Skip to content

BRK238 Breakout Wed 18 Nov 10:15 PST

Frontier tuning: Optimize AI agents for real-world business outcomes

Kalyan Ayloo, Ranveer Chandra, malik Pradhan, Ben Ambrosino

Summary

From the recording / Microsoft AI summary

The session described tuning agents against both accuracy and cost: running evals, then 'hill climbing' on skills and harness, then choosing the best model and harness at each step.

Session description

Microsoft's catalogue description

AI success depends on more than the model. Learn how Frontier Tuning helps organizations optimize models, tools, skills, knowledge, and evaluation frameworks to improve performance on real business workflows. See how Copilot Studio and Foundry work together to deliver better outcomes, higher quality, and improved cost efficiency at scale.

Demo timestamps

Microsoft's player does not support links to a specific time, so open the recording and skip to the time shown.

  1. Skip to 00:17:44

    Frontier Tuning CLI and GitHub Copilot plugin: creating a learning environment

    Walkthrough of Frontier Tuning using the Frontier Tuning CLI and a GitHub Copilot plugin. The presenter asks it to create a marketing demo learning environment from a marketing operations template.

  2. Skip to 00:20:02

    The 'build content brief' skill in the learning environment

    Opens the 'build content brief' skill and talks through its instructions within the learning environment.

  3. Skip to 00:22:33

    Open interview in Cowork: generated prompt and questions

    'Open interview in Cowork' shows the kind of prompt the environment would generate and the questions it would ask.

  4. Skip to 00:25:04

    Packaged video: flow across canvases for an invoice review skill

    A packaged video follows the flow across canvases, with one persona building an invoice review skill and another acting as the user.

  5. Skip to 00:43:12

    Partner video: Frontier Tuning agent learning environment in an enterprise

    A partner video shows setting up a Frontier Tuning agent learning environment in a real enterprise context.

Watch on-demand (opens in a new tab)

Resources named in the session

← All sessions