Build and deploy AI at the edge for real-world impact | OD837

Cosmos Darwin and Roycey Cheeran present a Microsoft Build 2026 session on deploying AI to edge devices using Azure Local on small form factor hardware with Foundry Local built in, focusing on low-latency, offline-capable “physical AI” scenarios like robotics and factory-floor automation.

Overview

What the session is about

The session explains what changes when AI moves from the datacenter to the physical world (“physical AI”), where latency, intermittent connectivity, and real-time interaction with people and machines make cloud round-trips impractical.

Core technologies highlighted

• Azure Local (public preview on small form factor devices)
  • Positioned as a way to bring cloud-consistent infrastructure to compact edge hardware.
• Foundry Local (built in)
  • Used to run AI models locally so systems can keep working when connectivity is limited and to reduce latency.
• Azure Adaptive Cloud
  • Framed as the broader set of technologies enabling edge + cloud consistency.
• IoT Operations
  • Mentioned as part of the Adaptive Cloud stack used in the solution approach.

Scenario and demo described

• The session includes a demo of a basic agentic robot performing pick-and-place actions.
• The robot is controlled via natural language, and the presenters discuss the agentic behavior and how it adapts.

Pipeline components discussed

The presenters describe an end-to-end local AI pipeline that includes:

• Vision
• Speech
• Reasoning

Session resources

• https://aka.ms/build26-next-steps
• https://aka.ms/build/foundrydiscord

Speakers

• Cosmos Darwin
• Roycey Cheeran

Session context

• Microsoft Build 2026 on-demand session (OD837)
• Track/theme: Agents & apps
• Language: English (US)

Chapters (from the video description)

• 0:00 - Introduction and session overview by Cosmos Darwin
• 00:00:39 - Framing the concept and opportunity of Physical AI
• 00:03:04 - Exploring work domains where AI has yet to transform physical tasks
• 00:04:05 - Emergence of Agentic Robots enabled by local AI models
• 00:04:40 - Demo: Basic agentic robot performing pick-and-place actions via natural language
• 00:06:05 - Analysis of agentic behavior and adaptability in robot demo
• 00:07:18 - Technical walkthrough: Hardware and software setup overview
• 00:09:00 - Detailed explanation of AI models, vision, speech, and reasoning pipelines
• 00:13:01 - Use of Azure Adaptive Cloud technologies: infrastructure, Foundry Local, and IoT Operations
• 00:21:27 - Conclusion and call to action: Try Azure Adaptive Cloud features for physical AI