Industry news roundup #5
The monetisation problem for the agentic web
The dominant web business model so far has been the following: users receive free information/entertainment/services by paying with their attention and personal data; content publishers earn cash by selling real estate for ads; advertisers gain eyeballs, awareness and clicks, paying in media dollars; Google and the ad-tech ecosystem take a cut of every impression by matching buyers and sellers.
This model has reached an incredible scale. Every HTTP request that lands on a page with an ad slot triggers an automated real-time bidding (RTB) auction. Industry estimates put RTB volume at 1.4 trillion ad impressions per day. In other words, the web already clears trillions of daily mini-transactions that we call “ad auctions.”
At Build 2025, during the keynote, Kevin Scott introduced the concept of the Open Agentic Web, based on MCP and an HTML equivalent called NLWeb.
The idea of the agentic web is that LLM-powered agents will call endpoints on user’s behalf. The problem is that an agent does not watch pre-rolls or click banners: enter the agents, and exit the eyeballs. If the entry point is a chat interface that instructs the agent to perform a task, direct human interaction with content disappears, and the ad market that funded millions of publishers is on shaky ground.
How can this play out?
One option is that major content and service providers survive, while the long tail starves. Sites such as TripAdvisor, O’Reilly, or Shopify stores will expose NLWeb endpoints so that agents can comparison-shop and transact. Premium tier-1 publishers such as the NYT, the WSJ, or the FT are in a position to strike wholesale licensing or revenue-share deals with an OpenAI/Google/Microsoft agenting applications. The rest of the web (hobby blogs, niche newsletters, indie sites) will lose the only monetisation path they ever had and quietly disappear. This is the less open and less interesting web that Kevin Scott described would have arisen if the early web had been dictated by a vertically integrated player such as a browser company.
There is an alternative scenario that keeps the rich and diverse open web intact, although it requires getting a key architectural component right, something that has been “the future” for a long time, but has never happened: micropayments.
In the agentic web, every HTTP request could become a payment transaction between a user (through an agent) and a publisher or service provider. Based on Cloudflare estimates and extrapolating to the entire web, the number of HTTP requests is about 20 trillion per day. That is a huge number, but it is not much larger than the number of RTB ad auctions that happen every day. Also, not every HTTP request would be a direct content hit that demands a payment, so in terms of volume RTB ad auctions are in the same ballpark.
However, there are many differences between the structure of the ad-auction marketplace and a hypothetical web of micropayments between users and content providers.
First, the trillions of RTB auctions executed every day are entries in a ledger, not real-time financial transactions. The RTB system selects the winning bid and logs the clearing price and relevant metadata (impression ID, buyer ID, seller ID, and so on), but no money changes hands immediately. Billing happens later, usually on a monthly basis, when advertisers and publishers settle their invoices based on the impressions delivered.
Second, programmatic advertising is a many-to-few-to-many hub-and-spoke system. The ecosystem consists of billions of users and millions of advertisers and publishers, but they do not interact directly with one another. Advertiser demand is aggregated by Demand-Side Platforms (DSPs), and the supply of ad space is aggregated by Supply-Side Platforms (SSPs), which also serve as ad exchanges. This market is very concentrated: Jounce Media estimated that 6 DSPs controlled 82% of all ad spend in 2024, whereas 15 SSPs drove 90% of publisher revenue.
In summary, millions of edge participants are funneled through perhaps a hundred core trading desks (DSP + SSP), and those trillions of daily RTB auctions resemble database entries more than financial transactions.
DSPs and SSPs not only aggregate demand and supply, but they also make the lives of advertisers and publishers easier. Advertisers need to set budgets, targeting rules, and frequency caps, while publishers need inventory optimisation, ad-quality controls, and fraud protection. All of this is provided by DSPs and SSPs. Meanwhile, the user is barely aware that they are the target of fierce competition for their eyeballs.
Returning to the hypothetical web of micropayments between users and content providers in the agentic web: the unintermediated version would require trillions of daily financial transactions to be settled within a 50ms latency window, between millions of market participants on a one-to-one basis (that is, agents acting on behalf of their owners).
These agents would need to be configured by users, for example by setting up a wallet with a monthly allowance or a per-request cap. Users cannot remain passive, lazy, unaware collateral victims of the ad bidding war, and they would have another chore to manage. I do not think this is a very realistic scenario from technical, financial, or UX point of view.
Maybe DSPs and SSPs will reinvent themselves, providing aggregation and clearance services to users instead of advertisers. They will solve the inherent problem of micropayments: they are too micro to be worth being processed. Maybe, in the future, we will receive a monthly bill from Google DV360 for all the content and services that our personal agent purchased on the agentic web. And if the bill is unexpectedly high, no Fortnite for a week for the agent. Because the alternative is much more grim: it could be the end of the rich and diverse open web that we are used to.
Do you like this post? Of course you do. Share it on Twitter/X, LinkedIn and HackerNews