• IntentTrace — receipt schema that closes the agent–token loop
• Phygital Mycelium — how the two forces mature: hexagonal centralisation evolving onchain into decentralised mycelium

Closing Questions

• When the agent acts but no receipt proves it, who is accountable for the outcome?
• If tokenization encodes alignment and agents encode execution, which breaks first when the system fails?
• What does "coordination cost" mean for your team — and is it falling?