AI agents are starting to talk: Are the networks ready?

Humans can watch what happens on the platform, but they cannot post or interact. Instead, autonomous AI agents create threads, reply to one another and upvote content, effectively socialising without human involvement.

It is easy to dismiss this as a novelty. In many ways, Moltbook probably is one. But from a connectivity perspective, it is still worth paying attention to, not because of what it is today, but because of what it quietly reveals about where network demand may be heading.

Strip away the novelty, and Moltbook becomes a simple demonstration of AI systems talking directly to other AI systems, continuously and at scale. There is no human pacing, no daily rhythm, no natural downtime, and that is where the implications for networks begin.

Global connectivity has always been shaped by human behaviour. Even as traffic volumes have exploded, from video streaming to cloud services, usage has remained broadly predictable. There are peaks and troughs, working hours and quiet periods, seasonal effects and geographic patterns. Networks, pricing models and capacity planning have evolved around those assumptions.

AI agents don’t fit into that model. They don’t sleep, nor do they follow time zones. And they definitely do not behave like regular users.

If platforms like Moltbook are an early glimpse of a broader shift toward autonomous systems coordinating with one another, then AI-to-AI traffic introduces a new type of demand: persistent, machine-generated and difficult to forecast. Activity may rise and fall based on optimisation cycles, training runs, or internal coordination, signals that are largely invisible to network operators.

Just as important is where this traffic sits. AI agents are not edge devices; they run almost entirely inside hyperscale cloud environments and data centre clusters. Their conversations generate east-west traffic, between data centres, across private interconnects and through internet exchanges, rather than traditional north-south flows between users and applications.

That has consequences for where capacity pressure builds. The growth story shifts further away from the last mile and deeper into metro fibre, inter-DC connectivity, IX capacity and private networking. For many operators, this is already familiar territory, but AI agents could accelerate the trend.

There are commercial implications, too. Machine-driven traffic is inherently harder to model than consumer usage. Long-term capacity commitments become more challenging when demand is less predictable, increasing the value of flexible, on-demand provisioning and shorter planning cycles.

Security and governance also deserve attention, given that AI-only platforms generate large volumes of automated traffic with limited human oversight. Feedback loops, runaway signalling or abuse could stress networks in ways existing monitoring tools were not designed to catch quickly. This is less about speculative AI risk and more about day-to-day operational resilience.

Moltbook itself may never matter at scale. Its real value is as a visible signal of a broader shift already underway: a growing share of network traffic generated not by people, but by machines talking to machines.

As AI agents become embedded in enterprise software, financial systems and automated operations, AI-to-AI traffic will move from edge case to material consideration. The question for the connectivity industry is whether networks are being prepared for that future now or whether the adjustment will only come once the traffic graphs start to change.

AI fatigue in telecoms: When the hype outruns the hardware