The question is no longer if agents will pay publishers
Amazon, Coinbase, Stripe, and OpenAI all said the same thing this month, in different ways. Here's the question that matters now.
A year ago, asking whether AI agents would pay publishers was a fringe question. OpenAI and Amazon both sent strong signals to answer this question.
Sam Altman went on Nick Thompson’s podcast at The Atlantic and said AI agents should pay publishers.
“I can give you my best theory, and I’ll caveat this by: no one knows. What I hope will happen, and what I’ve wanted to happen for a long time - but really makes sense in a world of agents - is that we try a sort of micropayment‑based approach.
So if my agent wants to come read Nick Thompson’s article, Nick Thompson or The Atlantic can set a price for the agent to read it. I might be different than a human reading it: my agent can read it, pay 17 cents, and give me a summary of that. If I want to go read the whole article, I can pay a dollar right then, or whatever—however that works.”
If you’re a regular reader, you know how crucial I think it is for publishers to explore the protocols themselves. And learn how they can surface the demand data behind every read. Which questions agents are asking. Which parts of the content surface. Which gaps competitors can’t fill. The signal in the traffic, not just the cents.
But when selling to agents is not your core business, you might feel it’s a bit far-fetched. This is where the next signal is interesting.
Amazon shipped rails
On May 7, AWS announced Bedrock AgentCore Payments (in preview). It’s a managed, end-to-end payment system purpose-built for AI agents. Built with Coinbase (running the x402 protocol and stablecoin rails) and Stripe (handling fiat funding via Privy).
You tell your agent how much they can spend: “To get started, developers connect their agent to a wallet or payment service provider, register a funded payment source, and set spending limits per session.”
Read the announcement and the use case they reach for first is striking:
“A financial research agent can dynamically access real-time market data feeds and paywalled publications, paying for the articles and data points it uses on behalf of the end user.”
Yes, articles.
Where the question moves
So the question moves. It’s no longer will agents pay publishers?, but what kind of publisher actually benefits when the rail goes live?
We can approach this question purely from a revenue perspective, but I think it’s broader than this. Our core thesis at Mizal is that the current information value chain breaks in an agentic world because of three things:
Lost value: when AI systems use creator knowledge, there is often no reliable way to attribute, price, license, or compensate that usage.
Invisible demand: creators do not know what agents and users are asking for, where demand is emerging, or which parts of their archive could answer it.
Flattened meaning: generic AI strips away the creator’s judgment, confidence, reasoning, tone, context, and distinctive perspective.
If we treat agentic monetization as a one-off payment — opt in, accept the per-read rate, move on —, we will get something like a syndication-agency relationship from 2008. The check comes in, the relationship ends, the content keeps being consumed but the direction of the product roadmap doesn’t change.
The publishers who treat it as a data layer will get something very different. Because every paid agent read is also a signal: which questions agents are asking. Which parts of the archive get surfaced. Which gaps competitors can’t fill. Which jurisdictions, audiences, and intents are most underserved.
CJR’s Tow Center published a very interesting piece on this exact tension based on the recent discussions we had in Perugia. The last paragraph:
“Compensation is just the first step. As the digital ecosystem rapidly evolves, publishers need to understand changing audience behavior and needs. Existing deals from AI companies may pay news organizations, but they don’t guarantee accurate representation or consistent attribution, nor do they always provide granular insight about how audiences engage. Securing greater control over these things can help publishers better serve their audiences while preserving their editorial priorities.”
The signal in the traffic, not just the cents. And that signal is where the next decade of editorial value and product strategy will come from.
That’s the layer worth building.
P.s.: I’ll be in Europe next week for the Nordic AI Journalism in Copenhagen and the WAN-IFRA World News Media Congress in Marseille. Come say hi if you’re around!
