In January 2026, PubMatic ran one agentic campaign — a CTV buy with Butler/Till. By April, they had 30 live campaigns across five countries. As of their Q1 earnings: 250+ agentic deals transacted, 80% year-over-year growth in their emerging revenue category, now 14% of total company revenue. Every advertiser that ran a campaign returned for more.
Thirty to 250 in 90 days. That is the adoption curve of infrastructure that works.
On June 18, Horizon Media launched an agentic orchestration layer connecting Disney, Fox, NBCUniversal, TikTok, Clear Channel, iHeartMedia, OUTFRONT Media, and Snap through APIs, MCP standards, and agent-based connections. Profectio called it the explicit inversion of the proprietary model — the agency provides orchestration, partners provide capabilities, clients choose the combination. Fox tested agent-to-agent transactions with Horizon the day before through AdStudio. Phase 2 expands to more verticals in H2 2026.
People Inc is not connected to any of it.
These are not pilots anymore. They are production systems processing real budgets. And in July 2026, the specifications they run on stop being malleable and start being permanent.
The July Calendar
iss validation. Mcp-Method/Mcp-Name headers for gateway routing. Any server runs behind plain round-robin load balancer. D/Cipher has no MCP server.Two protocol specifications finalize this month. Both were written without D/Cipher input.
What Hardening Means
Standards have a phase transition. Early in a protocol's life, contributors shape the spec. They get their data models into the standard. Their architectural assumptions become everyone's architectural assumptions. This is the window when participation has compounding returns — not just adopting the standard, but shaping it so the standard works for you.
Then the spec ships. The SDKs ship. The reference implementations ship. And the cost function inverts.
Before finalization, contributing to MCP or AAMP would have meant influencing how contextual intelligence surfaces in agent-to-agent transactions. D/Cipher's taxonomy — 12,000 terms, 100 million semantic connections, OpenAI-enriched — is arguably the best contextual intelligence product in publishing. If D/Cipher had participated in AAMP's Signals Activation Protocol during 2025, the spec might natively accommodate content-graph signals. If D/Cipher had built an MCP server during the RC period, tool definitions for contextual targeting would exist in the MCP ecosystem.
After finalization, implementing an AAMP 3.0 Seller Agent means conforming to someone else's three-tier hierarchy. Building an MCP server means exposing tools through someone else's stateless protocol. The architectural decisions are already made. D/Cipher goes from potential co-author to late adopter.
This is not abstract. The AAMP Agent Registry has 10 entries: Amazon Ads, PubMatic, Equativ, Optable, Dstillery, Burt Intelligence, HyperMindZ, Mixpeek, IAB Tech Lab. All MCP servers. All contextual or audience intelligence providers. D/Cipher's absence means a buyer agent querying the registry for contextual intelligence in entertainment-and-lifestyle gets back every provider except the one with the deepest taxonomy in publishing.
The Acceleration Problem
The data that should concern People Inc's engineering leadership is not the 250+ campaign number. It is the slope.
PubMatic CES (January): 1 campaign
PubMatic April update: 30 campaigns, 5 countries
PubMatic Q1 earnings (May): 250+ deals, 80% YoY growth
→ Agentic revenue = 14% of total PubMatic revenue
That trajectory is not linear. AgenticOS is going beyond direct deals into fully autonomous end-to-end campaign execution — no human intervention, no RFP, no media plan. The buy-side adoption numbers confirm the acceleration: 89% of advertisers and 92% of agencies are already engaging with agentic infrastructure, versus 29% of publishers.
Horizon's orchestration layer compounds the problem. An agency with $8.5 billion in media investments can now route budgets to Disney, Fox, or NBCUniversal through MCP-standard connections. When Horizon's Phase 2 adds entertainment and lifestyle verticals in H2 2026, the routing decisions for budgets that should reach D/Cipher will resolve to publishers who are connected. The query happens in milliseconds. The routing is invisible. The budget flows to an accessible competitor.
