Within ten years, AI could leave society far better informed. Almost nobody will notice. That is the point, and the challenge.
There are no altars to the gods of showers, card payments or traffic lights. Nobody wakes up in grateful wonder that the water runs hot, that money moves at the tap of a card, or that a busy junction does not descend into carnage. These are among the great achievements of modern life, and they are almost entirely invisible. Wake up to find any one of them broken, though, and your day is ruined. Possibly worse.
The information environment of the near future should work the same way. Most people do not think about it, and they should not have to.
Too tired to be informed
Consider a new mother today, six months in and running on broken sleep. This week's headlines told her the baby-food pouches she has been relying on are loaded with sugar; she feels guilty rather than helped. The government has just changed its childcare entitlements, and she stands to save thousands. She has seen the news. But her circumstances are slightly complicated, and none of the coverage quite answers whether, or how, the change applies to her. The answer exists. It is public, accurate and free, sitting in a policy document she never has the time or energy to read. She is not misinformed. She is too tired to be informed. The money goes unclaimed.
Ten years from now, another new mother is just as tired. But an agent acting on her behalf knows the rules have changed, works out what she is entitled to given her particular circumstances, and drafts the application for her to check. When a local investigation exposes failings at the maternity unit where she is booked for a follow-up, she hears about it that morning, with her options laid out. The decisions remain hers. If she asks how it knows, it can show her: where each claim came from, who checked it, and how other parents nearby have tested it, questioned it or disagreed. Then she goes back to sleep.
By 2036, there could be a step change in how well informed people are, and organisations, communities and governments too, delivered by an agentic information economy that works. Most of society's decisions will rest on it, and much of society’s critical functions will depend on it. Yet to the average person, it will barely seem to exist. Nobody will build an altar to it. Ingratitude will be the sign that it works; we might notice only if it breaks.
Emergence of an agentic information economy
Two linked shifts make this possible. The first is that machines are becoming the primary audience for information. A growing share of what is published, recorded or measured will be read first, and often only, by AI systems acting for people, and increasingly by agents acting for other agents. That leads to the second shift: information changes category. For most of its history, news, information and knowledge have been a final good, often produced antecedent to consumption and read by a person who may or may not act on it. In an agentic economy, they become an intermediate input into a decision, a transaction or an automated process. In most cases, they may never be rendered for a human at all.
Together, these shifts could give rise to an agentic information economy: the epistemic bedrock on which the wider agentic economy rests. It could be far bigger than the information economy of today, because machines consume on people's behalf without the limits of human attention. It could be far more complex: glance at a map of a language model's embeddings, or at an event-centric knowledge graph, and imagine most of our knowledge and context interlinked like that.
The price of being right
Unlike today's information economy, an agentic one could run in favour of accuracy, truth-seeking and self-correction.
To see why, start with what an error is. When a person misreads something, the error becomes an impression. When an agent acts on it, the error becomes an action: a wrong purchase, a missed entitlement, a recalled product back on the shelf. In agentic systems, outcomes loop back as inputs, so errors compound at machine speed and scale. And because so many systems draw on the same few models, they can fail together. This is more than "garbage in, garbage out": correlated, automated error is a different class of risk from misinformation.
It is also a different class of opportunity. For as long as there has been a digital information market, being right and sounding right have fetched roughly the same price, except in finance and a few other high-stakes niches. Once the cost of being wrong becomes measurable, and at potentially far greater cost, that need no longer hold, and the exception could soon extend to almost all information. Provenance, accuracy and the capacity to self-correct could become things that are priced. Across much more of the information economy, being right could start to pay.
But here’s the catch. The best version of an agentic information economy would span almost everything we know: personal data, niche expertise, the knowledge inside organisations, government records, real-time events and the findings of investigations. For it to be truth-seeking and self-correcting, it needs certain conditions, such as ownership and control that are spread out rather than concentrated, and no moral hazard in how information is gathered and verified. This will need a myriad of native protocols, institutions and rules to govern it. That is no easy feat, and the best case is unlikely to emerge by default.
Complex information systems are not new
Finance offers a useful comparison. An agentic information economy, in which news, information and knowledge are sourced, verified, exchanged, priced and recombined by agents on behalf of people and of each other, could become as complex as, if not more complex than, the current global financial system, and far less legible. Most of us have only a folk understanding of finance, yet we hand it our salaries and pensions. We don't fully understand it, because we don't really need to. We can rely on it because of protocols, institutions and regulation to which we delegate trust, legibility and risk: shared identifiers and messaging standards, custodians, rating agencies, insurers, clearing houses, and the regulators that oversee them.
