By IAB Tech Lab named its agentic advertising umbrella initiative AAMP on February 26. Four weeks later, Kochava shipped the first place where those protocols actually run.
On March 24, Kochava announced that its StationOne platform, now exiting closed beta after launching with invitation codes in November 2025, hosts the AAMP Buyer Agent SDK as an open-source workspace. StationOne is a desktop application, not a cloud service. It connects AI models to ad tech tools through API keys and Model Context Protocol (MCP) integrations. Every agent action, every system pre-prompt, and every skill definition sits in viewable source code.
For measurement and attribution teams, this is the first concrete artifact to evaluate. Not a roadmap slide. Not a blog post about the future of AI in advertising. A downloadable workspace with 19 agent skills you can read line by line.
What the spec actually contains
The AAMP workspace inside StationOne organizes 19 skills across eight functional areas that track the campaign lifecycle from setup to reporting. Each skill routes through IAB Tech Lab's reference implementation MCP Server via the StationOne Connector.
The eight areas, per PPC Land's detailed breakdown:
- Foundation and setup: Organization management, account linking, tiered pricing navigation across four levels (Public, Seat, Agency, Advertiser) with discounts up to 15%
- Planning and strategy: Campaign brief creation via OpenDirect protocol, audience planning that balances identity-based targeting against contextual targeting, budget allocation across channels
- Inventory discovery and deal structuring: Natural language inventory search, comparison analysis, deal structuring for Programmatic Guaranteed, Preferred Deal, and Private Auction
- Campaign execution: Brief-to-booking automation through OpenDirect, order creation, line item building with impression quantity calculation from budget and CPM inputs
- Creative management: Asset registration with approval workflows, IQG rating tracking, rotation weight configuration for A/B testing
- DSP activation: Deal ID extraction from programmatic direct bookings with platform-specific activation steps for The Trade Desk, DV360, and Amazon DSP
- Performance and optimization: Delivery pacing and budget utilization monitoring
- Reporting and compliance: Campaign reporting and compliance verification
That is a full buy-side workflow, from org setup to post-campaign reporting, expressed as agent-executable skills. The Agentic Audience Planner skill is the one measurement teams should look at first: it estimates coverage percentages across inventory and explicitly balances identity signals against contextual signals. That balance, and the match rates it produces in practice, will determine whether agent-driven audience planning actually works or just looks good in a demo.
The measurement stack question nobody is asking yet
Here is what concerns me. AAMP covers campaign execution thoroughly. The planning skills reference audience signals. The DSP Activation skill bridges OpenDirect deal structuring to buy-side platforms. But the gap between "agent executes a campaign" and "agent measures what that campaign did" remains underspecified.
Kochava is a measurement company. StationOne runs on your machine, not in their cloud. That architecture choice matters for attribution: if the orchestration layer is local, how does it ingest conversion data? Through Kochava's existing postback infrastructure? Through a new MCP-based measurement connector? The press materials describe "agentic automation for campaign, measurement, and compliance workflows," according to Kochava's announcement, but the 19 published skills lean heavily toward the buying side.
Meanwhile, IAB Tech Lab's broader agentic portfolio includes two other specifications worth tracking. The Agentic Real Time Framework (ARTF), released for public comment in November 2025, defines how agent services operate within host platforms and handle bidstream processing through containerized deployments. And Agentic Audiences (formerly UCP, or Unified Context Protocol), also at v1.0 since November 2025, standardizes how agents exchange identity, contextual, and reinforcement signals representing real-time consumer intent.
AAMP, ARTF, and Agentic Audiences together form a three-layer stack: AAMP handles managed campaign workflows, ARTF handles real-time bidding delegation, and Agentic Audiences handles the signal exchange between them. The measurement connective tissue, how an agent closes the loop from impression to outcome, is the piece that still needs building.
Why the open-source choice matters for trust
Charles Manning, Kochava's CEO, told Digiday that StationOne is an "integrative AI hub" and a "Slack-like desktop interface" where users plug in their own model APIs and connect them via MCP. He described it as an "orchestration layer" for connecting AI models and ad tech tools.
The open-source decision is the right one for an industry that has spent years arguing about whether intermediaries are skimming margin. When every agent skill, every system prompt, and every decision pathway is readable source code, the black box argument gets harder to make. Anthony Katsur, IAB Tech Lab's CEO, framed it directly: "AAMP is built to support AI-driven workflows on open, trusted standards. Making this workspace open-source within StationOne opens the door for broader industry participation and continued development."
But open source is table stakes. The real test is interoperability. Right now, StationOne is the only runtime for AAMP's Buyer Agent SDK. Multiple competing approaches to agentic advertising are already circulating, from various vendor-led frameworks to other protocol proposals. As Digiday noted, without convergence, the industry risks replaying earlier interoperability failures, this time compounded by LLM-driven workflows.
What to do Monday morning
If you run campaigns, download StationOne and read the 19 skill definitions. Not to deploy them. To understand what IAB Tech Lab thinks the agent-driven campaign lifecycle looks like, because that is the taxonomy your tools will eventually conform to or compete against.
If you run measurement, pay attention to the Agentic Audiences spec (formerly UCP). That is where identity resolution and signal exchange get standardized for agents. The match rates and coverage percentages that spec produces in real implementations will determine whether agent-driven buying actually outperforms the spreadsheet-and-Slack workflows it is trying to replace.
If you run ad ops, the DSP Activation skill, which handles Deal ID extraction and provides platform-specific trafficking steps for TTD, DV360, and Amazon DSP, is the one to watch. That is the skill most likely to affect your daily workflow first.
No one is running live transactions through AAMP yet. This is a testing environment. But the spec is real, the code is readable, and the campaign lifecycle it describes is comprehensive enough that ignoring it would be a mistake.
Devon Park covers ad tech measurement and privacy for The Daily Vibe.
