Google I/O 2026 named it: the Agentic Web. Chrome 149 ships a WebMCP origin trial. Prebid released an open-source seller agent. Microsoft is onboarding demand partners for its Publisher Content Marketplace. Magnite embedded agent capabilities in SpringServe. In two months, the infrastructure for agent-mediated content consumption went from white papers to production code.
People Inc sits at the center of three of these protocols. They co-designed one. Their crown jewel depends on another. And the third threatens to make both irrelevant.
Three Layers, Three Controllers
The agentic web isn't one standard. It's three protocol layers emerging simultaneously, each controlled by different parties:
| Layer | Function | Protocols | Controller |
|---|---|---|---|
| Discovery | How agents find content | WebMCP, llms.txt, AGENTS.md | Google + Microsoft (W3C) |
| Monetization | How publishers get paid | Microsoft PCM, Perplexity rev share | Microsoft + publishers |
| Commerce | How agents transact | AdCP, UCP, Prebid Sales Agent | Open consortium (AAO) |
Notice the familiar pattern. Discovery goes through the browser — which means through Google. Monetization goes through a platform provider — which means whoever runs the marketplace. Commerce is trying to be open — which means it needs adoption before one player captures it.
This is the session web's power structure, reconstructed for agents.
Where D/Cipher Breaks
D/Cipher works because users visit pages. Each pageview generates intent signals — what content the user reads, how long they stay, which terms match D/Cipher's 12,000-category taxonomy. Those signals feed real-time contextual targeting. Advertisers pay a premium because the targeting is accurate and privacy-compliant.
In the agent-mediated web, that loop doesn't exist.
An agent queries a WebMCP endpoint. It receives structured data. It synthesizes a response for the user. No pageview. No session. No dwell time. No taxonomy match. D/Cipher's signals evaporate.
This is the same threat that AI Overviews posed — extracted consumption without attribution — but worse. AI Overviews at least happened on a Google page with Google ads. Agent consumption happens in a browser sidebar, a chat window, or a third-party app. There's no page to put an ad on.
The Prebid Parallel
In 2015, publishers faced a similar lock-in: Google's ad server controlled access to demand. Prebid emerged as the open-source counter-play — a way for publishers to expose inventory to multiple buyers simultaneously, breaking Google's single-auction monopoly.
History is rhyming. Prebid has now taken over AdCP's sell-side agent code. The Prebid Sales Agent is an open-source tool that any publisher can deploy to expose their inventory to AI buyer agents. Magnite is doing the same thing proprietary — building seller agents into SpringServe.
The question for People Inc: does D/Cipher become a signal provider to these seller agents, or does it try to become one itself?
Four Engineering Decisions
1. Ship WebMCP endpoints (weeks, not months)
Expose Graphene-Selene's structured content through WebMCP's declarative API. Chrome 149's origin trial makes this testable now. Low cost, high optionality. The alternative — being invisible to agent browsers — is unacceptable.
2. Finish PCM integration (already co-designing)
People Inc co-designed PCM with Microsoft. They should be first to production. Yahoo is onboarding as demand partner. The usage-based licensing model maps cleanly onto content that Graphene-Selene already taxonomizes. Ship this in H2 2026.
3. Build D/Cipher as AdCP signal provider (the hard bet)
D/Cipher's taxonomy is valuable. The question is whether it can function without a session. Option: pre-compute intent embeddings per content unit and expose them via AdCP's Signals Activation Protocol. An AI buyer agent asks "what food-related inventory matches high purchase intent?" — D/Cipher answers with pre-scored content segments. This preserves the targeting value without requiring a live pageview.
4. Deploy the Prebid Sales Agent (open-source hedge)
Prebid's Sales Agent is available on GitHub. It's the header-bidding play for the agentic era: expose inventory to all buyer agents, don't lock into one demand path. For People Inc — which already doesn't use client-side Prebid (they use Exchange Bidding) — this is a chance to adopt open infrastructure before getting locked into Google's agentic ad stack.
The Architectural Tension
People Inc built D/Cipher for a world where users visit websites. The agentic web eliminates the visit. But it doesn't eliminate the need for what D/Cipher provides — contextual understanding of content, privacy-compliant targeting, premium signal quality.
The engineering challenge is decoupling D/Cipher's intelligence from its delivery mechanism. The 12,000-term taxonomy, the 100 million entity connections, the OpenAI semantic layer — none of that depends on a browser session. It depends on content existing. Content still exists in the agentic web. It just gets consumed differently.
The protocol stack is how publishers negotiate the terms of that consumption. People Inc is better positioned than most — they're co-designing the monetization layer, their content is already semantically structured through Graphene-Selene, and D/Cipher's taxonomy is the raw material that AdCP buyer agents will want.
The risk is the same one they faced with Google: letting someone else control the infrastructure layer while you provide the intelligence on top. Discovery goes through Chrome. Commerce goes through Prebid or Magnite. If People Inc doesn't act on layers 1 and 3, they'll find themselves in the same position — dependent on someone else's pipe — just with different names on the valves.
The session web's lesson: own your signals, rent your infrastructure. The agentic web's corollary: own your protocols, or become someone else's API.