The agentic information economy will need analogous systems, and probably more. Finance has SWIFT and shared identifiers, which let banks message each other and know exactly which party and which asset a transaction refers to; information needs a shared protocol and a standard unit, so that a claim can be identified and traced. In the agentic information economy, agent protocols such as MCP and content credentials such as C2PA are early versions of comparable systems. Finance has custodians, who hold assets on their owners' behalf; information needs trusted holders of dispersed, niche knowledge. Data cooperatives such as Switzerland's MIDATA, whose members decide who may use their health data, are an early version. Finance has rating agencies; information needs calibrated confidence in its sources. Probabilistic news, which attaches explicit likelihoods to claims, is an early version. Finance has clearing houses; information needs corrections that reach every system that used the original. Crossref's Crossmark, which flags corrected and retracted research, is an early version.
These systems also decide where value flows and where it is captured. In an agentic information economy, value is likely to pool at the interface, with whoever knows the most about the user. With the right rails, market mechanisms and institutions, some share of it, and of the demand signal (knowing what the user wants), could flow back along the chain to those who originate, verify and maintain information. A crude example from finance shows it can be done: card interchange fees send a slice of every card payment back to the bank that issued the card.
The right blueprints matter
Finance also shows how such institutions fail. The rating agencies that blessed mortgage-backed securities before 2008 were paid by the issuers whose products they rated. And the same infrastructure that makes finance work has also enabled extraordinary concentrations of wealth and power. Journalism and science encode the opposite instincts: independence from those you cover, disclosure of sources and methods, scrutiny by peers, public correction. Those are the right blueprints. Whether they are used is being decided now, while the standards are still soft. The agentic information economy is so early that it does not yet have a standard way to represent a piece of information, let alone the institutions to build on one. In October 2026, Meta and Sierra announced a protocol for how personal agents deal with businesses, with Walmart, Shopify and Stripe on board. The foundations of the agentic economy are being laid, but the information systems it runs on are lagging behind.
This is a space journalism should occupy, though not in the form it takes today. Its institutions already hold much of the DNA: verification, editorial standards, archives, corrections policies and a working ethic of accountability. Within ten years they could be taking on functions that look little like journalism today: originating machine-ready, verified events with their provenance attached; acting as custodians of a community's local and niche knowledge; rating sources and attaching calibrated confidence to claims; underwriting the information agents act on, with liability attached; and running the clearing systems that carry a correction to every system that used the original.
Journalists will also need to watch this new system. A system this consequential and this opaque will need watchdogs of its own. Covering the agentic information economy will become a beat, much as finance and technology are today. Someone will still need to hold information power to account; the power will simply be harder to see.
The invisible fourth estate
None of this displaces storytelling as we know it. Stories are how people come to care, and they will flourish alongside. But they will not be the only way people are informed. There is room for a new discipline of information and knowledge to emerge, one that no existing field, from writing and journalism to economics to computer science, fully addresses today.
So what should newsroom leaders do now? Start from first principles. That means asking what an agentic information economy will actually need, and looking well beyond the news industry for answers, rather than searching for a best-in-class version of something that already exists. It means rethinking what the unit of information, and therefore of value, actually is, building the economics around that, and developing the skills to work with information in ways the newsroom was never designed for. And it means a level of coordination and getting into the rooms where the standards for this economy are being written, while they are still soft.
If the future agentic information economy works for a better-informed society, with its value and power more evenly distributed, most of what it does will be invisible, and most people will never think about it. They should be able to trust it, without fully understanding it or inspecting it for themselves. But invisibility cuts both ways. A system in the background that can quietly leave you better informed can just as quietly manipulate you, and you would probably never even know it was happening. That is all the more reason to build it to serve citizens rather than to steer them. It also sits uneasily with journalism's cultural identity. As the fourth estate, journalism is used to occupying a visible seat in society, and wearing its trust on its masthead. That will have to be reckoned with. But our future selves deserve what journalism has always offered at its best: accountability, legibility, a willingness to be corrected, and the nerve to bark at power. There is a role for journalism in that future. It just might be unrecognisable.