When the telecom buyer is a bot

Insights

• AI agents are already consuming telecom network services autonomously, but most operators' commercial systems were built only for human buyers.
• Serving machine customers requires a new commercial layer, one where pricing adjusts dynamically, governance enforces itself and settlement completes automatically.
• Decades of operating under demanding regulatory and compliance frameworks position telcos as the natural trusted party in machine-to-machine commerce.
• Five revenue pools — from identity and fraud services today to an agent-to-agent marketplace at the horizon — give operators a practical path forward.

For most of history, customers have been human. They browse, compare, decide, and pay. A new kind of customer is now arriving: the AI agent, one that does all this automatically on behalf of a business or a person.

This shift is already visible in commerce. Amazon launched "Buy for Me" in April 2025, an AI agent that visits third-party retailer websites and completes purchases without the consumer navigating to the third-party site. Stripe, which provides payment technology to businesses worldwide, has built a dedicated payment infrastructure for agents that allows an AI agent to create a virtual payment card and complete a transaction on a customer's behalf. As Stripe put it, the agent doesn't assist with the purchase; it is the purchaser. Analysts at Gartner have been tracking this shift since 2015, calling it the machine customer era.

The same shift is beginning to reach telecom. AI agents embedded in enterprise workflows are already consuming network services, such as identity and antifraud application programming interfaces (APIs), without a human initiating each individual request. In Brazil, when Itaú, the country's largest private bank, suspects an account takeover, its fraud detection system automatically calls Telefónica's SIM Swap API in real time, checking whether the customer's SIM card was recently changed, and blocking or allowing the transaction in milliseconds. Cabify, the mobility platform with 50 million users, uses Telefónica's Number Verification API to confirm a new user's identity silently during app registration, without the need for any one-time password verification. Orange LiveNet has built a dedicated business unit to bring exactly these kinds of network services to enterprise buyers. Ericsson Research confirmed in April 2026 that the primary user of a network API is increasingly a machine, and therefore it requires that APIs be understood by both people and machines. That is a standard most telecom commercial systems do not yet meet.

This is an early stage of a longer journey. Today, the contracts behind these automated API calls were negotiated by human teams. Tomorrow, agents will need to discover telecom services themselves, assess pricing, verify trust, consume and settle, and complete the process without a human in the loop. The commercial infrastructure to support that does not yet exist at most operators. That is the gap this paper addresses.

The commercial gap

Telecom commercial systems were built for a world where every transaction begins with a human. A person calls a sales team, a procurement officer signs a contract, a developer registers for API access, a finance team receives and pays an invoice. Every step in the commercial chain assumes a person is present, reviewing and approving.

That assumption is breaking down, and when the customer on the other side is a machine, not a human, four gaps stand out: discoverability, accountability and governance, speed and real-time execution, and competition. They must be solved in sequence, because each one depends on the last.

Discoverability

Before an agent can buy a service, it must be able to find and understand it. Today, service catalogs are designed for human readers. They are written in natural language and structured for web pages and sales conversations. An agent cannot parse a PDF brochure or navigate a portal built for procurement teams. But the problem runs deeper than format. As Tom Loosemore, a UK expert in digital services, has observed that companies are already placing deliberate restrictions on what agents can see and access, often for legitimate safety and security reasons.

The result is that agents have a partial view of available services, not the complete picture they need to make a purchasing decision. A human buyer can ask a sales team for what is not on the website. An agent cannot. For machine customers to consume services autonomously, those services must be published in full, in machine-readable formats that an agent can query, interpret, and act on without human translation.

Accountability and governance

Once an agent can find a service, the next question is whether it can be trusted to buy it. With human customers, the answer has always been “Know Your Customer,” which is the regulated process of verifying who someone is before doing business with them. With machine customers, there is no equivalent. Tom Loosemore calls this the “Know Your Agent” problem, and it is one that the industry has not yet solved. Who owns this agent? What authority has been delegated to it? What can it spend, and on whose behalf? Who carries the liability if it acts outside those limits? How are disputes resolved when neither party has a human record of negotiation? Commercial systems assume a person carries legal and financial responsibility for every transaction. When an agent acts autonomously, that entire accountability framework breaks down.