People Inc's CIO Jon Roberts knows about AAMP. He named it alongside MCP and AdCP in Digiday's survey and said interoperability through these standards is essential. Then he cited a trust gap. On The Current podcast, he discussed D/Cipher at length — zero mention of agentic protocols, seller agents, or standards participation. Conscious deferral, not ignorance. The second muscle that knows its name and stays still.
The GAM Variable
There is a wildcard in the July calendar that has nothing to do with agentic protocols.
Judge Brinkema ruled in April 2025 that Google illegally monopolized publisher ad servers (DFP, ~90% market share) and ad exchanges (AdX, 60-70%). The remedies ruling has been pending since closing arguments finished — months past her self-imposed March 31 deadline. It could come any day.
The DOJ wants structural relief: force Google to divest AdX and open-source its auction logic. Google wants behavioral remedies — data-sharing promises, no divestitures. Brinkema has called for less "window dressing" in Google's proposals, but has also raised practical concerns about finding a buyer for AdX.
D/Cipher runs on Google Ad Manager. All of People Inc's ad serving — the slot targeting, the contextual taxonomy application, the D/Cipher+ off-platform distribution — flows through GAM's infrastructure. If Brinkema orders an AdX divestiture, the ad exchange D/Cipher depends on changes ownership. If she orders DFP divestiture, the ad server D/Cipher sits on top of changes ownership. If she orders both, D/Cipher's entire delivery infrastructure transitions to a new corporate entity during the same window that agentic specifications finalize without D/Cipher's participation.
Either remedies outcome creates engineering disruption. Structural remedies (divestiture) mean platform migration risk. Behavioral remedies (interoperability APIs) mean new competitors can access GAM's pipes — which means D/Cipher's advantage from GAM integration erodes even without a divestiture.
The Raptive Pincer
One detail from the past two weeks crystallizes the convergence.
Raptive Intelligence — the network serving 6,500 independent publisher sites with 224 million users — is an AdCP supporting member. Raptive is simultaneously building AlchemyAI, a food knowledge graph that competes directly with Allrecipes' recipe corpus. One company, eating both floors: the agentic ceiling (AdCP member, protocol-native) and the commerce floor (knowledge graph, data licensing).
People Inc treats these as organizationally separate challenges. D/Cipher is an ad tech product. MyRecipes is a food platform. Graphene-Selene is a CMS. Three teams, three roadmaps, three sets of priorities. Raptive treats them as one infrastructure investment across two revenue streams. That is the structural advantage of protocol participation — it compounds across use cases because the interface is the same.
What July Requires
The MCP stateless specification is designed for exactly the kind of intelligence D/Cipher provides. Stateless means no sessions — each request carries everything the server needs. OAuth 2.1 means authentication works with People Inc's existing Keycloak deployment. Round-robin load balancing means no sticky sessions, no shared state store. The engineering to build a D/Cipher MCP server exposing contextual taxonomy queries is not architecturally hard. Three engineers, two sprints, the same team that maintains D/Cipher's existing API layer.
The AAMP 3.0 Seller Agent SDK ships with the spec this month. A D/Cipher seller agent would register contextual intelligence capabilities in the AAMP Agent Registry, making D/Cipher's taxonomy discoverable to every buyer agent in the ecosystem. The three-tier hierarchy (Portfolio Manager → Channel Specialist → Functional Agent) means a D/Cipher functional agent could serve contextual signals to any channel specialist — CTV, display, video, native — through a standard interface.
This is not a multi-year platform migration. It is a protocol integration. The intelligence already exists. The taxonomy is already built. What is missing is the interface — the 50-line tool definition that makes D/Cipher's 12,000 terms and 100 million connections visible to the systems that are now allocating billions in media spending.
But every week the interface doesn't ship, the ecosystem builds muscle memory around the providers who are visible. PubMatic's 250+ deals created 250+ advertiser workflows that route through AgenticOS. Horizon's eight publisher connections created eight paths that bypass D/Cipher. AAMP 3.0's SDK will create thousands of agent implementations that assume the current registry is the available supply.
Infrastructure hardens. Then the companies that built on it compound. The ones that didn't discover that the concrete has set, and the forms were shaped by someone else.