GitHub Copilot Invades Slack and Teams to Fix Bugs During Meetings

GitHub just pushed a batch of Copilot updates that stretch the agent well beyond the editor. The headline change is that @GitHub is now a first-class participant in Slack and Microsoft Teams, and the release also bundles a new customization hub, multi-session support in the CLI, a fresh crop of models, and multilingual dictation in VS Code.

Copilot moves into the chat room

The most consequential piece is the new Slack integration, which brings the agentic capabilities of Copilot CLI and the Copilot app into Slack channels in public preview. Mentioning @GitHub in a direct message, channel, or thread starts an agent session that can read the surrounding conversation and act on it.

Once triggered, the agent can:

• Answer questions about your code and GitHub activity.
• Triage bug reports, update existing issues, or create and label new ones.
• Investigate failures, implement changes, and validate its work in a secure cloud sandbox.
• Open a pull request and post a link back to the conversation for review.

The interesting design choice is that these sessions are multiplayer. Agent runs in Slack are shared, so a team can collaborate on the work where the request began instead of one person prompting privately. Working in the open also makes effective agent workflows easier to pick up, since developers can watch how teammates prompt the agent. Copilot can also spin up a dedicated Slack Code channel so long-running work does not pollute the original thread.

The Microsoft Teams version behaves the same way. Anyone in the conversation can ask questions, add context, and help plan or steer the work, while participants with write access to the repository can trigger Copilot to make changes. The practical unlock is that when a meeting produces an action item, you can hand it to Copilot during the discussion and start resolving it before the meeting ends.

Both integrations are in public preview on paid Copilot plans. Admins can require an extra approval on any PR authored by the Copilot identity, so agent output still passes through a human before merge.

One home for MCP servers, plugins, and skills

The Copilot desktop app now has a Customize tab that is generally available. It pulls MCP servers, plugins, skills, and canvases into one place, replacing the separate flows that previously scattered these options across the product.

Inside the tab you can:

• Discover ways to customize Copilot across MCP servers, plugins, skills, and canvases.
• Start with featured customizations selected from each section.
• Find MCP servers faster by exploring trending options and browsing by category.

One featured canvas is an Azure DevOps integration that lets you triage issues, prioritize backlogs, assign follow-ups, and hand work to Copilot to investigate, implement, or prepare for review.

Juggling parallel agents in the CLI

Copilot CLI now has a Sessions sidebar for running multiple agent conversations at once. Click the left arrow to open it, then use shortcuts to open new sessions (n), close current ones (x), and move between them. If you have ever had one agent grinding on a refactor while you wanted to ask a quick unrelated question, this is the fix.

The same CLI release adds two more useful primitives:

• /worktree (experimental) creates an isolated worktree and begins a separate conversation, giving you another workspace for exploring changes without disrupting your current work.
• /rewind works without Git to restore the conversation and files Copilot changed, while preserving your subsequent edits.

New models and on-device voice

Three models are now generally available across the Copilot app, CLI, and VS Code: Gemini 3.7 Flash, MAI-Code-1.1-Flash, and Kimi K3, an open-weight model hosted by Fireworks AI. The Flash-class options target the fast, cheap end of the spectrum where a lot of agent tool calls happen, so they matter more for cost-per-task than for headline benchmark numbers.

VS Code dictation also got a privacy-friendly upgrade. It now uses a multilingual on-device model by default, keeping audio on your machine. It can follow your configured language, use your system or browser locale, or auto-detect the language as you speak. For anyone dictating comments or prompts in a non-English codebase, that combination of local processing and automatic detection is the actual quality-of-life win.