Agentic commerce is AI agents discovering, comparing and buying on someone’s behalf, over protocols such as MCP, ACP, UCP and AP2 that have shipped since late 2024. For telecom, it means software buys connectivity: an eSIM attached to a trip, IoT data plans procured by an agent. Providers with machine-readable pricing, purchase APIs and instant eSIM provisioning make the shortlist. Quote-only providers do not.
When the Buyer Is an AI Agent
In May 2026 Google said it would take its Universal Commerce Protocol beyond retail, into hotel booking and food delivery. Connectivity was not on the list. It should be next. No product is easier for a software agent to buy than an eSIM, and outside travel eSIM few industries are less ready to sell to one.
A year ago this argument would have been futurism. Between November 2024 and June 2026 the plumbing for agent purchasing shipped as named, open protocols with card networks attached. An agent can now call a provider’s systems (MCP), run a checkout (ACP, UCP), prove that a human authorized the spend (AP2, Visa’s Trusted Agent Protocol, Mastercard Agent Pay) and pay another machine directly (x402, MPP). Each layer has a published specification and a reference implementation you can run this week.
The 2025 to 2026 Protocol Timeline
Keep the four layers apart. Most confusion about agentic commerce comes from treating a tool protocol, a checkout protocol and a payment mandate as the same thing.
| Layer | Protocol and Owner | Status, Dated |
|---|---|---|
| Agent-to-tool How an agent calls your systems. | MCP , created by Anthropic, governed by the Agentic AI Foundation under the Linux Foundation | Open-sourced November 2024. Donated to the Linux Foundation on 9 December 2025 , at which point it reported 97 million monthly SDK downloads and more than 10,000 public servers. Supported by ChatGPT, Claude, Gemini, Microsoft Copilot and Cursor. |
| Commerce and checkout How an agent completes a purchase. | ACP from OpenAI and Stripe; UCP from Google with Shopify, Etsy, Wayfair, Target and Walmart | ACP: Instant Checkout launched in ChatGPT on 29 September 2025, opened to all US users in February 2026, and was scaled back to discovery-and-redirect in March 2026 . The protocol continues as an open standard adopted by PayPal, Shopify and Adyen. UCP: announced at NRF on 11 January 2026 with more than 20 endorsers including Visa, Mastercard, Stripe and Adyen; cart, catalog and identity linking added in March ; Universal Cart and expansion to Canada, Australia and the UK announced 20 May. Integrates over MCP and A2A, and is compatible with AP2. |
| Authorization How an agent proves it was allowed to spend. | AP2 from Google with 60-plus partners; Visa Trusted Agent Protocol; Mastercard Agent Pay | AP2 announced 16 September 2025: signed Intent, Cart and Payment mandates carried as verifiable credentials, with a “human not present” mode for delegated purchases. In May 2026 the FIDO Alliance formed working groups around AP2 and Mastercard’s Verifiable Intent, with Visa and Mastercard chairing the payments group. On 10 June 2026 Visa and OpenAI announced tokenized agent payments , and Visa launched Agent Score , a rating of how well agents can navigate and transact on a merchant’s site. |
| Machine payments How one machine pays another. | x402 from Coinbase, governed by the x402 Foundation with Cloudflare; MPP from Stripe and Tempo | x402 launched May 2025, v2 in December 2025, with Stripe support from February 2026. MPP launched 18 March 2026 with more than 100 integrated services. The Tempo ecosystem lists Anthropic, OpenAI, Shopify, Mastercard and Revolut. Volumes are small: x402 cleared roughly $28,000 a day in early March 2026 . |
Two things this timeline forces me to say. The rails are volatile: OpenAI launched, expanded and retreated from in-chat checkout inside six months, and machine-payment volume is still counted in tens of thousands of dollars a day. So build for interoperability: publish assets any agent can read and call, and let the rails fight it out. And every fact above carries a date because a six-month-old claim in this space can be wrong. This page is reviewed quarterly.
