Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchStripe and Tempo release an open standard letting AI agents pay for services programmatically — with Visa contributing credit and debit specs. The protocol supports fiat, crypto stablecoins, and microtransactions on mainnet today.
AI, startups & growth insights. No spam.
TL;DR: On March 18, 2026, Stripe and Tempo — a Paradigm-backed blockchain startup — jointly released the Machine Payments Protocol (MPP), an open standard that lets AI agents pay for services programmatically without human approval at each step. Visa contributed the credit and debit card specifications, extending MPP to its global network. Tempo's mainnet went live the same day after 3.5 months of testing. Early adopters include Browserbase, PostalForm, and Prospect Butcher Co. The protocol supports fiat, stablecoins, and microtransactions, and integrates with Stripe's existing PaymentIntents API in a few lines of code. This is the infrastructure layer the agentic economy has been missing.
The Machine Payments Protocol is an open standard — not a product, not a closed network — that defines how an AI agent requests a service, receives a payment demand, authorizes spending, and receives the resource in return. The entire exchange happens programmatically, without a human approving each transaction.
That framing matters because it distinguishes MPP from anything that has come before it. Today's payment systems were designed for humans. When a person buys something online, the flow is obvious: you navigate to a pricing page, choose a plan, enter card details, complete a CAPTCHA, and receive access. Every step assumes a human with a browser, eyes, and judgment.
AI agents do not have any of those things. An agent executing a multi-step task — researching competitors, summarizing news, generating a report, mailing a document — needs to interact with dozens of services. Each one has its own account creation flow, its own payment process, its own authentication requirements. Without a unified payment primitive, the agent either stalls at every paywall or requires a human to handle payments manually, which defeats the purpose of automation.
MPP solves this by giving services a standard way to signal that they require payment, and giving agents a standard way to authorize and complete that payment — across fiat currencies, stablecoins, and eventually any rail the ecosystem extends to.
The specification is deliberately minimal. As Tempo cofounder and Paradigm managing partner Matt Huang described it: "Our team just came up with what we thought was the most elegant, minimal, efficient protocol that anyone can extend without our permission." That open architecture is intentional — MPP is designed to be a foundation that anyone builds on, not a walled garden that Stripe or Tempo controls.
Tempo is the blockchain layer underneath MPP's stablecoin payment path. It is not a consumer product and not a general-purpose chain — it is optimized specifically for fast, low-cost payment settlement, designed for the volume and granularity that machine-to-machine payments require.
Tempo was incubated as a joint project between Stripe and Paradigm, the crypto investment firm. It raised $500 million at a $5 billion valuation in 2025, with backing from Paradigm, Thrive Capital (Joshua Kushner), and Grenoaks. The company spent the subsequent months building and testing its mainnet — which went live on March 18, 2026, after 3.5 months in test operation.
The significance of the mainnet launch is that MPP is not theoretical. Stablecoin payments through the protocol are live today, on production infrastructure, with real early adopters already processing transactions.
It is important to note that while MPP currently runs on Tempo, the protocol is designed to be rail-agnostic. The specification does not require Tempo — it can be extended to other blockchains, to traditional payment rails, and to any future infrastructure the ecosystem decides to support. Visa's card-based extension, described below, is the clearest demonstration of that architecture: it adds an entirely different payment rail without changing the underlying protocol.
The technical flow of MPP is straightforward, which is part of its power.
When an agent needs a resource — a web scrape, a document print job, an API call, a headless browser session — it sends a request to the service's endpoint. The service responds with a payment request: a structured message specifying the amount, currency, and accepted payment methods. The agent, working within a pre-authorized spending cap, completes the payment. The service delivers the resource.
For Stripe merchants, this maps directly onto existing infrastructure. The integration uses the standard PaymentIntents API:
const paymentIntent = await stripe.paymentIntents.create({
amount: 1,
currency: 'usd',
payment_method_types: ['crypto'],
confirm: true
});That is the entire payment leg for a one-cent stablecoin transaction. The agent's host application creates the payment intent, confirms it, and the resource is delivered. The merchant's standard Stripe infrastructure handles the rest: tax calculation, fraud protection, reporting, and payouts on the normal schedule in the merchant's default currency.
One of the more elegant architectural decisions in MPP is the concept of "sessions" — described in the design documents as OAuth for money. Rather than authorizing individual transactions one by one, an agent authorizes a spending cap once per session. Within that session, it can stream micropayments continuously as it consumes services: every API call, every browser second, every document page. The human who deployed the agent set the cap; the agent operates autonomously within it.
This session model solves a real problem. If every micropayment required a separate authorization, the latency and overhead would make the economics of microtransactions impractical. By front-loading authorization into a session cap, MPP enables the sub-cent, high-frequency payment patterns that agent workloads actually generate.
