YOU ARE AT:6G4G was for humans, 5G was for machines, 6G will be for...

4G was for humans, 5G was for machines, 6G will be for AI agents (Reader Forum)

In the 6G era, AI agents will issue requests, negotiate prices, and buy and sell network slices or compute, connectivity, and energy in real time

We live in a world dominated by digital services. From a technological perspective, we’ve never innovated at this pace or scale before, and networks are racing to keep up. Like laying a track in front of a moving train, networks are scrambling to be ready for the next era of telecommunications, but from now on, there’ll be one key difference: Humans are no longer the main customers.

More than 74% of new webpages contain AI content. As of November 2024, 70% of Fortune 500 companies have rolled out Microsoft 365 Copilot, with 85% entrusting Microsoft with their integration and innovation decisions. Even email, one of the oldest and most conservative channels, is going bot-first, with 25% of employees using AI to write messages. Add in Gartner’s forecast that 80% of enterprises will have Gen AI in production by 2026, and it’s no stretch to say that machines will soon mostly be talking among themselves.

Yet, telecom networks are still being built with people in mind. That was the fatal mistake with 5G: operators designed it for human consumption, when its true users were always going to be machines. With 6G, that mistake can’t be repeated. Because this time, it’s not just about machines, it’s about the AI agents running on them.

These agents won’t just generate content or execute commands. They’ll issue requests, negotiate prices, and buy and sell “network slices” or compute, connectivity, and energy in real time. If operators want to stay relevant in that world, their networks and the software that supports them need to be redesigned for a new kind of customer: AI code.

Why 5G won’t cut it

The premise of 5G was sound: ultra-low latency, high throughput, and more reliable connections to serve the rising tide of connected devices. But in execution, most operators misjudged the real audience. Rather than preparing networks for industrial automation, smart infrastructure, and machine-to-machine (M2M) interactions, the ecosystem stayed focused on human-facing use cases, including faster video streaming, better gaming experiences, and high-speed mobile data.

The result? An expensive, underutilised rollout that failed to unlock the full value of 5G.

Today, Cisco reports that M2M connections already account for half of all connected endpoints, and AI applications are expected to push uplink demand beyond what current spectrum allocations can deliver as early as 2027. Even Ericsson projects that total mobile data traffic will triple again by 2030, long before 6G spectrum is widely available. The assumptions baked into the 5G era are quickly becoming obsolete.

AI agents don’t work like people. They don’t consume content casually or log on during business hours. They act continuously and programmatically, exchanging data with other agents, sensors, or back-end systems in real time. That means the traditional architecture of mobile networks, built around static provisioning, long-term SLAs, and one-size-fits-all pricing, simply won’t work.

These agents will demand flexible, low-latency access to not just connectivity, but also compute, storage, and energy resources. And they’ll expect to negotiate these parameters dynamically, based on goals, priorities, or constraints like carbon budgets. For that to happen, operators must start building networks that are no longer optimised for coverage or capacity alone, but for negotiation, orchestration, and intent-based execution all at faster-than-human speed.

The 6G blueprint: AI by design

Unlike its predecessors, 6G isn’t just a faster version of what came before. It’s a structural rethink of what a network is and who it serves. The International Telecommunication Union’s (ITU) IMT-2030 framework, which sets out the official vision and requirements for 6G, positions 6G as AI-native from the outset, embedding artificial intelligence into both the radio interface and control plane. It introduces concepts like integrated sensing and communication (ISAC), where the same electromagnetic signals that transmit data can also perceive and interpret the physical environment.

That means networks won’t just carry information. They’ll observe, react, and coordinate autonomously. A swarm of drones responding to a wildfire, for instance, won’t rely on manual control. They’ll navigate hazards, adjust routes, and relay data to one another in real time, cooperating entirely via machine-to-machine interaction. 5G might enable a drone to receive instructions, while 6G will allow drones to sense and self-organize based on their surroundings.