What Agentic Commerce Means for Connectivity
Agent-ready commerce: Agent-ready commerce means a product an AI agent can discover, compare, buy and receive without a human in the loop: public machine-readable pricing, an API or MCP tool for ordering, a payment path that accepts an agent acting under a signed mandate, and instant digital fulfilment.
Why Connectivity Is Unusually Agent-Buyable
Most of what agents are being taught to buy is physical. Shoes need sizes, returns and a delivery address. A data plan needs none of that. It is fully digital, it fulfils in seconds through eSIM, and it compares on attributes an agent can parse: coverage, price per gigabyte, allowance, validity, activation time. A travel agent can attach an eSIM to a booking the way it attaches a seat. A procurement agent can compare per-SIM rates across five providers before a human has opened the first PDF.
The fulfilment layer arrived on schedule. GSMA Intelligence puts eSIM smartphone penetration at 5 percent at the end of 2025, 10 percent by the end of 2026, and 4.9 billion connections by 2030, after Apple took its eSIM-only iPhone global in late 2025. On the IoT side, the GSMA’s SGP.32 specification lets a device switch operator profiles remotely, with no one touching the hardware. The product is already digital. The open question is which providers will be legible to agents when the agents arrive.
Three Scenarios: Travel eSIM, IoT Procurement, Plan Management
Travel eSIM (near term). A traveller tells an assistant to get her to Tokyo on the 14th and sort out data. The agent books the flight, then queries connectivity providers for Japan. It reads each catalog over MCP or from a pricing file, filters on coverage, allowance and activation time, and picks a 10 GB, 15-day plan at the best price. The purchase clears through a checkout protocol under a mandate the traveller signed once. The eSIM profile is on her phone before the plane lands.
Every component exists today. Travel eSIM is already sold through APIs, eSIM delivery is instant, and the checkout and mandate layers shipped this year. Only the assembly is new. My prediction, stated so it can be checked: travel eSIM will be the first telecom product an agent buys end-to-end for a consumer, and the purchase will happen inside a travel product, because the trip is the job and the SIM is a line item.
IoT procurement (emerging). An operations team is deploying 5,000 asset trackers across three countries. Its procurement agent pulls per-SIM rates, coverage, contract terms and data-pooling rules from every provider that publishes them, shortlists, and places a standard order through the winner’s API. The comparison half works today wherever prices are public. The execution half waits on mandates a finance team will accept for a 5,000-line order, which is what AP2’s delegated mode is built for. The condition for making the shortlist at all: a price the agent can read.
Plan management (the retention flip). An agent inside a subscriber’s phone, or an enterprise’s operations stack, watches usage and right-sizes plans under policy: it downgrades idle lines, tops up busy ones and moves a fleet to a cheaper operator where the current one overcharges. Every switch costs nothing, which removes the inertia most connectivity businesses live on. Sheerbit puts churn on competitive consumer plans near 5 percent against about 1.5 percent for enterprise lines; agents push every segment toward the consumer number unless you are the best plan on the attributes the agent scores. Being the best plan starts to matter more than owning the customer’s habit.
What Makes a Provider Agent-Ready
Agent-ready telecom provider: A telecom provider is agent-ready when an AI agent can find its plans, read prices and terms, place an order, pay under a mandate and receive a working eSIM, with no human in the loop and no form to fill.
You can test your own readiness this afternoon. Ask an assistant with web access to buy a 10 GB Japan eSIM and report which providers it considered, which prices it could read and where it got stuck. If your brand is missing from its list, the checklist below explains why. Visa now sells this diagnosis as a product: its Agent Score, announced in June 2026, rates how well agents can navigate and transact on a merchant’s site. Telecom sites will be scored the same way.
The Agent-Readiness Checklist
Seven items, ordered by effort. The first two take days.
Agent-readiness checklist
Many sites block AI crawlers wholesale and are absent from agent answers as a result. Decide which agents may read and which may act, allow the ones you want in robots.txt, and log what they do.
#Standard SKUs purchasable end-to-end without a human: create order, pay, receive the eSIM activation code, confirm activation. Instant eSIM fulfilment is the delivery layer; a three-day SIM in the post fails the agent’s task.