Parag Agrawal, founder of Parallel Web Systems — a per-API-call web access service that is one of MPP's early adopters — described the integration: "Parallel is built for a world where agents are the primary users. We integrated machine payments with Stripe in just a few lines of code."
Visa's involvement in MPP is the detail that elevates this from a crypto infrastructure story to a mainstream financial infrastructure story. Visa did not simply endorse MPP — it contributed the formal specification for how agents pay using credit and debit cards, and it made the protocol available on the Visa Acceptance Platform.
Specifically, Visa released three components as part of its MPP contribution:
Card-based MPP specification. A formal technical spec that allows merchants, payment service providers, and acquirers to participate in card payment flows within MPP. This means a service accepting MPP payments can receive traditional Visa card payments alongside stablecoin payments — using the same protocol, the same integration, the same request-response flow.
Software Development Kit (SDK). A developer SDK that implements the card spec, so any merchant integrating MPP gets card support without writing custom card integration code.
Security integration. MPP card transactions get access to Visa Intelligent Commerce and the Trusted Agent Protocol — Visa's trust and fraud infrastructure applied to machine payment flows.
Cuy Sheffield, Visa's head of crypto, framed Visa's motivation clearly: "We look at MPP as another way that you can have a very clear, defined protocol around how an agent communicates with merchants."
Rubail Birwadker, Visa's global head of growth products and strategic partnerships, added: "By extending Visa's network, we're bringing trust and resilience into these new forms of commerce."
What Visa's participation does, at a structural level, is remove the bifurcation between "crypto payments" and "card payments" from the MPP ecosystem. A service integrating MPP does not have to choose whether it wants crypto users or card users — it gets both, through a unified protocol. That significantly lowers the friction for adoption among businesses that have existing card infrastructure and do not want to add a parallel crypto payment stack.
It also provides the institutional legitimacy that a new protocol needs to be taken seriously by large financial institutions, enterprise procurement teams, and regulators. When Visa contributes specifications to a new payment standard, compliance officers at Fortune 500 companies read that differently than when a crypto startup releases an open-source spec on GitHub.
Four businesses were named as early adopters in Stripe's launch announcement, and they collectively illustrate the breadth of the use cases MPP enables.
Browserbase offers headless browser sessions — cloud-hosted browser instances that agents use to navigate the web, fill forms, extract data, and complete web-based tasks. The problem MPP solves for Browserbase is pricing: browser sessions are consumed in seconds, and charging for them per-session rather than per-month requires a payment mechanism that works at that granularity. An agent can now spin up a Browserbase session, pay for exactly the time it uses, and close it — without a monthly subscription or per-user account.
PostalForm lets agents pay to print documents and mail them. This is a particularly concrete illustration of the MPP value proposition: an AI agent completing a workflow that requires a physical document to be mailed can now close that loop autonomously, paying PostalForm to handle printing and delivery without a human needing to do anything beyond the initial task delegation.
Prospect Butcher Co. is a New York sandwich shop accepting agent-placed orders for pickup or delivery. The use case is deliberately grounded: a lunch-booking agent can autonomously order and pay for food. This is not a headline use case for enterprise software teams, but it demonstrates that MPP works across the full spectrum of commerce — from API calls to physical goods.
Parallel Web Systems provides per-API-call web access, charging agents for each call rather than through a subscription or data plan. Parallel's founder described the integration as "a few lines of code" — and their model of per-call pricing is only commercially viable because MPP makes sub-cent transactions practical at scale.
Together, these four cases cover SaaS compute (Browserbase), physical fulfillment (PostalForm, Prospect Butcher), and data API access (Parallel Web Systems). They demonstrate that MPP is not a solution looking for a problem — it is a protocol designed around actual commerce patterns that agent workloads generate.
The economic model MPP enables is fundamentally different from how software services are priced today.
Current pricing models — monthly subscriptions, annual enterprise contracts, per-seat licensing — were designed around human users with predictable, bounded consumption patterns. A team of 20 people uses a SaaS tool roughly 20-person-equivalents per month. That predictability makes subscription pricing rational.
Agents do not have consumption patterns like that. A single agent running a complex research task might make thousands of API calls in minutes. Another agent might be dormant for days and then burst through a month's equivalent of human API usage in an hour. Subscription pricing penalizes low-volume use and fails to capture value from high-volume bursts. Per-seat licensing does not apply at all — agents do not have seats.
MPP enables a different economic model: pay-as-you-go at the granularity of the actual resource consumed. A browser second. A document page. An API response. A search result. This is the pricing model that makes agentic workloads economically legible for service providers — and it is the pricing model that makes agentic services commercially viable to build.
