The Infrastructure Race for Agentic Commerce: An Analytical Overview
The burgeoning landscape of agentic commerce is marked by the emergence of distinct infrastructural frameworks that have begun to define the operational capabilities of artificial intelligence applications. As organizations strive to harness the potential of AI, a competitive race has materialized, yielding notable winners within the sector.
Key Protocol Developments
Prominent among these advancements is Anthropic’s Model Context Protocol (MCP), which now operates across an expansive network of over 10,000 public servers. The protocol has garnered significant traction with an impressive 97 million monthly SDK downloads, effectively facilitating the integration of AI applications with diverse external tools and data sources.
In parallel, Google’s Agent-to-Agent (A2A) protocol, launched in April 2025, commenced with a robust partnership comprising 50 initial collaborators and has since scaled to include over 100 supporting enterprises. This initiative transitioned under the governance of the Linux Foundation, emphasizing its growing significance within the industry.
Furthermore, on January 11, 2026, Google introduced the Universal Commerce Protocol (UCP), which has attracted strategic endorsements from key industry players such as Shopify, Walmart, Target, Mastercard, Stripe, Visa, and American Express. This protocol aims to establish standardized methodologies for agents navigating real-time checkout processes.
Coinbase’s x402 protocol represents another significant advancement in this domain by providing a payment transport layer that enables automatic stablecoin transactions over HTTP. By late 2025, this framework reported processing over 100 million payments across various APIs and applications integrated with AI agents.
The Standardization Paradox
Despite these advancements in standardization within a technology category that was largely nascent just three years prior, it is critical to note that each of these protocols addresses a specific aspect—namely, the connectivity and coordination of agents in initiating payments. However, they collectively overlook a fundamental commercial challenge positioned one level deeper in the operational stack: determining who adjudicates the completion of work.
| Protocol / Standard | Functionality | Limitations | Relevance to Current Discourse |
|---|---|---|---|
| MCP (Model Context Protocol) | Facilitates connections between AI applications and external tools, APIs, and data sources. | Lacks mechanisms for verifying whether task outcomes have been successfully delivered. | Focuses on the tool/data layer, neglecting the essential trust layer surrounding completed work. |
| A2A (Agent-to-Agent) | Enables communication and coordination among agents across disparate systems or organizations. | Does not incorporate escrow mechanisms or assess deliverable quality. | Solves agent interoperability, yet fails to address conditional settlement requirements. |
| UCP (Universal Commerce Protocol) | Standardizes agent-driven commerce and checkout processes. | Does not ascertain whether a purchased service or task was satisfactorily accomplished. | Paves the way for deeper engagement in real transactions while highlighting the absent verification layer. |
| AP2 (Agent Payment Protocol) | Employs signed payment mandates to confirm authorized expenditure by an agent. | Validates permission but omits verification of whether the desired outcome was achieved. | Serves as an authorization standard rather than a work-verification benchmark. |
| x402 | Facilitates automatic payments via HTTP, encompassing stablecoin transactions. | Affects monetary transfers but does not ensure that funds are disbursed only upon verified completion of work. | Acts as the payment transport mechanism, devoid of escrow or adjudication components. |
| Mastercard Verifiable Intent | Establishes a trust and audit framework for user purchase authorization verification. | Mainly focuses on approved purchases and dispute resolution trails without addressing actual task completion. | Delineates how incumbents are formalizing intent and accountability, while still lacking comprehensive outcome verification. |
| ERC-8183 | Outlines a job-based escrow process: funds are locked, work is submitted, and evaluators confirm or reject submissions before expiry triggers client refunds. | Does not resolve evaluator trust issues or disputes regarding “agentic” identity independently. | This proposal serves as a crucial link to address missing conditional payment / verification mechanisms. |
| ERC-8004 | Presents a trust/reputation framework for agents and their counterparties. | Lacks built-in escrow functionalities or payment-release mechanisms. | This standard is likely to serve as a foundational composition layer, enhancing trust for ERC-8183-like evaluations. |
| Oracle / Staking / zkML / TEE-style Trust Systems | Presents potential methodologies for verifying outcomes or reinforcing evaluator judgments with heightened assurances. | No single standard has emerged yet as universally applicable within agentic commerce frameworks. | This represents potential solutions to the article’s pivotal query: whether entities are empowered to verify job completion? |
The Escrow Mechanism: A Fundamental Component
The ERC-8183 draft Ethereum standard published on February 25 stands as a significant advancement towards making judgment processes programmable within agentic commerce frameworks. Simplifying its design reveals a minimal state machine dedicated to task-based transactions: clients lock budgets into escrow accounts; service providers submit deliverables; and evaluators subsequently determine job completion or rejection. This architecture follows a clear sequence—Open, Funded, Submitted, Terminal—while explicitly assigning evaluation responsibility solely to designated evaluators upon receipt of work outputs.
This design framework is arguably narrower than its characterization as “agentic commerce” might suggest. Critics within the Ethereum Magicians discussion forum have articulated that there is “nothing especially ‘agentic’” about this proposal; one commentator aptly referred to it as “a job registry with escrowed funds.” Such critiques are not only valid but also instrumental in refining our understanding of ERC-8183’s practical implications. Ultimately, what ERC-8183 delineates is a programmable escrow primitive applicable across various task-based transactions—be they human- or machine-driven—which predates the concept of agentic functionalities entirely. The more compelling inquiry lies in whether this structure constitutes the missing component within the current operational stack for agentic commerce.
