After three months in its Frontier preview, Microsoft is rolling out Copilot Cowork for general availability—an agentic system designed to execute multi-step, long-running workplace tasks end to end. Microsoft says it’s already used by more than half of the Fo
On a typical workday, the hardest part of getting a project done isn’t starting it. It’s keeping every step moving—matching the right files, stitching together dependencies, spotting when a workflow quietly stalls. Microsoft’s newest AI agent, Copilot Cowork, is built for that moment. The company is now rolling the system out for general availability after what it describes as rapid early adoption during a three-month Frontier preview.
Copilot Cowork is designed to do the work, not simply suggest what a person might do next. Microsoft says it’s meant to execute long-running. multi-step workflows on a user’s behalf—from start to finish—rather than just generating drafts or answering questions. And during the Frontier preview. Microsoft says Copilot Cowork is already used by more than half of the Fortune 500. with organizations including Accenture. Zurich Insurance. and Capital Group. among others.
Microsoft also points to how quickly the preview turned into scale. The rollout is described as one of the fastest-growing launches in the history of Microsoft’s Frontier program. For Microsoft. that speed matters: it signals that companies aren’t just experimenting with AI text generation. but are willing to hand more operational tasks to agent-style systems.
What Copilot Cowork can actually take off a team’s plate depends on the workflow. Microsoft says customers have used it to compare thousands of files across product versions. automate spreadsheet-heavy workflows. generate dependency charts. and identify stalled sales opportunities. Under the hood. Microsoft attributes these capabilities to a mix of cloud-based processing. enterprise security controls. and Work IQ—a context engine that lets the AI pull information from the tools and systems businesses already use.
The company is also selling flexibility in a way that reflects how varied enterprise environments can be. Copilot Cowork can tap into different AI models depending on the task. rather than requiring customers to rely on a single model. At launch, it runs on Anthropic’s Opus 4.8 and Sonnet 4.6 models, and Frontier customers can also access GPT-5.5. Microsoft says an in-house model called Cowork 1 is expected to arrive in the coming weeks.
All of that is paired with a pricing approach that aims to match how work actually happens inside organizations. Copilot Cowork requires a Microsoft 365 Copilot subscription, but Microsoft says usage is billed separately through a consumption-based model. Instead of a flat fee. organizations are charged according to the resources required for each task—covering model usage. context retrieval. tool calls. and runtime.
To make those costs easier to estimate. Microsoft says it has identified three categories of work: light. medium. and heavy tasks. Light tasks are described as simple requests involving limited reasoning. Medium tasks involve larger jobs that require more reasoning and information. Heavy tasks pull data from multiple sources and require deeper analysis. Microsoft argues that this design lets organizations scale usage based on need rather than paying for unused capacity.
Microsoft’s internal testing adds another selling point: the company claims Copilot Cowork is roughly 30% to 40% cheaper per prompt than competing enterprise AI offerings that use Microsoft 365 connectors. With Copilot Cowork now available worldwide. Microsoft is effectively positioning the next phase of workplace AI around delegation—handing complete projects to an agent and receiving back finished work instead of faster drafts.
The shift is subtle but significant. Copilot Cowork isn’t presented as a feature for individual answers—it’s a system for managing the messy middle of work: the long-running workflows where people usually spend time stitching steps together. double-checking what changed. and chasing down where progress actually stalled.
Microsoft Copilot Cowork Frontier preview agentic AI workplace AI Work IQ Microsoft 365 Copilot consumption-based pricing Anthropic Opus 4.8 Anthropic Sonnet 4.6 GPT-5.5 Cowork 1 enterprise security AI agents dependency charts sales opportunities
So it’s basically an intern that never sleeps, right? Sounds cool until it messes up a spreadsheet and everyone’s like “who did this”.
I don’t get it, isn’t Copilot already doing stuff? Like when they say “execute end to end” do they mean it actually clicks buttons and moves files? Or is it still just suggestions but they call it an agent.
Wait it’s taking full workflows and matching files and “spotting stalls”… so basically it can mess with your whole project timeline? If it’s used by half of Fortune 500 then it must be perfect?? Idk I feel like automation always finds a way to stall too.
Accenture + insurance companies + finance… yeah that tracks, they love stuffing new software everywhere. But the part about “Work IQ” pulling info from… like where? Your emails? Your drive? This is one of those things where they say it’s secure and then later you find out it wasn’t. Also “long-running” sounds like it’ll be down when you need it.