First Impressions of the Advertising Context Protocol (AdCP) (dennisy.me)

AdCP is a task-first, adtech-specific semantic layer built on top of general agent protocols (MCP for tools, A2A for agent-to-agent messaging). The spec standardizes a small, fixed set of advertising tasks (about nine media-buy actions) with JSON schemas — for example create_media_buy — so compliant agents and platforms expose the same tool names, payloads and workflow ordering. That promises easier interoperability, publisher-led workflows and direct activation across sellers: a buyer agent that can launch a campaign with Seller A should be able to do the same with Seller B without bespoke adapters. The trade-offs are significant for the AI/ML community. By baking high-level domain semantics into the protocol, AdCP reduces the need for runtime discovery and free-form agent planning — agents become form-fillers for large, predefined commands rather than autonomous planners discovering APIs. That constrains agent expressivity and may collapse “agentic” flows back into familiar REST-like patterns. The spec hints at new roles (an Orchestrator, DSPs becoming strategy/optimization layers) and uses a signals protocol for audience buys, but it doesn’t directly address ecosystem-wide problems (open web vs walled gardens) and largely reuses existing trust constructs (e.g., ads.txt). Adoption and whether this truly enables agentic automation versus standardization remain open questions.