Five weeks ago, in The Missing Seat, I noted that D/Cipher was absent from every agentic advertising standard — AAMP, AdCP, Agent Registry, PubMatic AgenticOS, Magnite's seller agent. The absence was concerning but abstract. Standards are governance documents. You can join a working group later.
That changed in the last five weeks. The industry didn't just write standards. It built runtimes.
What Happened
Three SSPs shipped containerized partner execution environments — infrastructure where someone else's intelligence runs inside the auction, at the point of decision, on signals the partner could never access externally.
| Runtime | Launched | Protocol | Partners Running Inside | Latency |
|---|---|---|---|---|
| Index Cloud | Apr 21 | ARTF | Bedrock (full DSP bidder), InterMedia, Navigator | <5ms |
| PubMatic Decision Fabric | Jun 1 | ARTF | inPowered AI, MiQ, Chalice AI, SWYM.AI | ~1ms |
| Magnite Seller Agent | Jan 6 | AdCP / MCP | MiQ, Warner Bros. Discovery, LG Ad Solutions | Agent-mediated |
Adform and Adsquare implemented ARTF on the DSP side. Kochava open-sourced StationOne with the AAMP Buyer Agent SDK. IAB Tech Lab released AAMP 2.0 for programmatic. AdCP hit 92 members.
This isn't pilots. PubMatic processes 2.7 trillion bids per day across 100,000+ properties. Index Exchange runs the first containerized DSP bidder in production. The runtimes are live.
Why Runtimes Are Different from Standards
A standard tells you how to talk. A runtime tells you where you live.
D/Cipher's intelligence today runs in its own infrastructure. An advertiser buys a D/Cipher segment via a deal ID. The segment was pre-computed — OpenAI analyzed the content, mapped it to the 100-million-connection taxonomy, scored the intent. By the time the bid request fires, D/Cipher's work is done. The intelligence travels as metadata attached to inventory.
Containerized runtimes invert this. The partner's model — not its metadata — deploys inside the SSP. It sees impression-level signals the partner could never access externally: real-time user context, full auction dynamics, inventory characteristics that don't survive the oRTB pipeline. The model runs, decides, returns — all before the bid request leaves the building.
Intelligence is upstream. SSP sees a label, not a model.
Intelligence is inside. SSP provides compute. Partner provides logic.
The difference matters because D/Cipher's value proposition — we understand what content means better than anyone — is exactly the kind of intelligence that improves when it runs closer to the decision. A D/Cipher container inside an SSP could score content and see auction dynamics and match against buyer intent in the same execution frame. Pre-computed deal IDs can't do that.
Four Paths
Jon Roberts, People Inc's Chief Innovation Officer, told Digiday there's "a trust gap to be solved before this really gets deployed." Fair. But the infrastructure is deploying regardless. The question for People Inc's engineering org is which deployment model to build for.
Path 1: Containerize
Package D/Cipher's scoring model as an ARTF-compliant container. Deploy into PubMatic Decision Fabric, Index Cloud, and whatever Magnite builds next. D/Cipher's intelligence runs at the point of decision on signals it's never seen before.
Engineering: Model serialization, latency optimization to <5ms, container security, API extraction from monolith. Risk: Signal leakage — the SSP now hosts your scoring logic. Commoditization pressure — if the model runs everywhere, what's proprietary?
Path 2: Build Agents
Create D/Cipher buyer and seller agents on AAMP and AdCP. D/Cipher becomes a participant in agent-to-agent negotiation — translating campaign briefs into intent matches, surfacing inventory with contextual intelligence no competitor can replicate.
Engineering: Agent orchestration layer, MCP/A2A integration, LLM orchestration for brief interpretation, real-time inventory matching. Risk: D/Cipher's publisher scale (175M monthly) may be too small to attract agent integrations vs. SSPs processing trillions of bids.
Path 3: API Bridge
Register as an ARTF signal provider. SSP containers call D/Cipher APIs at auction time for contextual enrichment. D/Cipher intelligence stays home. Partners come to it.
Engineering: Sub-5ms API endpoint, massive concurrency (trillions of daily calls), edge deployment. Risk: External call adds latency. At auction speed, even 2ms is expensive. SSPs will prefer partners who deploy locally.
Path 4: Stay Direct
D/Cipher continues as a direct-sold and PMP-delivered product. D/Cipher+ extends to third-party inventory through managed service. No agentic infrastructure investment. Bet that premium relationships outperform programmatic at scale.
Engineering: Minimal — continue current roadmap. Risk: The programmatic market builds an intelligence layer without D/Cipher. When agencies deploy buyer agents, those agents negotiate with SSP seller agents. If D/Cipher isn't inside the negotiation, it doesn't exist to the buyer.
The Actual Decision
Path 4 is the default. It's also the most dangerous, because it feels like a choice but is really an absence of one. D/Cipher+ is growing — 50%+ of direct digital revenue, orders 50%+ larger. The direct-sold business works. But direct-sold is a human-mediated channel. When the buy side automates, the sell side that hasn't automated gets routed around.
Path 3 is the easiest engineering lift but the weakest competitive position. An external API call can't compete with in-auction inference on latency. Every millisecond is a lost auction.
Path 2 is the most ambitious and the hardest to justify with D/Cipher's current scale. You need massive inventory to attract agent integrations. PubMatic and Magnite have it. A single publisher — even one with 175M monthly uniques — doesn't.
Path 1 is the one I'd recommend.
Not because it's risk-free — the signal leakage concern is real. But because containerization lets D/Cipher's intelligence compete where it's strongest: at the moment a machine evaluates what a piece of content means. Every other contextual provider (GumGum, Seedtag, IAS) pre-computes and ships labels. If D/Cipher can ship models — scoring logic that runs live, on impression-level signals, inside the auction — that's a category distinction, not a feature.
ARTF containers are cryptographically signed by the partner and verified by the host. The SSP provides compute; it doesn't own the logic. Index Cloud's architecture explicitly prevents the host from inspecting container internals. The signal leakage risk is real but architecturally bounded.
Roberts is right that there's a trust gap. But the gap is narrowing on a timeline that doesn't wait for People Inc. PubMatic has 1,000 AI-powered deals running. Bedrock's entire bidder lives inside Index Exchange. When the gap closes, the partners who are already inside the runtime will have data, relationships, and operational experience that late entrants can't replicate.
The engineering priority: extract D/Cipher's contextual scoring model from its current monolithic deployment. Serialize it. Make it container-portable. Build toward ARTF compliance. Then decide which SSP to deploy into first. PubMatic's GPU-accelerated inference servers are the obvious starting point — they're built for exactly this kind of intelligence.
D/Cipher's intelligence is the best in publishing. That's established. The question is no longer whether the intelligence is good enough. It's whether the intelligence is portable enough. The industry just built a new place for smart things to run. The smartest thing in the room is still at home.