The Landscape Right Now
The agentic payments space just went from theoretical to real — and it happened fast. In the span of about six months, four distinct protocols emerged to solve the same fundamental problem: AI agents can’t pay for things on the internet.
Here’s who’s doing what:
MPP (Machine Payments Protocol) — co-authored by Stripe and Tempo, launched March 18, 2026 alongside Tempo’s mainnet. It’s the session-based, hybrid-rail protocol. Agents pre-authorize a spending limit, then stream micropayments within a session — stablecoins, cards, BNPL, Lightning — all through one interface. Over 100 services integrated at launch, including OpenAI, Anthropic, Google Gemini, Dune, fal.ai, and Browserbase. Tempo raised $500M at a $5B valuation from Thrive Capital, Greenoaks, Sequoia, and Ribbit Capital. Design partners include Visa, Mastercard, Shopify, Nubank, Revolut, DoorDash, Standard Chartered, UBS, Deutsche Bank, and Klarna.
x402 — Coinbase’s open-source answer. Lightweight, crypto-native, permissionless. Embeds stablecoin payments directly into HTTP request-response cycles. No accounts, no intermediaries, just a wallet and a 402 status code. Integrated into Google’s AP2 framework. Partners include Cloudflare, Google, Visa, World (Sam Altman’s identity project), and Anthropic. Zero protocol fees — you only pay gas on L2s like Base.
ACP (Agentic Commerce Protocol) — Stripe’s checkout layer for shopping agents. Launched in ChatGPT’s Instant Checkout in February 2026. Handles cart management, product discovery, and structured checkout flows. Stripe, Shopify, Salesforce, and PayPal support it.
AP2 (Agent Payments Protocol) — Google’s authorization and trust layer. Uses cryptographically-signed “Mandates” to prove a user authorized an agent to spend. Coalition of 60+ organizations including Adyen, Amex, Mastercard, PayPal, Revolut, and Salesforce. Has an x402 extension for crypto settlement.
Visa TAP (Trusted Agent Protocol) — extends existing card network trust to AI agents. Leverages Visa’s interchange infrastructure but carries card-network economics (1.5-3.5% fees), making micropayments uneconomical.
The critical insight: these protocols are not in a winner-take-all race — they’re building a layered stack. AP2 handles authorization, ACP handles checkout, x402 handles permissionless settlement, and MPP handles session-based streaming payments. Stripe explicitly supports both MPP and x402 through separate integration paths, owning the abstraction layer while letting protocols compete underneath.
Where This Is Going
Three shifts are converging that make this moment genuinely different from every previous “micropayments will fix the internet” pitch:
1. AI agents are real buyers now, not hypotheticals. The agentic AI market was valued at roughly $7-10 billion in 2025 and is projected to hit $100-200 billion by 2033-2034 at a ~40-47% CAGR. More concretely, 63% of enterprise executives believe agentic commerce will be crucial within two years. BCG projects AI agent-led shopping could represent over a quarter of e-commerce spending in the near future. McKinsey pegs the US B2C agentic commerce opportunity at up to $900 billion by 2030. These numbers are obviously squishy, but the directional signal is unmistakable: agents need to spend money, and they currently can’t do it programmatically.
2. Stablecoins became real financial infrastructure. Stablecoin monthly transaction volumes are approaching $700+ billion. Stripe acquired Bridge (stablecoin infra) for $1.1B. Tempo’s mainnet is purpose-built for stablecoin settlement with sub-second finality. This matters because stablecoins solve the micropayment economics that killed every previous attempt — you can actually settle a $0.001 transaction profitably on an L2.
3. The MCP + payments stack is crystallizing. Model Context Protocol (MCP) gave agents a standard way to discover and use tools. MPP (and x402) give them a standard way to pay for those tools. This is the “last mile” — once you can discover, use, AND pay for any API in a single request cycle, the internet becomes a vending machine for AI agents.
My thesis: Within 18 months, every API that charges for access will offer at least one agentic payment protocol. MPP will likely win the “high-frequency, enterprise, hybrid-rail” segment because of Stripe’s distribution and Visa/Mastercard’s participation. x402 will dominate the “long-tail indie developer” segment because it’s permissionless and zero-fee. The authorization layer (AP2 or similar) will become mandatory for any agent spending >$100/day as compliance requirements catch up.
The real money, though, isn’t in the protocols themselves — it’s in the tooling, infrastructure, and services that sit on top of them.
The Whitespace Map
1. Agent Wallet Infrastructure & Management
The gap: Every agent needs a wallet. Right now, setting up wallets for agents is manual, insecure, and fragmented across chains. There’s no “Stripe for agent wallets” that handles key management, spending policies, multi-chain support, and compliance out of the box.
Who feels the pain: Developers building agent systems who need to manage wallets for hundreds or thousands of autonomous agents, each with different spending limits and permissions.
Why incumbents aren’t solving it: Coinbase has partial coverage via x402, Crossmint is building here (agent wallets + virtual cards), but the market is early enough that no one owns the developer experience yet. Stripe is focused on the protocol layer, not wallet management for agent fleets.
Size: If even 10% of the projected $10B+ agentic AI market needs wallet infrastructure, that’s a $1B+ TAM within 3 years.
2. Agent Payment Analytics & Observability
The gap: When agents are making thousands of micropayments per day across multiple protocols (MPP, x402, card rails), there’s no unified dashboard to track spending, detect anomalies, enforce budgets, and generate financial reports. Stripe’s dashboard shows MPP transactions, but doesn’t cover x402, doesn’t aggregate across protocols, and doesn’t offer agent-specific analytics.
