Build automated agents using optimized AI Foundry models on Snapdragon (DEMSP380)

Meghana Sreenivasa Rao and Darren Oberst present a Microsoft Build 2026 session on building local, enterprise-ready agentic workflows on Snapdragon X Series PCs using GenAI models optimized for the Qualcomm Hexagon NPU.

Overview

The session focuses on building performant and secure local agents at the edge, customized for enterprise use cases. It highlights how using models optimized for the Qualcomm Hexagon NPU architecture enables true NPU acceleration for agentic workflows that can be automated and scheduled.

What the session covers

Challenges with complex, repetitive workflows

• The presenters describe enterprise workflows that are complex and repetitive, and therefore good candidates for automation and scheduling.

Enterprise workflow example: Jira across data silos

• The session uses Jira as an example of an enterprise workflow.
• It discusses integrating across multiple data silos as part of the workflow.

Summarizing Jira activity

• The presenters demonstrate running summarization over Jira activities as part of the automated workflow.

Moving from prototype to production

• The session calls out challenges when transitioning an agentic prototype into a production-ready solution.

Backend service and API exposure on Windows

• The presenters introduce a backend Windows service.
• They describe exposing an API to run the workflow and retrieve results.

Retrieving outputs

• The workflow produces outputs that can be retrieved after completion.
• Outputs are packaged as a ZIP file.

Q&A: Snapdragon advantages for AI

• The Q&A segment discusses advantages of Snapdragon processors for AI tasks.

Using multiple models in one use case

• The session discusses mixing and matching multiple models for different stages of a single use case.

Session metadata

• Event: Microsoft Build 2026
• Session code: DEMSP380
• Level: Intermediate
• Topic area: Agents & apps
• More sessions: https://build.microsoft.com