Get things done just by talking
The slowest part of software was never the doing — it’s the finding. Which screen, which button, which setting buried two menus deep. By the time you’ve navigated to the thing, the thought that sent you there has half-evaporated. Voice collapses that distance: you say what you want while you’re still thinking it.
That’s the Copilot. It’s voice-first by design — click the globe, speak, and it interprets the intent rather than making you spell out the steps. It knows your workspace well enough to pick the right tool, the right agent, or the right conversation, so “get me last month’s numbers and start a draft reply to the Acme thread” becomes one spoken sentence instead of a dozen clicks.
Say it, and it acts
Click the globe and speak
Open the Copilot and say what you want in plain language — the way you’d ask a colleague across the desk. No command syntax, no setup, no hunting for the right menu. Just the request.
It figures out the approach
The Copilot interprets what you meant and decides how to get it done — which tool to use, which agent to run, which conversation to act in. You described the outcome; it works out the route, grounded in your workspace.
It advises or it acts
For a question, it answers. For a task, it directs the workspace to do it and reports back. It takes action across your tools, agents, and conversations — so the work happens, not just a description of how you’d do it.
A request to the Copilot sounds like something you’d actually say out loud:
Because it learns how you work and understands your business, the spoken request can be loose and it still lands right — it knows which “Acme thread,” which “growth channel,” and how you like the summary. You’re not writing a precise command; you’re talking to something that already has context.
Frequently asked questions
The Copilot is voice-first — click the globe and just say it — but it understands typed requests too. Either way, you describe the outcome in plain language rather than navigating menus or learning command syntax.
It takes action across the tools, agents, and conversations in your workspace — pulling data, drafting messages, running an agent, posting to a channel. It interprets the intent, picks the approach, and either advises you or directs the workspace to carry it out.
You stay in the loop. The Copilot acts within what you authorize, and for anything consequential it confirms before going ahead. You can also just tell it to stop or do it differently, out loud, mid-task.
No. It runs alongside your main chat rather than taking it over, so a voice request runs in parallel and your primary conversation stays right where you left it.