The new agentic digital economy

The internet's two hemispheres

Over the past year the web has quietly split into two hemispheres: AI-blocked and AI-open. Platforms that monetize primarily through advertising are erecting barriers to AI crawlers and agent traffic, while sites that make money from subscriptions or purchases are leaning in because AI agents look like a fresh distribution channel. It's not just cultural resistance; it's economics.

If your revenue depends on humans seeing and clicking ads, non-human traffic threatens the value of your inventory.

If your revenue comes from content subscriptions or commerce, an agent that brings a paying reader or a buyer is welcome.

Three surfaces where AI agents touch the web

To make sense of the policy and money flows, it helps to separate how AI systems touch content and services.

First is deferred crawling, where models or providers collect data to build training sets and reference corpora.
Second is real-time retrieval, where an AI answers a live query by reading sources on demand.
Third is action-taking, where an AI acts on behalf of a user to get something done-writing a post, booking a flight, purchasing a tool set.

These three surfaces-train, read, and act-look similar from afar, but the incentives, permissions, and economics are different enough that they deserve distinct treatment.

Case 1: B2B licensing for training

Training access is settling into straightforward business-to-business licensing between AI platforms and content owners. Publishers and platforms that don't want to be mined for free data block training crawlers and negotiate rights instead.

This isn't a consumer product story; it's contracts, data scopes, indemnities, and usage caps. Expect more of it, not less. Blocking remains a bargaining chip, not the end state. The equilibrium is "pay to train, attribute, and don't exceed the licensed scope."

Case 2: Bundled access and rev share for real-time answers

Live retrieval sits closer to consumer value, so the monetization pattern shifts from B2B to B2B2C.

The likely shape is publisher bundles inside AI apps and answer-placement revenue shared back to sources. Instead of asking users to manage dozens of tiny paywalls, AI apps will package access to clusters of premium sources-news, research, niche trade publications-and settle the economics behind the scenes.

Alongside bundles, some AI search products are piloting revenue shares when a source materially powers an answer. The old search engine dynamic of "send traffic and sell ads" gives way to "surface content inside an answer and share revenue or pay a license."

The subscription model of media propagates into AI through these bundles, while rev-share programs cushion the impact of fewer outbound clicks.

Case 3: Commerce actions via MCP

Action-taking AI agents are the most sensitive to trust, fraud, and breakage, which is why the future here is not headless browsing but sanctioned tools. The pattern emerging is Model Context Protocol (MCP) servers that wrap a platform's APIs with clear scopes, rate limits, and delegated user consent, exposed in a form AI agents can call reliably.

Commerce is the natural early fit.

A merchant or platform provides discovery, cart, and checkout as MCP tools; an AI consumer app invokes those tools with the user's authorization; attribution and payments flow through familiar affiliate, CPA/CPL, or rev-share mechanics. The Shopify and OpenAI precedent shows how this can feel native to AI agents while still riding proven commerce rails under the hood.

For subscription products, the agent analogue to "family plans" appears as agent seats-add-ons that explicitly grant an assistant permission to read or act within an account.

Some of the eccomerce and subscription providers may feel tempted to block AI agents and provide their own AI-driven experiences, and that may work for the biggest players -like Amazon- but it will not work for the rest. The distribution power of the big Ai-consumer app providers -ChatGpt, Claude, Gemini, etc- is way too tempting.

The ARPU replacement problem

A useful way to forecast pricing is to ask what revenue a platform must replace when agent behavior displaces human scrolling.

For ad-funded feeds, the lost minutes and impressions have a calculable value. The recovery mechanisms are bundles for reading, revenue shares for answer placement, and paid, scoped access for actions.

For commerce, the value is clearer still: did the agent produce a sale, a qualified lead, or a repeat customer? If yes, attribution unlocks the familiar palette of rev-share and performance pricing.

Agent seats for subscription products then mirror human seat licensing, with policies that define which actions the agent may perform and how those actions are metered.

The governance stack that makes it all safe

The emerging agent economy needs recognizable guardrails. Identity and consent must be portable so an agent can prove who it acts for and what it's allowed to do.

Tokens and scopes need to be narrow and revocable. Payments and liability need clear handoffs, particularly in regulated flows like card-not-present checkout in regions with strong customer authentication.

MCP provides the agent-facing contract; underneath, platforms still run hardened APIs, fraud controls, and compliance checks. The important point is that the user experience is agent-native while the operational reality remains enterprise-grade.

As a technological society, we just started to glimpse the need of a governance stack that makes it all safe. We are not there yet, but we are getting closer.

The digital agent economy

Put together, the emerging monetization map is clear.

Training becomes B2B licensing.
Reading becomes bundled source access and answer-placement revenue sharing inside AI apps.
Acting becomes commerce and subscription transactions mediated by MCP tools that wrap platform APIs, with revenue shares for merchants and agent seats for subscription services.

It isn't a war between open and closed so much as a re-plumbing of value: don't train for free, do read with consent and compensation, and do act through sanctioned channels that measure and pay for outcomes.

Once you see AI agents as members of a team -and proxies of humans-, the spending patterns look familiar.

Organizations will buy content bundles so assistants can read what their analysts read. They'll purchase agent seats for the SaaS they already use. They'll enable MCP tools for their storefronts so assistants can bring qualified demand. Consumers will subscribe to source bundles in the same places they already pay for premium content, and they'll allow their AI agents to act where the outcomes are worth it.

In other words, AI agents will deliver value to users while also participating in a parallel economy with its own licenses, seats, and rev-shares. That economy is already taking shape; now it needs the discipline of good contracts, good tooling, and clear attribution to scale without friction.

What's next?

My question is, who is going to capitalise on the governance and economic transactionalities of the new agentic digital economy? Who is going to capisalise on this brand new layer of value?