Every morning Michelle Seo, a public relations consultant in San Diego, gets a briefing from Chattiana, the AI agent inside her Muse app. It lists her agenda and to-dos, then plucks conversation starters out of the headlines in her newsletters, links included. She describes it as training for an introvert who would rather not sit silent on a Zoom call. That is the first half of a shift now running through publishing. The second half is that newsrooms are starting to produce for the agent rather than the human behind it.
Axios announced Axios Direct at the Digiday Publishing Summit in September: feeds built for AI agents. The first went live this autumn, giving the agents of Axios Pro Deals subscribers access to its mergers-and-acquisitions reporting. A second, due next year, will serve as a source for training customers' in-house models. A third, also next year, is aimed at consumers so their agent receives Axios reporting directly. All three look like RSS: standardised lines of text, not designed for the human eye. "Time is money for crawlers," says Axios chief operating officer Allison Murphy. "They don't want to have to go three quarters down the page to find the main point."
Chief revenue officer Jacquelyn Cameron frames the move around the state of the open web — deteriorating, with search shrinking, hyperpersonalisation rising and large language models becoming a bigger part of how people consume news. "We have always said that we want to be where the audience is," she says. An all-access Axios Pro Deals subscription costs $2,499 a year.
Every, a tech publication and software company, attacks the same problem from the other side. Founder Dan Shipper describes the typical difference from a normal article this way: wherever possible, writers embed prompts a reader can paste straight into a model. Guides carry buttons reading "Read With Claude" or "Read With ChatGPT" that open a prompt — a tight summary first, then an offer to go deeper. The effect is that writers can pack in detail without worrying about losing the reader, because the agent will pull out what matters. An 8,000-word piece on Codex, Shipper says, can be reshaped from its companion database into an article, a podcast or a video, depending on what a subscriber wants. The annual subscription costs $288 — and explicitly includes the agent. "I may not want to read all of your stuff every month," Shipper says. But if the subscription covered his agent, "I may want my agent to know it."
Whether any of this pays is unsettled. Rasmus Kleis Nielsen, a professor of communication at the University of Copenhagen, calls the market for publishing to agents "emerging" but immature and "extremely uneven", with open questions about price points and how agents will interact with content. Visibility inside model answers, he notes, may work like reach on TikTok or Instagram — channels that also rarely send audiences back to a publisher's own site. Murphy prefers a calmer reading: this is not the first time media has followed its audience onto a new channel, after social, short video and podcasts. "It isn't really AI as the audience," she says. "It's AI as a tool that helps us reach our audience."
The surrounding terrain is contested. Many publishers, including WIRED's parent Condé Nast, have signed deals letting AI platforms surface their work. Newsrooms chase "generative engine optimisation" so their articles appear in model answers, and the WordPress plugin AIOSEO has been shipping LLM.txt files since last year to help machines interpret a site. Others litigate: the New York Times sued OpenAI and Microsoft over copyright in 2023, and the case is still running.
What is genuinely new is the direction of travel. Search engine optimisation taught publishers to write for an algorithm that then delivered them to humans. Feeds for agents cut out that step: the publisher writes for something that curates for a human, and hopes the brand survives the answer. Whether a subscription that includes the agent is worth more than one for the human alone is the open question underneath all of it — and it decides whether "the agent as audience" becomes a business model or just a new kind of reach that pays nothing.
