Agentic media operations: Run at scale | BRK465

Ashok Kalidas, Don Hejna, and Kanio Dimitrov present live demonstrations of agentic advertising media workflows with Kantar. The session focuses on running image and video optimization loops at production scale.

Image and video optimization loops

The session demonstrates two complete optimization loops—one for images and one for video—built using Microsoft Agent Framework, Foundry, and Azure. Each follows the same sequence:

  1. Score the advertising media.
  2. Recommend improvements.
  3. Generate revised media.
  4. Re-evaluate the results.

Scalable graders power the evaluation stages, supporting advertising media optimization throughout the workflow. The demonstrations connect scoring and recommendations with generation and subsequent evaluation, rather than treating each activity as a separate task.

Production-scale performance

The session reports improvements in key component processing times from days to hours at production scale. This claim applies to key components, rather than a stated end-to-end processing time for the entire creative workflow.

Session details and resource

For further guidance, see the Agentic Business Applications next steps resource.