Michelle Seo does not start her day on a homepage. In an October 27, 2026 WIRED story, Kate Taylor described the San Diego public-relations consultant getting a morning briefing from an agent she calls Chattiana: agenda, to-do list, newsletter headlines framed as water-cooler topics, and links back to the sources. She uses the briefing the way other people use a commute, so she has something to say on Zoom. She told WIRED she is challenging herself, as an introvert, to start more conversations.
The habit is ordinary now. What publishers are building in response is not. Taylor reported that some publishers have started treating AI agents, not only human readers, as an audience. If agents become the default reader, the homepage is a lobby few people walk through. The story has to survive a reader that never scrolls.
A feed built for a machine, sold first to a company
Axios Direct sits at the center of that account. Axios Direct, on WIRED’s clock, was announced at a September Digiday Publishing Summit. Axios launched the first feed in fall 2026 and planned the other two for 2027. The fall 2026 feed gives AI agents of Axios Pro Deals subscribers access to merger-and-acquisition reporting. The 2027 pair would do two jobs. One would supply material for training in-house models. A consumer feed would send Axios reporting to an individual subscriber’s agent. All three, WIRED said, would resemble RSS: standardized lines of text, not pages built to please a human eye.
Jacquelyn Cameron, Axios’s chief revenue officer, put the bet bluntly. The open web is deteriorating. Search is shrinking. Hyperpersonalization is surging. Large language models matter more in how people get news. She said Axios wants to be where the audience is.
We do believe that a significant portion of news consumption will move into AI systems, LLMs, and devices.
That is an executive belief. It is not a measured share of the market. The product description is more concrete. The feeds cut narrative in favor of structured data. Cameron compared them to a Bloomberg terminal more than a series of articles.
Axios is starting with companies, not consumers. An all-access Axios Pro Deals subscription, WIRED reported, costs $2, 499 a year. Companies already pay that because a slight edge in deal news can be worth real money. WIRED disclosed no price for any of the three Direct feeds. A pipe a corporation will test is not proof that a consumer’s agent will pay for the same thing.
Allison Murphy, Axios’s chief operating officer, pulled the thesis the other way. She argued that Axios’s short, highly structured style already makes it a favorite source for AI answers, because crawlers do not want to hunt three quarters of the way down a page for the point. “Time is money for crawlers, ” she said. She also refused the existential framing. AI is not really the audience, she said. It is a tool for reaching the audience, in the same family as the moves toward social, short video, and podcasts.
Hold both lines at once. One executive is designing distribution for agents. The other says the agent is a courier. A company can believe both and still not know which invoice gets paid.
The article that can be remixed
Every, the tech media and software company, is the other experiment in the piece. Less data terminal, more workshop. Founder Dan Shipper is working an old editorial anxiety with a new exit door. A writer has to choose what goes into the story, he told WIRED, and those choices may be wrong for a particular reader. His answer is to stop pretending one cut serves everyone.
Writers at Every include prompts so readers can test the ideas. Some guides carry “Read With Claude” or “Read With ChatGPT” buttons. The handoff tells the model: “Start with a tight summary: one paragraph, bulleted. Then offer to go deeper.” Two years before WIRED’s story, Every had shipped what it called an extendable piece. That was a feature plus a database of reporting, including interview recordings, news articles, and YouTube videos, that a chatbot could query. More recently, WIRED said, an 8, 000-word article about Codex, OpenAI’s coding agent, let subscribers’ agents draw on that database. Those agents could recast the work as a bulleted summary, a podcast, a video, or a short film of interview bits. Shipper’s line for what comes next was plain: “People are gonna do crazy, crazy stuff with this.”
WIRED put an annual Every subscription at $288. Shipper treats that price as covering a human reader and that reader’s agents. He may not want every story himself, he said, but he may want his agent to know the work well enough to tap him when something matters. That is a theory of willingness to pay. It is not a renewal chart.
Do not confuse a new audience with a faster back office
By fall 2024, major publishers were already putting AI on the money side of the business. A Digiday preview of that September’s Publishing Summit had Business Insider’s chief subscription officer, Katie Friedman, set to discuss an AI paywall and a reported 75 percent higher conversion rate than the old approach. Treat that figure as a claim she was going to make on stage. It was not an audited result. The surrounding conversation was about fragmented attention, changing search, and social platforms that keep people inside their walls. None of it was an announcement of agent feeds.
Operators should keep the two stories apart. AI automation for business, in the 2024 version, meant sharper paywalls, faster ad plans, and AI for sales workflows a company already runs. The 2026 version WIRED reported is a different decision. It asks who the story is for, and whether the unit of value is a narrative a person finishes or a structured record a system can query. A better conversion rate on a human paywall does not answer that.
