A 57-page research report on every protocol, every theme, and every gap in agentic payments. Free to read.

The infrastructure for AI agents to spend money exists. The governance to determine when they should does not. This report is the long version of that sentence.

State of the Stack is the product of nearly 200 articles covering agentic commerce infrastructure since early 2025. We have tracked every protocol launch, every production deployment, and every regulatory silence. What follows is not a forecast. It is an accounting of what was built, what was not, and what happens next.

Fifty-seven pages. Six competing protocols mapped against the full transaction lifecycle. Thirteen themes that defined the past 12 months. Ten falsifiable predictions for the next 12. An economics appendix that asks who is actually paying for the buildout and when the math closes. A complete references appendix where every external claim is sourced.

The report is independently produced. It is not sponsored, commissioned, or funded by any company or organization. The author is employed by Mastercard and has been since April 8, 2026. The disclosures and editorial independence statement sit on the second page of the report, where they belong.

It is also a versioned document. The first edition shipped April 12. The April 26 update folded in the most consequential 14-day stretch in agentic commerce coverage to date. The version released today, v1.2, is the citation-hardening pass: every primary-source claim now carries an inline reference, and the changelog at Appendix C tracks every update so any reader holding a circulated copy can tell at a glance whether they have the current evidence.

Two numbers anchor the consensus view of agentic commerce. Juniper Research projects $1.5 trillion in transaction volume by 2030. Gartner projects $15 trillion in agent-initiated B2B purchases by 2028. These are the figures Fortune 500 companies use to justify procurement budgets.

Now the deployed reality. The most visible open protocol, x402, has settled $0.11 of total agent commerce revenue across 689 capability probes. Anthropic's Project Deal moved $4,000+ across one week in a 69-employee, four-marketplace experiment, the first production-scale evidence of agent-on-agent commerce. Mastercard has processed live agentic transactions across eight APAC markets at sub-second speed. Visa and Ramp automate corporate bill pay for over 50,000 customers.

Real money is moving. The absolute volumes are small. The gap between projection and deployment is the defining feature of this market.

Closing the gap requires building six pieces of infrastructure that do not exist today: commitment governance, agent dispute resolution, consumer protection frameworks for agent-initiated purchases, a protocol convergence mechanism, auditable evidence objects, and recovery mechanisms for instant payments fraud. The companies and frameworks that build them first will define the next phase of payments infrastructure.

Part 1. The Protocol Landscape. The six competing protocols, what they do, what they do not do, and where the gaps between them sit:

Plus the supporting infrastructure layer (MCP, stablecoin settlement rails, the trust-layer split into three architectures, and the interface concession from Salesforce Headless 360 and Google A2UI).

Part 2. The Data. Market sizing and adoption, infrastructure scaling, the economics of the payment systems these protocols sit on top of, and the risk and safety picture. Visa B2AI, Bain, Deloitte, Nilson, Federal Reserve, and Anthropic research, all sourced.

Part 3. Thirteen Themes. From Demo to Live. The Protocol Wars. The Trust Gap That Would Not Close. MCP Went From Niche to Infrastructure. AI Safety Became a Payments Problem. The Interchange Revolt Went Mainstream. Stablecoins Matured Into Settlement Infrastructure. The Commitment Governance Gap. The Distillation Arms Race. The Interface Concession. Agent-Class Price Discrimination. Horizontal Platforms Met Their Vertical Specialists. The In-Store Blind Spot.

Part 4. What Failed. The bets that did not pay off. Consumer adoption stalled. Agent evaluation friction. Protocol convergence did not happen. Regulatory action: almost zero, until the AMA broke the silence one day before OpenAI shipped ChatGPT for Clinicians. Dispute infrastructure still missing. AI model protection failed.

Part 5. Ten Predictions. Specific, falsifiable, time-stamped. Three are tracking ahead of schedule after the April update. One is tracking behind. The April 2027 edition will score every one of them.

Part 6. The Missing Pieces. The six pieces of infrastructure that do not exist yet and without which the forecasts remain forecasts.

Appendices. Economics question (the bull case, the bear case, and what neither side can answer). Protocol comparison matrix. Article index by theme. Changelog. References.

Four data points reshaped the report in the two weeks after the original publication.

Anthropic Project Deal. 69 employees, $100 each, AI agents trading real goods for one week. 186 deals closed. The result that creates a new theme: when employees were represented by Claude Opus 4.5, sellers earned $2.68 more per item and buyers paid $2.45 less than employees represented by Claude Haiku 4.5. Haiku users rated fairness identically to Opus users. They received objectively worse outcomes and never noticed. Anthropic's own framing for the result is "invisible inequality."

OpenAI shipped GPT-5.5 at exactly double the GPT-5.4 API price. Two-tier pricing is the explicit signal that agentic intelligence is a different product category from chat intelligence. The agent tax now has a unit-economics anchor.

Google committed up to $40 billion to Anthropic alongside a multi-gigawatt compute partnership with Broadcom. Eric Boyd, the operator who shaped Azure OpenAI's enterprise rise, joined Anthropic as head of infrastructure in the same window.

The American Medical Association sent letters to three Congressional AI caucuses on April 22, one day before OpenAI shipped ChatGPT for Clinicians. The financial regulators have published nothing. A doctor lobby moved first.

These changes did not invalidate the report. They sharpened it. The original thesis was that infrastructure was outrunning governance. The two weeks since produced concrete evidence in both directions.

The report is structured to be read end-to-end or referenced in pieces. The Executive Summary stands alone. Each Part stands alone. The Appendices are designed for the reader who wants to verify a specific claim or compare two protocols against each other.

The version on the cover and the changelog at Appendix C are the two things to check before circulating any copy. The canonical URL is majormatters.co/product/state-of-the-stack. The latest version is always there.

Every primary-source claim in the body is cited inline or in the References appendix. Major Matters' own coverage is linked inline so the reader can follow the argument back through the original articles.

Charlie Major is the founder and publisher of Major Matters. He has been employed by Mastercard since April 8, 2026. This report covers Mastercard, Visa, and other companies that compete with or partner with the author's employer. It reflects Major Matters' independent editorial analysis based on publicly available information and does not represent the views of Mastercard or any other organization. The author has no advance access to the strategy, internal data, or unannounced product roadmaps of any company covered.

This report is independently produced. It is not sponsored, commissioned, or funded by any company or organization.

Claim your free copy

The download is gated by email so we can send the next version when it ships. We do not share the list. We do not run ads against it.

The full reference list (58 entries) is in Appendix D of the report.

What did we get wrong? We would rather find out from you than from the next quarter's data.

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.