Social Rocket
Carat and NBCUniversal are launching an AI-supported solution for buying and optimising premium video advertising. At the same time, Tinuiti is opening its cross-channel measurement data to AI assistants. Both announcements in early October 2026 show that agentic media buying is becoming concrete. But they also show: data access, recommendation and binding campaign booking are three different things.
Carat connects AI agents with the upfront business
On 6 October, Carat, NBCUniversal, FreeWheel and Newton announced a bespoke agent for planning, buying and optimising premium video inventory. It combines data and audience signals from Dentsu with NBCUniversal's advertising environment and FreeWheel's infrastructure.
A luxury retailer, not yet named, is using an individually configured agent for its upfront investment. Upfronts are agreements in which advertising inventory is bought in advance. The partners promise faster analyses and adjustments, while also stressing the role of human media experts. The announcement does not give concrete performance figures or the maximum budget changes the system may make independently.
NBCUniversal and FreeWheel have been working on this model for some time: In January 2026 they presented, with RPA and Newton, an approach for joint buying across linear TV and streaming. The new Carat use case builds on this development.
Tinuiti opens data instead of triggering bookings automatically
The agency Tinuiti introduced its Bliss Point MCP Server in early October. Via the open Model Context Protocol, compatible AI assistants can retrieve approved measurement data across channels. Named examples include ChatGPT, Claude, Microsoft Copilot and in-house enterprise systems. Scope and use depend on client rights and settings.
Possible tasks include campaign reporting, budget pacing and flags for unusual developments. According to Tinuiti, the cosmetics provider e.l.f. is an early test client. The product therefore primarily addresses data and analysis access – a different step from directly booking advertising inventory.
The real issue is approvals and accountability
At Advertising Week, representatives of several large agencies described different levels of control. According to Digiday of 8 October, WPP Media emphasised explicit approval thresholds and human accountability, especially when decisions cost money or change campaigns.
For a practical pilot, clear questions therefore matter more than product promises: Which data can an agent read? May it only make suggestions, or also change campaigns? From which amount is human approval required? How are actions documented and reviewed?
These points also belong in contracts. Our overview of AI clauses between agencies and advertisers covers data rights, tool disclosure and liability. Control over automated decisions is no longer a theoretical side issue.
What actually changes for the German market
Large agencies are already building the technical infrastructure. Stagwell has reorganised its media business around an agentic platform. Amazon is also testing AI for campaign decisions and running into transparency questions.
It is therefore plausible that agents will be able to take over a larger share of the groundwork in future. However, the public reports do not prove lower error rates or better advertising returns than conventional, human-supervised processes. For brands, the order should therefore be: first clean data and traceable rights, then supervised automation, and only then more autonomy based on measurable results.