Who feels the pain: CTOs and finance teams at companies deploying agent fleets. They have no visibility into what their agents are spending, where, and whether it’s generating ROI.
Why incumbents aren’t solving it: Too early. Stripe sees itself as the payment processor, not the observability layer. DataDog and similar monitoring tools haven’t built payment-specific agent monitoring.
Size: Adjacent to the $20B+ observability market. Agentic payment observability could be a $500M niche within 3-4 years.
3. MPP/x402 Middleware for Legacy APIs
The gap: There are millions of existing paid APIs (weather data, financial data, geolocation, etc.) that charge via API keys and monthly subscriptions. None of them support MPP or x402. Someone needs to build the proxy layer that wraps any existing API in an MPP-compatible interface.
Who feels the pain: API providers who want agent revenue but don’t want to rebuild their billing systems. Agent developers who want to use APIs that don’t speak MPP.
Why incumbents aren’t solving it: MPP’s docs mention “Proxy an existing service” as a guide, but there’s no hosted solution. Cloudflare could build this but hasn’t prioritized it beyond basic x402 support.
Size: The API economy is worth $5B+ and growing. Capturing even the “conversion layer” fee would be substantial.
4. Agent-to-Agent Payment Orchestration
The gap: Current protocols handle client-to-service payments. But the emerging pattern is agents hiring other agents — an orchestrator agent that needs to pay a research agent, which pays a data agent, which pays a compute agent. Multi-hop payment flows with splitting, escrow, and settlement aren’t well-supported by any protocol yet.
Who feels the pain: Anyone building multi-agent systems (CrewAI, AutoGen, LangGraph) where agents need to compensate each other for work done.
Why incumbents aren’t solving it: MPP supports split payments (it’s in their docs), but the tooling for complex multi-hop flows is minimal. The multi-agent orchestration frameworks haven’t integrated payments at all yet.
Size: Multi-agent systems captured 66% of the agentic AI market share in 2024. As these systems need to transact, the orchestration layer could be a $500M-$1B opportunity.
5. Compliance & Authorization for Regulated Industries
The gap: Healthcare, financial services, and government sectors want agent automation but face strict compliance requirements around who authorized what payment and why. AP2’s mandate system is a start, but there’s no turnkey compliance layer for regulated agent deployments.
Who feels the pain: Enterprise compliance officers who are told “we want to deploy agents” but have no framework for audit trails, spending governance, and regulatory reporting.
Why incumbents aren’t solving it: Google’s AP2 addresses authorization but not industry-specific compliance. The protocols are built for developers, not compliance teams.
Size: Financial services alone is projected to contribute 20% of the $632B AI spending growth through 2028. Compliance tooling for agent payments in regulated industries could be a $1B+ opportunity.
Signals to Watch
Stripe’s Agentic Commerce Suite adoption metrics — Stripe said URBN, Etsy, Coach, and Ashley Furniture are onboarding. Watch for volume numbers at Stripe Sessions or earnings calls. If transaction volume on MPP is in the millions/month by Q4 2026, the market is real.
x402 v3 and multi-chain expansion — x402 currently focuses on USDC on Base. Watch for expansion to Solana, Arbitrum, and Ethereum mainnet. More chains = more developer reach = faster adoption.
IETF standardization progress — MPP was submitted to IETF recently. If it gets traction as an official internet standard, it becomes the default. Watch for working group formation and draft iterations.
Google AP2 production deployments — AP2 has a massive coalition but limited production usage. Watch for actual transaction volumes, not just partner announcements.
Agent framework payment integrations — When LangChain, CrewAI, or AutoGen ships native MPP/x402 support as a first-class feature, that’s the inflection point for developer adoption.
Consumer trust numbers — Only 16-29% of consumers currently trust AI to make payments. Watch surveys from Visa, Mastercard, and consumer fintech companies. If trust crosses 50%, the floodgates open for B2C agent commerce.
Regulatory moves — The EU’s AI Act and potential US legislation around autonomous agent transactions could either accelerate (by providing clarity) or slow (by adding friction) adoption. Watch for agent-specific payment regulations.
Contrarian Take
MPP’s biggest risk isn’t technical — it’s that the “agent economy” stays an API-to-API affair and never reaches consumer commerce.
Everyone is projecting trillions in agentic commerce by 2030. But here’s the uncomfortable truth: Stripe’s own co-founders called it “overhyped too early” and expect 95% of AI-driven sales in 2026 will still complete on merchant websites, not inside the AI. Only 11% of enterprise AI pilots make it to production. And only 16% of US consumers trust AI to make payments.
The part of the agent economy that’s real right now is narrow: AI agents paying for API calls (LLM inference, data feeds, compute). That’s a meaningful market — probably $5-10B within 3 years — but it’s not the $3-5 trillion consumer commerce revolution everyone is pitching.
MPP is well-positioned for the boring-but-real market (API micropayments), but its valuation and ecosystem reflect the hype-driven projections of the bigger vision. If consumer trust in agent payments stays below 30% through 2028, the market for protocols like MPP will be a fraction of current projections — still large, still important, but not world-reshaping.
The smart play for builders: ignore the trillion-dollar projections and build for the $5-10B API payments market that exists today. The tooling gaps are real, the customers are identifiable, and the competition is still nascent. Let the big players fight over consumer commerce while you build the picks and shovels for the developer economy.
This analysis reflects publicly available information and the author’s independent assessment. It is not investment advice.