The best lever of economic margin for investors to track is power to price. In classical economics, pricing power is not merely a reflection of market share, but rather the capacity of an economic actor to minimize transaction costs while maintaining strategic control over data, risk, and user experience. Historically, eCommerce has operated under a macroeconomic paradigm where merchants absorb the operational and financial frictions of the conversion funnel, while payment networks and processors leverage their scale to price security, identity, VAS and settlement infrastructure.

The emergence of agentic commerce is disrupting this equilibrium. The introduction of open, interoperable protocols ( x402 standard and Machine Payments Protocol (MPP)) is shifting the strategic landscape at both ends of the funnel.  At the top end, agentic platforms will be forced to pay to access Universal Commerce Protocol (UCP) and Model Context Protocol (MCP) data feeds. At the back end of the funnel, autonomous agents will need to pay to register themselves and complete purchases under new risk-mitigation (ie Auth) frameworks. This report analyzes these macroeconomic and technical dynamics, examining how the power to price is being structurally redefined across the commerce value chain.

Pricing Power in Commerce

In my past blogs discussing Transaction Cost Economics (TCE) I outlined why this framework succeeded in explaining why coordination of economic activity through the market price mechanism entails significant friction (namely, the costs of searching, negotiating, and enforcing contracts). In digital commerce, transaction costs are highly asymmetric.

The power to price in commerce today sits firmly with the merchant, but it comes at a high premium. Merchants pay for virtually every operational friction point across the funnel: top-of-funnel customer acquisition through advertising, mid-funnel conversion optimization, back-end payment processing fees, and the substantial costs associated with fraud mitigation and data analytics. 

As I outlined in checkout evolution ( economic context of the last mile redesign), the interface between the processor and the merchant has historically been a primary battlefield for margin capture. Within payments, the networks (Visa, Mastercard, American Express) have demonstrated an unparalleled ability to price transaction delivery through interchange, assessment fees, and Value-Added Services (VAS), However, large merchants have proven adept at constructing competitive advantages within this structure, Tier-1 retailers leverage closed-loop debit networks, bi-lateral agreements (some driven by co-brand) and payment optimization to achieve massive 100-150bps cost advantage, which is significant in a retail industry where average net margin is also around 150bps. 

Agentic Platforms Failure to Disintermediate

The early, model-first vision of agentic commerce anticipated a rapid disintermediation of the commerce funnel (see Death of Instant Checkout). AI platforms wanted to create instant checkout flows intended to collapse the transaction entirely within their conversational interfaces. The hypothesis assumed that merchants would willingly pay a 2% to 5% commission on “purchase orders” generated by these platforms, rendering the retailer’s traditional eCommerce user interface and data capture obsolete.

This vision met a swift demise because it ignored the “operational void”, the complex reality of transaction liability, risk management, and the commercial unit economics that govern physical commerce. Google listened to retailers and created UCP with an embedded checkout flow keeping merchants in control of the final phase of agentic purchases (which included the cost of payments and VAS used).

The Commercial Reality: Data is Not Free

The early era of web search relied on free, open web crawling to power search indexing. In the agentic era, however, this “free lunch” assumption has collapsed. For an agent to provide accurate conversational recommendations, it must access real-time, high-fidelity merchant data regarding stock keeping units (SKUs), local inventory, and dynamic pricing.

Merchants have no incentive to grant free, unmitigated API access to third-party scraping bots, particularly when interacting with their most loyal customers.  While a retailer may be willing to pay an acquisition fee for a net-new customer, they will aggressively put friction (such as rate limits, paywalls, or strict authentication prompts) on external platforms that attempt to insert themselves between the merchant and a pre-existing, loyal loyalty member.  Consequently, agentic platforms must pay to access UCP/MCP data flows, creating a new economic model for search and discovery. 

How MPP and x402 Price UCP Data Flows

The “original sin” of the internet was the omission of a native, open-standard payment layer within the TCP/IP suite, forcing the web to rely on ad-hoc, human-centric credit card forms and centralized databases. This omission has rendered autonomous machine-to-machine (M2M) commerce functionally impossible without constant human authorization. The structural resolution of this issue is occurring via the convergence of the x402 standard and Stripe’s Machine Payments Protocol (MPP).  Note that MPP and x402 are converging to be the same.   

M2M Value Exchange

UCP, MCP, and x402 are all evolving to enable more structured data flows. Obviously, data inquiry in the agentic era cannot operate as a public utility; it requires a direct commercial construct. Its not just a pricing issue, many wonderful data sets are locked up because of permissioning and control of use (ie bank, healthcare, financial). I see four macroeconomic pricing strategies:

Strategy	Mechanism	Primary Economic Use Case
The Toll (Fixed Fee / Pay Before Inquiry)	Micropayment charged per API call or query. Requires orchestrators to have “skin in the game”.5	Protecting premium databases (healthcare, travel, deep pricing) from free scraping.5
The Bounty (Pay for Success)	Performance fee or revenue-share paid upon transaction completion (the purchase).5	Lower-funnel product recommendations and high-intent checkout conversions.5
Market Pricing	Dynamic fees where specialists delivering sales charge a premium, and scrapers deliver no value and pay.5	Monetizing premium, highly localized inventory data while deterring value-extracting scrapers.5
Bi-lateral / Multi-lateral Pricing	Complex value swap where value flows both ways and is complicated by reciprocal services.5	Strategic data-for-compute or data-for-traffic exchanges between massive platforms (e.g., Walmart and Google).5

Measurement in Agentic Commerce

As the founder and former CEO of Commerce Signals I can tell you that the power to price is completely toothless without the ability to measure (value). Every complex pricing model we have discussed, whether a simple toll, a success-based bounty, dynamic market pricing, or a complex bi-lateral swap, lives or dies by its measurement framework.