Speed and real-time execution

Even if an agent can discover a service and has the authority to buy it, the current commercial process cannot keep up. With humans, pricing is agreed in advance, entitlement is checked manually, billing runs in cycles, and settlement happens once the invoice is processed. Agents cannot wait for any of this. For instance, a logistics company's AI agent reserving network capacity for a port automation window needs confirmed pricing, authorization, and a service-level commitment before it acts — in milliseconds, not in the next billing cycle.

Competition

Telecom operators are not the only ones who see this gap as an opportunity. API aggregators and hyperscalers are building commercial systems for machine customers right now.

Aduna — the joint venture backed by Ericsson and 12 major telecom operators, including AT&T, T-Mobile, Orange, Telefónica, and Jio — has built a single global platform giving enterprises access to network APIs from 40 operators through one contract, one price, and one integration point. By February 2026, Aduna had agreements with three major US carriers, covering 300 million connections, accessible through a single commercial relationship. For the telecom industry, the convenience this brings is real, but so is the risk. If Aduna becomes the primary commercial interface for machine customers consuming telco network services, individual operators become wholesale suppliers behind someone else's commercial layer, with no direct relationship with the buyer.

Hyperscalers are moving in the same direction. Nokia and Google Cloud announced at MWC 2026 that enterprise agents can now discover, configure, and consume telco network services through Google's agentic AI framework, with no coding required. The enterprise agent talks to Google, and the telco provides the underlying capability. Similarly, Microsoft has embedded standardized telecom network APIs, developed under the CAMARA open-source framework, as native services in Azure Marketplace, making them directly accessible to enterprise developers and agents through Microsoft’s own platform. The pattern is the same in each case: discovery, pricing, entitlement and settlement move to the platform. The operator carries the traffic.

Commercial programmability

Closing these four gaps requires a fundamental shift in how telecom companies think about their commercial systems.

Done well, it reflects in product catalogs that machines can read, prices that adjust dynamically, governance that enforces itself, assurance that is verified in real time, and settlement that completes automatically. Together these capabilities define what it means for a telecom company to become a commercially programmable platform — one that can serve both human buyers and machine customers without friction.

This shift follows a pattern the industry has experienced before. Each generation of telco value creation expanded what operators could sell, and with it the commercial model that supported it (Figure 1). The next stage is the trust and assurance economy: where identity, policy, risk and settlement become the commercial foundation, and where AI agents are among the buyers.

Figure 1. Telecom value evolution and commercial implications

Source: Infosys

The trusted commercial orchestration

The mechanism that delivers commercial programmability is a trusted commercial orchestration layer — a real-time engine sitting between the telecom operator's network capabilities and the customers, both human and machine, trying to access the services. Unlike traditional commercial systems built for human-mediated processes, this layer operates at machine speed, across five core functions:

Service discovery and readability: The commercial orchestration layer exposes product catalogs and service offerings in structured, machine-readable formats, so an AI agent can find, interpret and compare options without human translation. This is the first and most fundamental requirement. An agent cannot buy what it cannot read.

Dynamic pricing: The layer prices services dynamically, adjusting in real time based on demand, service quality, risk, duration and business context. An agent reserving network capacity for a port automation window needs a confirmed price before it acts, not a quote that arrives the following week.

Charging and settlement: This meters consumption, generates charges, and settles automatically at transaction close, with audit-ready records built in. For machine customers operating at scale, settlement must happen at the moment the transaction completes.

Assurance and SLA control: It continuously verifies that the promised service level is being delivered. If not, compensation is triggered automatically, without a human raising a support ticket. This is what transforms a network service from a best-effort connection into a commercially accountable outcome.