The Authorization-Verification Dichotomy
The evolving landscape surrounding agentic commerce largely emphasizes authorization rather than verification. For instance, Google’s Agent Payment Protocol frames financial transactions through cryptographically signed mandates that substantiate what an agent is authorized to expend. Similarly, Mastercard’s Verifiable Intent—developed collaboratively with Google—introduces an infrastructure designed to validate user purchase authorizations along with an audit trail aimed at resolving disputes. While these frameworks robustly address inquiries surrounding whether purchases have been sanctioned, they fail to provide insights into whether resultant outcomes have materialized satisfactorily—thus revealing a critical gap within current operational paradigms in agentic commerce.
This dichotomy encapsulates a productive contradiction inherent within the operational stack. The A2A protocol guarantees inter-agent communication across organizational boundaries; MCP ensures access to requisite tools and data; AP2 along with x402 facilitates seamless monetary transactions; while ERC-8183 proposes conditional holding of funds until evaluators affirm that deliverables have met established criteria. The determination of who may serve as an evaluator remains open-ended; it could be a client entity, an oracle network participant, or possibly even derived from proof systems like zkML. Importantly, ERC-8004’s trust and reputation framework are explicitly recommended as foundational layers when addressing higher-value transactional engagements.
The Evaluator Role: A Central Point of Complexity
The role of the evaluator emerges as politically significant within this discourse. The security section of ERC-8183 warns that malicious evaluators possess the capacity to arbitrarily accept or reject submitted work outputs and advocates for implementing reputation or staking mechanisms particularly concerning high-value contracts. The absence of dispute resolution capabilities within the core specification highlights an unresolved complexity inherent in this structure. One contributor within the Ethereum Magicians thread astutely noted that “the Evaluator is where real complexity resides,” while another emphasized that “everyone verifies payment; nobody verifies work.” Such observations underscore a structural dynamic pervasive within any open marketplace involving agents: those who control evaluation inherently wield considerable power over market operations.
The design of ERC-8183 brings this tension into sharper focus. In enterprise contexts where clients also assume evaluator roles, complexities may remain manageable; however, in multi-party networks where providers submit tasks across organizational lines to clients situated elsewhere, evaluators become crucial bottlenecks susceptible to platform-level leverage dynamics. While ERC-8183 articulates this chokepoint clearly within its design principles, it currently lacks durable solutions addressing such complexities.
Status Quo Analysis: The Current Landscape of Agentic Commerce Infrastructure
The adoption metrics indicate that sectors surrounding these protocols are advancing at a pace outstripping developments in verification mechanisms. According to Gartner forecasts, by 2028 it is anticipated that approximately 33% of enterprise software applications will integrate agentic AI functionalities—up from virtually non-existent levels recorded in 2024—while around 15% of routine decision-making processes may be executed autonomously by then.
Deloitte estimates that the global market for agentic AI will reach approximately $8.5 billion by 2026 and could escalate toward $35 billion by 2030—with projections suggesting up to 75% of companies potentially investing in this technology by year-end. Furthermore, IBM and NRF highlighted findings indicating that roughly 45% of consumers already leverage AI during their purchasing journeys—41% specifically utilizing it for product research activities.
This burgeoning volume of agentic activity necessitates robust settlement infrastructures capable of managing transactional complexities effectively. The optimistic perspective surrounding ERC-8183 alongside its associated frameworks posits that open marketplaces encompassing research initiatives spanning code creation through inference capabilities will generate sufficient cross-organizational machine-to-machine commerce activities warranting on-chain conditional settlement solutions.
The Ownership Paradigm: Who Governs Judgment Mechanisms?
If Gartner’s projections regarding agentic AI adoption hold true—granting it substantial influence over enterprise procurement practices along with research outsourcing initiatives—the most lucrative segment within this operational stack will not necessarily be held by model providers themselves but rather by entities capable of governing conditional payment moments—the infrastructure responsible for holding funds while attesting outcomes before releasing payments contingent upon successful completion.
The proposition embodied by ERC–8183 may very well represent such pivotal infrastructure—or alternatively manifest itself as marketplace escrow equipped with superior branding capabilities.
The discourse captured within Ethereum Magicians threads insightfully notes how foundational structures predate contemporary AI paradigms altogether—a sentiment equally applicable across numerous financial primitives ultimately proving consequential over time.
This observation underscores how escrow mechanisms predate digital environments while conditional payments were established long before blockchain technologies emerged.
The ongoing theoretical exploration currently revolves around discerning whether challenges associated with verification processes inherent in agentic commerce are most effectively remedied through conventional authorization standards championed by Big Tech or via programmable on-chain escrow fortified by configurable trust layers.
Bearing witness to both methodologies actively unfolding without definitive resolution suggests forthcoming tensions wherein ultimate answers may hinge upon determining where agents engage predominantly in economically significant activities once adoption reaches thresholds necessitating strategic infrastructural battles.
