Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
This article describes the fixes and improvements to existing functionality that are included in this build.
Visit the Microsoft Copilot Studio Blog to stay up-to-date on the latest news!
Version 26.05.17-20 updates
New Features
Observability experience: The Evaluation area is renamed to Observability with updated labels and icon, a baseline hub with placeholder sections, improved inventory (default sort by last modified, corrected last published sort, owner column, icons, quick filters, and clearer empty states), and the Analytics tab now offering the full analytics experience with deep links and drilldowns.
Microsoft 365/Teams channel enhancements: Makers can submit agents to their organization catalog via an inline flow with in-panel navigation and error messaging; setup and submission panels have more consistent layouts, improved accessibility semantics, and clearer status announcements; edits to channel details are kept as drafts until saved.
Agent creation and management: Makers can delete connections directly in Microsoft Copilot Studio; creating agents via home tiles and natural-language prompts reliably uses classic one‑click defaults (including no authentication and automatic voice setup when available); the home creation tile is labeled “Agent.”
Analytics updates: Hybrid agent view includes sessions over time and a sessions summary; AI Summary analytics is correctly scoped for Enhanced task completion (preview) agents; the Effectiveness section adopts a new card layout and removes older sentiment/empty states; analytics summary adds a “Last 360 days” preset with custom ranges up to 360 days.
Voice and training: Classic voice agents automatically get default text‑to‑speech settings per language at enablement, when adding languages, and from the voice template; training results and status messages are clearer and only reference custom lexicon when enabled, adding a Pronunciation column when applicable; test chat defers voice until agents are fully provisioned.
Breaking change: Text‑to‑speech voice selection now requires a “mode” setting (defaults to Classic). Integrations that don’t supply the mode parameter might see different available voices.
Fixes and Improvements
Continued improvements to performance and reliability.
Microsoft 365/Teams channel stability and UX: Distribution options are disabled until an agent is published; preview and submission panels use consistent layouts and headers; button labels no longer wrap; status panels no longer get stuck during initial load or admin approval propagation and remain visible during propagation; channel descriptions use consistent sizing and alignment.
Evaluation and Observability: New agents’ Evaluation pages load correctly; tabs for Evaluation and Analytics only show on main pages (hidden on drilldowns); clearer loading spinners prevent flashing states; improved navigation, layout, and visuals in Maker Evaluation (including RTL emoji alignment), with correct updates when switching agents and clearer errors for unmanaged customizations; option to use a test chat conversation is removed when creating a new evaluation.
Analytics and UI: Enhanced task completion (preview) experiences appear as expected; tooltips display translated text; the Effectiveness section resolves overflow issues and removes older user sentiment and empty-states content; Total Reactions is removed.
Authoring and overview experience: The management evaluation page adds overview cards; the overview page no longer crashes on inaccessible Dataverse connections and shows access‑denied messaging; inventory and hub layouts corrected; details Save button behavior fixed; fallback label reads “Redirect to a topic”; home page greeting shows the user’s name; deprecated tools/triggers are removed from suggestions; “WorkIQ” section and teaching bubble are removed where no longer applicable.
Workflow and tools: Creating a workflow from the agent designer works in the Power Platform UX; MCP tool configuration correctly lists tools for connected agents; AI Prompt tools allow adding outputs as expected; tool‑call test sets retain expectations after edits; custom pronunciation lexicon entries support whitespace; Microsoft 365 and Teams approval/share flows guide publish decisions and surface share/download after channel add.
Security
Tenant Data Loss Prevention can block publishing copilots with AI features when disabled in the Power Platform admin center.
Users in geo‑restricted regions are prevented from selecting or testing Anthropic models in computer use agents.
The Analytics Viewer role can no longer be assigned to security groups and is limited to individual users.
Newly created agents correctly show the Draft agent status card in Security Health.
Return to the Microsoft Copilot Studio Release Notes.