AI agents fail to fix advertising fragmentation

In a Digiday sponsored column, Oz Etzioni, CEO and co-founder of Clinch, argues that simply adding more AI agents will not fix advertising's long-standing fragmentation. Per the column, AI can process engagement signals at scale and surface patterns that were previously invisible, but its impact is constrained when data and workflows are split across platforms.

Etzioni writes that agent-to-agent communication differs from true system-level coordination. The column says agents require access to shared data and consistent decision logic to align across creative, activation and measurement stages; when those inputs are spread across disconnected systems, agents end up optimizing in isolation and insights remain trapped in silos.

Editorial analysis: Companies and vendors promoting autonomous agents frequently emphasize agent interoperability, but industry reporting and practitioner experience show that interoperability alone does not create a single source of truth. Comparable transitions in martech and adtech have historically depended on data integration, canonical identity graphs, and consistent measurement schemas before automation could scale learning effectively.

Editorial analysis: For practitioners, the column underscores a common pattern - automation amplifies the capabilities of underlying infrastructure. In advertising, the payoff from models and agents is limited if pipelines do not deliver timely, reconciled signals about creative variants, audience exposure and conversion events. This matters for teams investing in MLOps, feature stores, and real-time telemetry for campaign analytics.

Analysts can monitor three observable signals: vendor claims about shared-data layers or unified activation platforms, improvements in end-to-end measurement latency, and evidence of learning compounding across campaign stages (for example, creative-level signals informing media optimization within the same measurement window). Observers should evaluate vendor architectures against those operational indicators rather than agent counts alone.