Today, 6G testbeds are already pushing the boundaries of latency, device density, and spectral efficiency. Terahertz and sub-terahertz frequencies will enable on-demand, terabit-per-second links. Device densities are forecast to reach 10⁸ per square kilometre, enough to support hyper-connected environments filled with sensors, digital twins, and edge AI agents. Latency will drop to sub-millisecond levels, opening the door to truly synchronous applications like holographic communication, immersive extended reality, and AI-driven financial micro-trades.

But these capabilities won’t be delivered in fixed packages or user tiers. Instead, they’ll be negotiated, brokered, and continuously optimised based on the specific, moment-to-moment intent of the agent in question. For telecom operators, this will require a departure from static provisioning and a move toward programmable, AI-managed infrastructure designed for real-time, dynamic allocation.

A new experience layer

While M2M takes on more of the provisioning work, the overall “experience” of the network needs to remain very much human. But instead of tapping through interfaces or clicking “confirm,” users will simply express their intent, and agents will act on it. For a B2C consumer, that might mean negotiating a ride, booking a hotel, or orchestrating a work task across multiple services.

Increasingly, these agents won’t be interacting with APIs designed for human latency. They’ll be interfacing with other agents in real time, haggling over price, bandwidth, availability, and even carbon constraints. For telecom operators, this means moving away from static service tiers and toward a marketplace model where requests are dynamic, conditional, and context-aware. In this world, “connectivity” isn’t just a commodity to be taken for granted. It’s a smart, tradable resource that gets priced and provisioned per transaction, per intent, per joule of energy it consumes.

Take something as mundane as a scheduled cab ride. Today, a user opens an app, chooses a ride, and confirms the booking. But in an intent-driven model, the agent knows the user’s calendar, understands their location and preferences, and issues a real-time query: “Find me a vehicle to arrive at 10:00 AM, under $12, within a five-minute wait window.”

That agent doesn’t just call one API. It negotiates with multiple transport providers’ agents, weighing price, carbon footprint, and latency. And in more complex enterprise scenarios, such as real-time digital twin orchestration or autonomous supply chain coordination, these transactions could involve per-second network slices bundled with compute and storage, activated and torn down in milliseconds. The experience layer will ultimately still be for people, but it won’t be negotiated by people. In short, it won’t be built for people.

Who owns the machines?

As AI agents become the dominant actors in the digital economy, the parameters that define “value” will shift dramatically. Pricing models will no longer be based purely on data volume or time. Energy consumption, measured in joules, carbon impact, or sustainability score, will become a core metric for both performance and billing. Do not be surprised if we have GB/joule charging and billing models in the future.

With resource-intensive AI and blockchain processes trading slices in real time, operators will need to track, manage, and monetise not just bandwidth, but the energy required to deliver it. That creates new opportunities: eco-efficient agents, low-carbon slices, and intent-based service plans that optimize for environmental as well as economic outcomes. But it also introduces new responsibilities.

As networks grow more autonomous, they must remain auditable, transparent, and accountable. Trust will depend on open telemetry, programmable guardrails, and the ability to detect and remediate exploitative behaviours, like slice hoarding or pricing bias, without human intervention. This will demand a reinvention, or at least a reevaluation of telecom software.

Business support systems (BSS) must evolve to support dynamic, millisecond-level mediation, intent-based service catalogs, and real-time marketplaces for slices bundled with compute, connectivity, and carbon. Traditional CRM models will no longer apply because customers won’t be people, they’ll be agents. Success won’t be measured by NPS scores, but by continuous agent quality (CAQs), service uptime, negotiation efficiency, and energy footprint.

And while the end-user brand may still own the relationship, the operator will own the infrastructure of trust, arbitrating disputes, enforcing policies, and ensuring fairness across billions of machine-to-machine transactions. In this world, and it’s coming soon, operators will evolve from service providers to economic orchestrators. The challenge now is to build the systems that make that future not just possible, but reliable, sustainable, and safe.

AI agents won’t be waiting around for networks to catch up. They’ll be shaping their own demands, brokering their own connections, and expecting infrastructure to respond. All of this change will still be in service to humans, but the operators that succeed will be the ones who treat code as their number one customer and build networks ready to trade on its terms.

ABOUT AUTHOR