Licenses, a file the crawler can read, and a lawsuit that has not settled this
Publishers are not answering AI with one strategy. WIRED described several running side by side: licensing so platforms can surface content, including deals signed by major publishers and by WIRED’s parent, Condé Nast, along with generative engine optimization, or GEO, meant to get articles into answers to prompts, and technical files that help models interpret a site. WIRED said the WordPress plug-in AIOSEO launched such a feature the year before its story, and spelled the file LLM.txt. The name more often used in the industry is llms.txt. Then there is litigation. Mash those together and you get a slogan. Keep them separate and you get a risk map.
The New York Times sued OpenAI and Microsoft in 2023, alleging illegal training on its reporting. As of October 27, 2026, WIRED described that case as still ongoing. The clearer legal marker from the period around it is Anthropic. That case is not a ruling on agent feeds. In June 2025, U.S. District Judge William Alsup largely accepted fair use, on summary judgment, for training on lawfully acquired books. He let the claims over allegedly pirated books proceed. In September 2025, Anthropic agreed to pay authors $1.5 billion. That settlement spoke to how books were gathered and used in training. It did not hold that an agent summary is fair, illegal, or anything in between.
The question a publisher should ask before celebrating that summary is simpler than a docket. Does it send a reader back, or does it replace the thing they used to buy? Nothing in the Times case or the Anthropic settlement decides the legality of Axios Direct or of Every’s remix tools. Training lawsuits, licensing pressure, and agent-facing formats can sit next to each other without being the same bet.
Visibility is not a business model
Rasmus Kleis Nielsen, a University of Copenhagen communication professor quoted by WIRED, called the market for publishing to agents emerging, immature, and extremely uneven. Price is unsettled. So is future agent behavior: how a system retrieves, attributes, summarizes, or quietly drops the caveat you spent a paragraph earning. His comparison was to publisher clips on TikTok and Instagram, which can build a brand and still almost never send anyone home to the site. Useful analogy. Not a traffic study.
Say this once, because it covers the rest. WIRED published list prices and executive hopes. It did not publish adoption counts, citation share, referral loss, or revenue for the Direct feeds. Murphy’s crawler argument, Nielsen’s off-platform analogy, and Shipper’s idea that a subscription should cover the subscriber’s agents are three theories sitting next to each other and getting mistaken for one result. Any of them might be true. None of them has been shown, in public numbers, to pay a newsroom.
What WIRED did say people will still pay for is information other publications and models do not already have. In-the-weeds AI guidance at Every. Fast, market-moving deal reporting at Axios Pro Deals. Commodity explanation is a hard sell once a chatbot can draft it. A proprietary feed, or a database the agent can query because you hold the interviews, is a different product. That is closer to selling a terminal than selling a magazine, and terminals are priced like tools, not like content.
Questions worth asking before you rebuild around agents
-
Are AI agents already a real media audience, or a prototype?
WIRED reported live experiments at Axios and Every, aimed first at corporate users and paying subscribers. Nielsen’s view in that piece is that the market is emerging and uneven: a prototype with a sales pitch, not a settled audience.
-
Who pays, the human or the agent?
The prices WIRED published are human subscriptions: $2, 499 a year for all-access Axios Pro Deals, $288 a year for Every, which Shipper treats as covering a reader’s agents too. The three Direct feeds were not priced, and nothing in the reporting says an agent pays on its own.
-
Does showing up in an AI answer replace the visit?
It can be a brand win and a traffic miss at the same time. Nielsen’s TikTok and Instagram comparison is the right shape for that risk: people may hear your name inside an answer and never walk back to the site.
-
What happens to judgment, and to copyright, when an agent remixes the reporting?
Every’s tools, as WIRED described them, let a subscriber’s agent turn one database into a summary, a podcast, a video, or a short film. That may fit the reader better than a single edit, and it may also replace the thing they used to buy. The Times case and the Anthropic settlement have not answered that for these products.
-
Is this the same project as AI automation inside ad sales and paywalls?
No. Speeding a paywall or an ad plan improves a business you already have. Designing a feed a crawler prefers is a decision about who the story is for. Winning at the first does not answer the second.
The practical read for a leadership team is narrower than the slogan. An agent is a distribution channel with little patience and no loyalty to your design. Structured reporting may travel farther inside other companies’ models. That does not, by itself, create a market, a price, or a reason for the model to send the human back. Media has changed containers before. This one can rewrite the piece and hand your subscriber a version you did not edit. Build for that reader if you must. Just do not confuse a courier with a customer until someone pays.