The session-based authorization model also creates a new category of financial product: agent spending budgets. A company deploying agents sets per-agent, per-session, or per-task spending caps. Those caps become the primary cost control mechanism for agentic operations, the same way API rate limits are the primary consumption control mechanism today. Finance teams will eventually manage agent budgets as a line item, the same way they manage cloud compute and SaaS subscriptions.
This is not speculative. The businesses currently integrating MPP are already building their pricing models around it. Browserbase and Parallel Web Systems are not offering agents a different tier of their existing subscription plans — they are building agent-native pricing from the ground up, with MPP as the payment primitive.
The obvious concern with a protocol that lets software spend money autonomously is: what prevents an agent from spending more than intended, or being manipulated into spending on malicious services?
MPP addresses this through a layered governance model.
Pre-authorization caps. The session model requires that a human (or higher-level orchestration system) authorize a spending cap before an agent can make any payments. An agent operating within a session cannot exceed that cap regardless of what services it encounters. The human approval is front-loaded, not eliminated.
Stripe's existing fraud infrastructure. For Stripe merchants, MPP transactions run through the same fraud detection, tax calculation, and risk scoring that protect existing Stripe payment flows. Agents do not get a separate, less-scrutinized payment path — they use the same infrastructure that Stripe has built to protect human transactions, applied to machine-initiated ones.
Visa's Trusted Agent Protocol. Visa's security integration for card-based MPP transactions includes the Trusted Agent Protocol, a framework specifically designed to verify the trustworthiness of the agent initiating a card transaction. This addresses a specific attack vector: a malicious service that responds to agent requests with inflated payment demands or attempts to capture card credentials outside the expected flow.
Audit trails. Because MPP transactions flow through Stripe's existing infrastructure, merchants get full transaction records, reporting, and accounting integration. The agent's spending is visible and auditable — not a black box.
What MPP does not yet have, and what will need to develop as the ecosystem matures, is a standardized framework for agent identity verification. When an agent initiates a transaction, the service receiving the payment can verify the authorization and the amount, but the question of whether the agent is who it claims to be — deployed by the organization claiming to deploy it, operating within its intended scope — is still an open problem. This is not a flaw in MPP specifically; it is a broader unsolved problem in agent identity infrastructure that several teams are working on in parallel.
MPP is not the only protocol competing for the role of default agent payment standard.
Coinbase's x402 is the most direct competitor. x402 is an HTTP-based payment protocol that uses the 402 Payment Required status code — dormant since the early web — as the trigger for agent payment flows. It is crypto-native and designed for the same use case: enabling agents to pay for HTTP resources programmatically. Coinbase released x402 earlier in 2026 and has its own developer ecosystem and adoption momentum.
Google's payments scheme, released in September 2025, supports both credit cards and stablecoins for agentic commerce. Google's approach leverages its existing payments infrastructure and its relationships with merchants through Google Pay, giving it distribution advantages that a protocol-only approach lacks.
The competition between these standards is real, and the outcome matters because the winning protocol will become embedded in how services price and sell to agents — making switching costs high once ecosystem lock-in develops.
MPP's advantages in this competition are substantial. Stripe's existing relationships with millions of merchants mean that MPP adoption can start with an upgrade to an existing Stripe integration rather than a new payment provider relationship. Visa's involvement gives MPP the card payment coverage that crypto-only protocols lack. And the open architecture — Matt Huang's explicit design goal of making something "anyone can extend without our permission" — gives MPP a credibility advantage over protocols tied to specific companies' product strategies.
The most likely outcome is not winner-take-all: different protocols will find different niches, and the best-designed services will eventually support multiple standards through adapters. But MPP's launch positions it strongly as the default for Stripe-integrated businesses — which is a very large portion of the internet's commercial infrastructure.
Every major technology transition has an infrastructure moment — the point where the enabling layer for the next wave of applications becomes commercially available and broadly accessible.
For the web, that moment was when SSL/TLS made secure transactions practical and browsers standardized around it. For mobile commerce, it was when Apple Pay and Google Pay made one-tap checkout possible. For cloud computing, it was when AWS released S3 and EC2 and made it cheaper to rent infrastructure than to buy it.
For agentic AI, the infrastructure layer has been emerging in pieces. Vector databases for memory. Tool-calling APIs for action. Orchestration frameworks for multi-agent coordination. Each piece removed a specific limitation on what agents could do autonomously.
Payments was the missing piece. Without a standardized way to transact, agents hit paywalls they could not cross, relied on pre-provisioned accounts that required human setup, or needed human approval for every purchase — effectively forcing humans back into the loop for the most commercially consequential actions.
MPP removes that limitation. An agent can now, for the first time, be deployed with a spending budget and a set of tasks, and complete those tasks — including paying for every service, resource, and tool the tasks require — without a human touching the process.