Google did not build a multi-billion-dollar empire because they had a cool search interface; they built it because the PageRank algorithm was, at its core, a measurement algorithm that recorded what users clicked on for every keyword. They created the definitive directory of the internet by keeping score. In the agentic era, where a “Super Agent” or orchestrator coordinates dozens of federated specialist models, measurement is the missing link.

As outlined in Federated Models Need Measurement, we face a massive attribution gap. Prior to participation, measurement of a specialist’s contribution is required for two reasons: attribution (determining who gets paid what share of the value) and tuning (refining the specialist’s response based on performance). The purchase (the conversion event) is the ultimate metric of success, but to unlock real market forces, every contributor in the chain needs to understand the exact incremental value they created.

In the advertising age, ad platforms graded their own homework. In the next phase, Sellers (and data contributors) cannot rely on the self-attested metrics of super-agent orchestrators (such as Google’s internal reporting), nor can they rely solely on the subjective reports of merchants. For agentic commerce to scale, measurement and attribution must be handled by a neutral, trusted intermediary. This entity must possess the capability to link the initial Intent Mandate to every single specialist model that contributed, all the way down to the final checkout transaction. (ie VAS opportunity for V/MA)

The Data Tipping Point

Many of the world’s best datasets remain locked up because as each is constriained. For instance, bank transaction data remains heavily siloed due to a lack of explicit consumer permissions and intense competitive sensitivity. Similarly, healthcare data represents high-intent preference and medical behavior signaling, but completely lacks standardized mechanisms for user permission and the strict post-authorization controls required to govern downstream use.

The fundamental macroeconomic challenge with data is that its value is highly non-linear, dependent entirely on context, use, and scarcity.  Consider the data scarcity paradox: your mobile location data might place you physically at a McDonald’s, while your credit card transaction data confirms you purchased a meal there. Individually, these datasets are incomplete, but when intersected, they create exponential commercial value. Yet, controlling the downstream use of these intersected data points is exceptionally difficult. As outlined in Data Tipping Point and Data Leakage, moving sensitive customer information out of its native environment inevitably leads to commoditization, allowing external platforms to capture the merchant’s margin. Managing these constraints and measuring the true value of data is hard enough; doing so within a collaborative, multi-party LLM is even harder. 

The underlying science of pricing the incremental value of specific data used to train a monolithic, trillion-parameter LLM is virtually impossible to quantify. How do you isolate the financial contribution of a single purchase record or medical history once it has been processed into generic neural weights? Because this incremental training value cannot be cleanly quantified, a far more elegant approach is to construct federated specialist LLMs. These specialists can know your specific, domain-specific preferences and act as a secure boundary, strictly controlling the leakage of your private data into broader public models. These specialists can inject high-value, contextual enhancements into the orchestrator’s responses, and their performance can be empirically measured and priced against a standard control set where the specialist did not participate. 

Of course, the platform giants have zero incentive to foster this federated specialist ecosystem; they want the raw data in its entirety.  Google is actively in this data-acquisition mode right now because agentic tokens and credentialless checkout mechanisms break many of their legacy data feeds.

Federated Specialist LLMs and Data Sovereignty

The early, monolithic vision of AI assumed that a single, all-encompassing LLM (a “central brain”) would orchestrate the entire global economy, bypassing existing businesses. The emergence of x402 and the Model Context Protocol (MCP) leads me to the belief that a more decentralized, resilient macroeconomic architecture makes sense: Federated Specialist Models. Not only does this unlock new data sets and capabilities it also enables pricing value and measurement. 

Preserving the Specialist Knowledge Gap

In this specialist world, instead of routing all transactions through a single proprietary platform, hundreds of highly specialized, domain-specific models add value to a transaction.  Consumers could even select which specialists they want and permission their data. In this model a consumer will coordinate with a trusted travel planning agent, a HIPAA-compliant healthcare model, or a highly secure bank financial advisor. 

This federated model seems to rely on three key pillars:

1. Model Context Protocol (MCP) Boundaries: By standardizing data exchange via MCP, domain specialists can expose specific functional outcomes to the orchestrating LLM without exposing their underlying training data, model logic, or proprietary data 
2. Cryptographic Authorization: Sensitive datasets (such as bank transaction histories, patient healthcare records, and proprietary merchant supply data) have historically been locked in silos due to competitive dynamics and strict privacy regulations (GLBA, HIPAA, GDPR).  
3. Data Leakage Mitigation: By establishing x402 micropayment handshakes, specialized models can charge a commercial premium to deliver targeted recommendations, ensuring that high-value proprietary insights do not leak into the public domain to train free foundational models.

Strategic Outlook

The practical realization of this is 48+ months away unless a major platform (like APple) enables. This is just another year beyond the 24 to 36 months I project for M2M (large platform w/ no specialists).   Operations and standardized rules must mature and achieve scale within individual, sovereign domains (e.g., inside Google’s ecosystem or within Walmart’s physical and digital footprint) before they can expand securely across distinct commercial boundaries. 

X402 and MPP create the pricing mechanics, but to unlock market forces there must be measurement of value (or another type of commercial construct). My personal hope? I have my own personal data on my phone with an LLM (on my phone) acting as a coordinator that “personalizes” all recommendations.. That no one else knows my real preferences. 

The ability to price machine interactions will ultimately unlock a massive, highly efficient wave of specialist models, allowing “super-specialists” to monetize niche market demands. However, these market forces cannot truly be unlocked until we establish the supporting measurement and governance structures to enforce rules on the exchange of value.