#Expose catalog lookup, ordering and account actions as tools an agent can call. MCP is the layer the ecosystem standardized on, UCP and AP2 both integrate through it, and agents built on ChatGPT, Claude, Gemini and Copilot can use it today.
#Accept agent-initiated payments through your payment provider’s agentic rails: ACP and UCP compatibility, AP2-style mandate verification, Visa and Mastercard agent tokens. Your PSP does most of this work; your job is to turn it on and stop betting on a single rail.
#Design how a verified human or business stands behind an agent’s purchase. The mandate carries the principal; your activation flow must bind the SIM to that principal and re-verify when the profile moves. Telecom regulation requires it in most markets, and mandates make it possible.
#Spend limits, velocity checks and anomaly detection enforced by infrastructure, per principal and per payment instrument. A policy written inside a prompt is advice to the agent. A policy enforced by your platform is a control.
#
Tick what you already have. Your readiness score appears here.
Everything on this list also serves human buyers. Public pricing converts better than a quote form, instant eSIM cuts support tickets, and spend controls prevent bill shock. If agents take three years longer to arrive than I expect, none of this work is wasted.
Implications for MVNOs, OEMs and Fintechs
For MVNOs and resellers, discovery shifts from brand to attributes. The agent knows your coverage map, your price per gigabyte and your activation time, and nothing about your logo. Standard SKUs get compared and commoditized faster, and the margin moves to bundles, controls and service quality that an agent cannot score from a table. Acquisition budgets that bought awareness start buying comparability.
For device OEMs and platforms, connectivity becomes a line item an agent attaches at order time: a fleet platform’s agent adds SIMs to a purchase order, a consumer device’s setup agent picks a plan during onboarding. The OEM whose connectivity is a callable tool gets attached; the one behind a portal login gets skipped. Our guide to IoT connectivity platforms covers the architecture this needs.
Fintechs and super apps already hold the two things an agent purchase needs: a verified customer and a payment instrument. Klarna, Nubank and Revolut have launched mobile offerings; an agent that buys connectivity for a verified account holder is the next step, and the fintech’s agent may become the buyer of record for its customers’ data plans.
Machine payments change pricing, too. With an MPP session or an x402 wallet, an agent can pay for 500 MB for an afternoon and stop, which makes per-megabyte or per-day connectivity an API call rather than a plan. Providers that already meter in real time are a configuration change away from that model.
Will agents replace telecom sales? For discovery and standard SKUs, yes, and soon. For enterprise deals, no. But the enterprise buyer’s agent will have run the comparison before the first call, so the sales conversation starts at the shortlist.
Risks: Trust, Fraud, KYC and Control
KYC is the hard problem, and it varies by market. In the US, prepaid SIMs need no registration; in India, Germany, Spain and most of Asia and Africa, activation attaches to a verified identity. So the agent can transact, but a verified principal must stand behind the mandate, which is what AP2-style signed mandates and the card networks’ agent tokens are built to carry: who authorized what, within which limits. The provider’s job is the binding from mandate to KYC record, including re-verification when an eSIM profile is transferred.
Fraud runs at machine speed. Anyone who can buy SIMs through an API can buy ten thousand of them, and SIM-farm and OTP fraud were automated long before agentic checkout. The controls are mandate verification, velocity limits per principal and per payment instrument, and anomaly detection on activation patterns, enforced in your platform rather than requested in a prompt.
Disputes need machine-readable rules. An agent buys the wrong plan; the mandate chain shows intent, cart and payment, which settles who authorized it and says nothing about whether you owe a refund. Publish change and refund terms in the same structured form as prices, or the agent cannot reason about them and the dispute lands on your support desk.
Protocol churn is the last risk. Instant Checkout’s retreat is the warning. Put your own API and MCP layer between your catalog and any commerce rail, so a rail can change without your product changing.
