Ericsson forecasts 38% will delegate tasks to AI agents by 2030: the figure is an expectation, not a measurement

By TeleSemana · September 17, 2026.

Ericsson ConsumerLab, the market research division of the Swedish telecommunications equipment maker, estimates that 38% of current smartphone users expect to delegate everyday tasks to artificial intelligence (AI) agents by 2030. Today, according to the same source, only around 4% use these capabilities. The figure comes from a survey of more than 43,000 people aged 15 to 69 across 27 markets — among them Brazil, Chile and Mexico — a sample Ericsson says is representative of some 1.7 billion smartphone users worldwide.

Agentic AI, unlike an assistant that has to be given an instruction for each query, can chain together and carry out several actions on the user's behalf without constant supervision. Among the tasks respondents imagine delegating are scheduling and coordinating activities, comparing products, monitoring recurring purchases and tracking health. Participants estimate this could save them up to an hour a day of manual browsing.

It is worth flagging a nuance the report itself acknowledges, even if the headline dilutes it: the 38% is a self-declared expectation about what people think they will do four years from now, not a measurement of actual adoption. The literature on surveys of future intent is consistent on this point: people systematically overestimate how much their habits will change. And there is a second nuance that does not appear in the press release: the study is signed by a network infrastructure vendor whose business depends on operators investing in capacity, identity and security to manage agent traffic. That does not invalidate the data, but it does require reading it for what it is: a piece of commercial argument with real data behind it, not a neutral verdict on the future.

Within the study itself there is a pattern more interesting than the headline figure. Ericsson already identifies a group of "AI superusers" today — around 10% of smartphone owners, defined as heavy users of multimodal tools (those combining text, voice and image) — who spend two and a half times more time on AI services than the average user. Ericsson projects that this group could grow to 30% by 2030, and that the usage gap with the average user would widen to fourfold. In other words: adoption is not advancing uniformly but is concentrating and accelerating among a minority already hooked, while the bulk of the user base follows a curve far slower than its own stated expectations. It is the same uneven pattern we have been observing in other areas of AI: the gap between those who use it intensively and those who barely touch it is growing faster than average adoption.

For mobile operators, the underlying issue is not only how much additional traffic agents will generate — Ericsson's own press release avoids giving a figure on that — but what kind. If a meaningful share of manual searches and comparisons is replaced by automated requests between apps, agents and cloud services, much of the network load stops being a person opening an application and becomes background machine-to-machine traffic. That forces a resolution to something that is unresolved today: how an agent acting on a human's behalf is authenticated, how an action is attributed to it, and how it is prevented from acting without permission. It is the same blind spot — the identity and oversight of autonomous agents — that we have been flagging as one of the most urgent regulatory and security fronts of this transition, and which so far has been addressed more through isolated incidents than by design.

Our reading is that the 38% headline says less about 2030 than about the script the infrastructure industry is beginning to write to justify investment in "agent-ready" networks. That does not make the underlying issue any less interesting: freeing up an hour a day of human browsing so an agent can do it is exactly the kind of time abundance that, multiplied many times over, is the underlying promise of this technology. The problem, in the short term, is that the promise is arriving ahead of the answers about who polices those agents, who is accountable when they act badly and who pays the bill for doing it securely. That gap — between stated ambition and the infrastructure of trust that underpins it — will determine, far more than the percentage itself, whether the transition to a life mediated by agents is as smooth as the survey promises.