No verifiable Microsoft AI announcements surfaced in the supplied research
The provided research notes and citations contain no primary-source items that meet the brief’s sourcing requirements. To avoid publishing unverified or speculative content, this edition includes no stories. If you share links to official Microsoft posts, product release notes (non-doc), or regulator filings from 2026-08-03 to 2026-08-09, I will produce a full digest.
Copilot changes this week
From Message Center and the public Microsoft 365 roadmap — admin actions first.
- Viva Learning: Copilot experience transitions to Learning Agent in Microsoft 365 Copilot
The Copilot experience in Viva Learning is being replaced; users selecting Copilot in Viva Learning are redirected to Learning Agent in Microsoft 365 Copilot, creating a more unified learning experience.
- Dynamics 365 Contact Center - Split recordings speaker-wise in closed conversation view (Preview)
Supervisors can now split conversation recordings by speaker in Dynamics 365 Contact Center, improving review by enabling direct navigation to customer or agent speech.
- Dynamics 365 Contact Center - Equip voice agents with custom voice
Dynamics 365 Contact Center adds public preview support for custom neural voice agents via integration with Azure AI Speech from August 21, 2026.
- Microsoft 365 Copilot: Unified management experience for connectors in the Microsoft 365 admin center
Microsoft is unifying Copilot connector and plugin management into a single experience in the Microsoft 365 admin center to simplify administration and discovery.
- Microsoft Teams: Interpreter agent support in Teams Rooms on Android
Interpreter agent brings real-time spoken language interpretation with voice simulation to Teams Rooms on Android for multilingual meetings, no Copilot license required.
Microsoft Learn: what changed
The most significant documentation changes this week.
- Use admin-connected models in cloud evaluations
This new documentation introduces support for using admin-connected models—models hosted behind enterprise AI gateways—in cloud evaluation scenarios within Microsoft Foundry. It details how to reference these models, the prerequisites, and the supported evaluation scenarios, marking a significant expansion in model integration capabilities.
- Understand model changes in GPT 5.1+
Page removed or relocated. The documentation explaining the behavioral and architectural changes from GPT 5.0 to GPT 5.1 has been removed. This leaves a gap for users and developers seeking guidance on adapting prompts, workflows, and agent instructions for the new model's intent-first, adaptive reasoning approach.
- Structured outputs
The documentation now details how to use structured outputs with the Azure OpenAI Responses API, including Python, .NET, and REST/cURL examples. This expands structured output support beyond Chat Completions and clarifies schema configuration for each API.
- Configure Microsoft Entra authentication for Foundry agent trace ingestion (preview)
This new documentation introduces support for configuring Microsoft Entra authentication for trace ingestion from Foundry agents to Application Insights, replacing key-based ingestion with identity-based access control. It details prerequisites, setup steps for both new and existing connections, and troubleshooting guidance, enabling more secure and manageable telemetry ingestion.
- Azure OpenAI reasoning models
The documentation introduces new controls and explanations for reasoning models in Azure OpenAI, including expanded 'reasoning_effort' levels, a new 'reasoning.mode' (standard/pro) for gpt-5.6 models, and the ability to preserve reasoning across conversation turns. These changes give developers more granular control over model behavior, cost, and continuity in multi-turn interactions.
- Quickstart: Optimize a prompt agent (preview)
This new page introduces the Agent Optimizer (preview) in Microsoft Foundry, providing step-by-step instructions for optimizing prompt agents using an interactive wizard. It details how to select agent versions, datasets, evaluators, and models, run optimizations, compare results, and promote improved agent versions, enabling users to systematically enhance prompt agent performance.
- How toolbox authentication works in Microsoft Foundry
This new documentation introduces and explains toolbox authentication in Microsoft Foundry, detailing how authentication is centralized on project connections and how agents can use various authentication types without handling authentication logic themselves. It provides step-by-step guidance for configuring OAuth identity passthrough, enabling per-user access to downstream services while maintaining user permissions and boundaries.
- Create a private skill catalog (preview)
This new documentation introduces the ability to create a private skill catalog in Microsoft Foundry using Azure API Center, enabling organizations to register, govern, and share agent skills internally. It details prerequisites, setup, governance, access control, and troubleshooting, representing a significant new capability for managing and distributing AI agent skills within an enterprise.
- What is Toolbox in Foundry?
This new documentation introduces the Toolbox feature in Microsoft Foundry, which centralizes tool management for AI agents. It describes how Toolbox enables centralized configuration, authentication, governance, and sharing of tools and skills across agents, including new preview features like tool search and skills.
- Types of tools in Foundry Agent Service
Page removed or relocated. The documentation page detailing the types of tools available in Microsoft Foundry Agent Service, including built-in and custom tools, toolboxes, and related management concepts, has been removed. This eliminates a comprehensive resource for understanding and implementing tool extensibility in Foundry-based Copilot agents.