Agent governance: The four functions listed above make commerce executable. But they cannot operate without answering the accountability questions raised earlier: who owns the agent, what authority it carries, what it can spend and who bears the liability. This requires telcos to incorporate ‘Know Your Agent’ as a practice — the commercial equivalent of Know Your Customer. Every transaction an agent initiates must pass through a governance check that resolves these questions before the transaction is allowed to proceed. Telcos already operate under some of the most demanding regulatory and compliance frameworks in any industry. Applying that governance rigor to the commercial layer means every automated transaction is verified, traceable and disputable.

The trusted commercial orchestration layer does not ask operators to build something entirely new. It asks them to apply what they already do to a new kind of customer. In the agent economy, the company that establishes trust, governs transactions, verifies delivery and settles commercial obligations across ecosystems will own the relationship. Telecom operators are already doing all this, at national scale, every day, in a way that hyperscalers and API aggregators are not.

Stages toward autonomous commerce

Commercial opportunity does not arrive all at once. It follows a staged progression (Figure 2).

Figure 2. Five-stage commerce progression

Source: Infosys

Most telecom operators and their enterprise customers currently sit at stage two of the autonomous commerce progression. Enterprise systems trigger service requests automatically, but the contract and invoice behind each request remain human. For stages three and beyond, enterprises require the trusted commercial orchestration layer described in this paper. The operators that build it now will be ready when those stages arrive.

Where to start

The commercial orchestration layer does not need to be built all together. Five revenue pools sequence the opportunity by commercial confidence and time horizon, giving operators a practical path from where most sit today to where the agent economy is heading (Figure 3). Initial pools, that belong to the stage two category, are already starting to generate revenue. The later ones anticipate more autonomous buying relationships of stages three and beyond.

Figure 3. Revenue pools by time horizon

Source: Infosys

Start with identity and fraud

The highest-confidence, nearest-term opportunity is already generating revenue. Telefónica, Orange and Telstra are already selling SIM swap and number verification APIs to banks, fintechs and digital platforms. The commercial model is proven. The demand is immediate. Banking and financial services leads the GSMA Open Gateway demand index, followed by media, healthcare, retail, and manufacturing.

Start here. Design identity and fraud services as a reusable commercial pattern from the outset, with product catalog, policy, pricing, assurance and settlement aligned from day one. That pattern becomes the template for every subsequent revenue pool.

Expand into quality-on-demand and edge

As the identity pattern matures, the next revenue pool opens. Quality-on-demand services, including dynamic network performance for media, industrial automation, mobility and healthcare applications, represent the next opportunity. Edge and compute orchestration follows, packaging connectivity, compute and security for latency-sensitive enterprise workloads.

Monetize network intelligence as service

As edge and compute services scale, a parallel opportunity opens in the data that operator networks already generate. Location signals, population density, movement patterns and device status are produced continuously as a by-product of serving subscribers. Transport authorities, city planners, and public safety agencies are beginning to pay for this intelligence as a packaged service at per query rates or in the form of subscription models. The commercial model simply requires packaging this data in compliance with privacy regulations and in formats that machines can read and act on.

Build for the agent-to-agent marketplace

The next is agent-to-agent marketplace — a machine-speed commercial environment where AI agents discover, negotiate, consume, and settle network services autonomously. These are stages four and five of the commerce progression. The operators that build the commercial orchestration layer now will be positioned to serve this market when it arrives.

One principle underpins the path forward. The operators best positioned to offer commercial orchestration as an external service are those that have already become practitioners of it internally. Frontier Telco describes how operators make that internal transformation. By deploying AI across customer service, network operations, software engineering and enterprise functions, operators build consumption management, governance and cost optimization capabilities that are proven in their own environment. Those same capabilities, once productized, extend to retail, small and medium businesses, and enterprise customers — turning internal AI adoption from a productivity initiative into the foundation of a new growth business.

In the long run, value will accrue to the operators that can serve every type of customer — human or machine — and across every type of service, by establishing the trust, governance, and commercial infrastructure that autonomous commerce demands. Telecom operators have spent decades building exactly that foundation. Extending it to machine customers is the natural next step.