By Matt Griffiths, SVP of Technology, Audigent, a part of Experian
Over the past year, programmatic’s center of gravity has shifted from cookie deprecation workarounds to a new promise: agentic AI that can plan, activate and optimize media with minimal human intervention.
But most of what’s currently labeled “agentic” is workflow automation. Large language models translate briefs into structured inputs. APIs move instructions between systems. Emerging standards like AdCP aim to formalize how agents communicate with platforms.
That’s orchestration.
Autonomy, particularly at the auction level, is something else entirely.
In real-time bidding, decisions are made in milliseconds. The agent can only evaluate what it can see in that moment. And today, much of the industry’s signal arrives too late, travels through too many hops or gets diluted along the way.
If agentic AI is going to meaningfully change programmatic performance, the signal layer has to change with it.
The Auction Is Still The Bottleneck
Agentic systems compress planning and activation cycles from weeks to minutes. But the auction has not changed. It still operates under strict latency constraints, and the value of data decays rapidly inside that window.
When enrichment happens upstream or after the fact, agents optimize against partial information. They rely on proxies instead of outcomes.
Consider a retail advertiser optimizing toward in-store sales. If contextual meaning, behavioral intent or first-party intelligence is appended outside the bid window, the agent evaluates a simplified version of value. Move that enrichment closer to the auction itself, and the system can price the impression based on richer, real-time signal instead of historical approximations.
Speed without proximity to signal just accelerates the wrong decisions.
Signal Quality Determines Autonomy
Agentic buying is often framed as a modeling problem. In practice, it’s a signal problem.
Models require inputs that are timely, interpretable and aligned to real outcomes. Every additional handoff in the enrichment chain increases the risk of latency, mismatch or signal loss. By the time data reaches the decision point, its fidelity may already be compromised.
Real-time data enrichment moves contextual and first-party intelligence closer to the moment of transaction. Page-level meaning, live behavioral indicators and environmental factors like weather or seasonality can be evaluated within the auction rather than bolted on later.
That doesn’t just make the pipes faster. It upgrades what flows through them.
If the goal of agentic systems is auction-time autonomy, then signal quality at the moment of bidding becomes the gating factor.
Sovereignty Is Now A Performance Variable
As activation becomes more automated, first-party data becomes more valuable and more exposed.
Many legacy workflows still require brands to pass data through multiple intermediaries before it becomes actionable. Each hop introduces governance complexity and increases risk. In a world where AI agents can trigger activation instantly, brands will not scale adoption unless they trust the chain of custody.
Data sovereignty is often discussed as a compliance issue. Increasingly, it’s a performance issue. If brands restrict activation because of exposure concerns, agentic systems simply won’t have the inputs they need.
Reducing transfers and tightening control over enrichment isn’t just about risk mitigation. It’s about unlocking participation in automated buying at scale.
Standardizing Agents Isn’t Enough
The industry is making progress on standardizing how agents communicate with platforms. That’s an important step.
But protocol standardization alone doesn’t solve the core issue: agents are only as effective as the signals they can access in real time.
Without modernizing the enrichment layer, agentic programmatic risks becoming a faster orchestration engine optimizing on yesterday’s data.
The next phase of programmatic competition won’t be decided by who builds the most elegant AI interface. It will be decided by who controls and delivers the highest-quality signal at the moment of auction.
Agentic AI will change how media is executed. Whether it materially improves outcomes depends on something less glamorous but more fundamental: the infrastructure that feeds it.
If the signal layer remains fragmented and delayed, autonomy will stall. If enrichment moves closer to the transaction, agents can finally optimize against real value instead of proxies.
The models are ready. The auction mechanics are not.