That is not an incremental improvement. It is a qualitative change in what autonomous agents can accomplish. The agent economy — where AI systems act as buyers and sellers of services, where software pays other software — becomes commercially viable at the moment when payments are as programmable as everything else.
The businesses that understand this earliest will have the most time to build agent-native services and pricing models before the transition becomes the new default. Browserbase, PostalForm, and Parallel Web Systems are early. Most businesses are not. The gap between those two groups will close faster than most industry observers currently expect.
The practical implications of MPP fall into three categories for businesses thinking about the next 12 to 24 months.
If you sell services to developers or other businesses, the question is whether your pricing model can serve agent customers. Subscription plans and per-seat licensing are not agent-compatible. You need to audit whether your services could be priced per-unit — per API call, per minute, per document, per query — and whether integrating MPP would open a new category of agent-customer that your current pricing model cannot capture.
If you are deploying agents for internal operations, the question is what those agents currently cannot do because of payment friction, and whether MPP integration in your agents' tool stacks would change that. Every workflow that currently requires a human to handle a payment step is a candidate for automation. Procurement workflows, subscription management, API access, vendor payments — all of these have payment steps that MPP can make agent-native.
If you are building AI products, the question is whether your product's value increases if the agents it deploys can spend money autonomously. Task completion rates go up when agents can pay their way past paywalls. Workflow automation becomes more complete when agents can procure the resources they need. The businesses building on top of agent infrastructure — orchestration platforms, copilot tools, vertical AI applications — should be evaluating MPP integration as a near-term roadmap item.
Integration itself is straightforward for Stripe merchants. The PaymentIntents API already supports the payment_method_types: ['crypto'] configuration. Supporting MPP is an extension of an existing integration, not a new infrastructure build. The barrier is not technical — it is commercial awareness that the use case exists and is live today.
MPP is an open standard that defines how AI agents request services, receive payment demands, authorize spending, and receive resources — all programmatically, without human approval at each step. It was co-authored by Stripe and Tempo (a Paradigm-backed blockchain company) and released on March 18, 2026. Visa contributed the credit and debit card specifications.
Giving an agent a credit card requires the agent to navigate account creation flows, pricing pages, and checkout interfaces designed for humans — which agents cannot reliably do. MPP creates a machine-readable payment layer: services respond to agent requests with structured payment demands, and agents complete those payments through a standard API call. No human-facing interface required.
MPP supports stablecoins (via the Tempo blockchain), fiat currency, and traditional credit and debit cards (via Visa's card-based MPP specification). It is designed to be rail-agnostic, meaning additional payment types can be added as the ecosystem develops.
Yes. Tempo's mainnet went live on March 18, 2026, the same day MPP was announced. Early adopters including Browserbase, PostalForm, Prospect Butcher Co., and Parallel Web Systems are already processing transactions through the protocol.
For Stripe merchants, MPP integration extends the existing PaymentIntents API. The protocol requires a few lines of code to add to an existing Stripe integration. Parag Agrawal of Parallel Web Systems described the integration as "just a few lines of code." Full documentation is available at Stripe's blog announcement.
MPP uses a session model — similar to OAuth — where a human or orchestration system pre-authorizes a spending cap before the agent begins operating. The agent cannot exceed that cap within the session. Stripe's fraud infrastructure and Visa's Trusted Agent Protocol add additional security layers for card-based transactions.
Visa contributed the card-based payment specification to extend MPP beyond crypto stablecoins to traditional credit and debit card payments. This makes MPP accessible to businesses with existing card infrastructure and gives agents the ability to pay with either cards or stablecoins through a single unified protocol. Visa's global head of growth products stated the goal is to bring "trust and resilience into these new forms of commerce."
Both address the same problem: enabling agents to pay for HTTP resources programmatically. x402 is crypto-native and leverages the HTTP 402 status code. MPP is rail-agnostic, integrates with Stripe's existing merchant infrastructure, and includes Visa's card payment support — giving it broader payment method coverage and a larger built-in merchant base. The two protocols are likely to coexist in different niches rather than one eliminating the other.
Sources: Stripe Blog, March 18, 2026, Fortune, March 18, 2026, PYMNTS — Visa Coverage, March 18, 2026, PYMNTS — MPP Overview, March 18, 2026, CoinDesk, March 18, 2026
OpenAI releases GPT-5.4 Mini — 2x faster than GPT-5 Mini with near-flagship reasoning — and Nano, a $0.20/M token model built for autonomous agent workflows. Free tier users get the Mini upgrade immediately.
NVIDIA Cloud Partners have cumulatively deployed more than one million GPUs across AI factories globally — doubling from 400,000 GPUs last year to 1.7 gigawatts of AI capacity spanning sovereign deployments across six continents.
Santander completes the first regulated bank transaction in the EU executed entirely by an AI agent — setting a precedent for autonomous financial services under the EU AI Act compliance framework.