The Agentic MVNO: Spenza’s Four-Layer Architecture
Agentic MVNO: An Agentic MVNO is a mobile service built on four layers: multi-operator connectivity sourcing, usage and spend controls, a branded experience and billing layer, and AI agents that operate the first three. Its products can be bought by agents and its operations can be run by them.
The framework above is industry-neutral and the checklist works for any provider. This section is ours. Spenza’s platform, which lets a company run a mobile service on other operators’ networks, has been three hubs for some time; the fourth layer is what changes the category.
TelecomHub, ControlHub, UXHub, and the Agents That Run Them
TelecomHub is the supply agents shop from: multi-operator connectivity across more than 180 countries, with plans and eSIM inventory structured so an agent can compare them. ControlHub is the guardrail layer: usage analytics, spend limits and anomaly detection, where checklist item seven lives. UXHub is the interface: branded storefronts, billing and self-service, serving human and machine customers from the same catalog. The agents are the fourth layer: support, provisioning, plan changes and monitoring across the other three, with the same functions callable by an external agent.
The proof points are on record. Spenza’s partner platform is exposed over MCP, so an agent can read the plan catalog, provision an eSIM and manage a subscriber through the protocol layer the rest of the ecosystem adopted; few connectivity providers can say that today. Phone Service for AI Agents sells telecom primitives to agents as a product line: numbers, voice, SMS and real-time audio streaming in more than 40 countries. RocketPhone built an AI-powered business mobile network on the platform, with enterprise eSIM, native voice and Salesforce integration. And the positioning that predates all of this, API-first, eSIM-native, launch in under seven days, matched the checklist before the protocols existed.
That gives Spenza both sides of a frame no one else occupies: agents as buyers of connectivity, which is this article, and agents as users of telecom, which is why AI agents need real phone numbers. One buys the SIM. The other answers calls on it.
Conclusion
Agentic commerce is set to change how connectivity is discovered, compared, purchased, and managed. As AI agents move from answering questions to taking actions, telecom providers will increasingly compete not only for human customers, but also for the machines making purchasing decisions on their behalf.
For MVNOs, OEMs, fintechs, and connectivity providers, becoming agent-ready is therefore becoming a product requirement. Machine-readable pricing, open APIs, MCP tools, protocol-aware payments, verified identity, strong fraud controls, and instant eSIM provisioning will determine whether a connectivity product can be discovered and purchased by an AI agent. The underlying connectivity infrastructure is already moving in this direction, with GSMA eSIM standards supporting remote provisioning and management across consumer and IoT use cases.
The biggest opportunity is eSIM connectivity because it is inherently digital, globally deployable, and easy to evaluate through structured attributes such as coverage, data allowance, price, validity, and activation time. This makes travel eSIM, IoT connectivity procurement, automated plan management, and machine-to-machine payments natural early use cases for AI agents in telecom.
The competitive advantage will not come from simply adding an AI chatbot to a telecom website. It will come from making the entire connectivity product callable by software. An agent should be able to discover a plan, understand its terms, compare alternatives, complete an authorized purchase, satisfy KYC requirements, provision an eSIM, and manage the subscription without relying on a human-operated portal.
The protocols will continue to evolve, so providers should avoid building around a single commerce standard. A durable agentic telecom architecture should separate the connectivity catalog, APIs, identity, payment, provisioning, and agent interfaces so that protocols can change without forcing the underlying product to change.
For Spenza and the broader MVNO ecosystem, this creates a new category: the Agentic MVNO. Connectivity becomes infrastructure that both humans and AI agents can buy, operate, optimize, and manage. The providers that make their networks, pricing, provisioning, and controls machine-readable today will be best positioned for the next generation of telecom distribution.
The future of buying connectivity will not begin when AI agents arrive. It is already being built. The question for telecom providers is whether their products will be visible, comparable, purchasable, and manageable when the buyer is no longer a person, but an AI agent acting on that person’s behalf.
FAQs
If you sell connectivity, agents will reach your catalog whether or not you prepared it for them. Start with the checklist; the first two items take days. If you would rather start on a platform that is already callable over MCP, talk to Spenza.
