Implicator
Microsoft on Monday made Copilot Cowork available to customers enrolled in its Frontier early-access program, expanding the Anthropic-powered agent from a limited research preview to a broader enterprise audience. The tool, built on the same technology platform as Anthropic's Claude Cowork, handles long-running tasks across Microsoft 365 by creating a plan and executing it in the background, with visible checkpoints where humans can steer. The rollout is part of Wave 3 of Microsoft 365 Copilot and arrives as Microsoft races to boost AI adoption amid competition from Google's Gemini and autonomous agents like Claude Cowork itself, Reuters reported.
Key Takeaways
- Copilot Cowork, built on Anthropic's Claude Cowork technology, is now available to Microsoft 365 Frontier program customers for multi-step task delegation
- A new Critique feature pairs GPT and Claude models to cross-check research, scoring 13.8% higher on the DRACO benchmark
- Microsoft 365 E7 launches May 1 at $99/user/month, bundling Copilot, Agent 365, and security tools at an $18 discount over separate pricing
- Microsoft's share price dropped 14%+ since Claude Cowork's January debut, pressuring the company to prove embedded AI beats standalone agents
From assistant to executor
Until now, Copilot has focused on single-shot tasks. Summarize this email. Draft that slide deck. Useful, sure. But each output still required a human to stitch the pieces into a coherent workflow. Copilot Cowork changes the model entirely. Users describe a desired outcome, and the system breaks it down into steps, reasons across Excel, Outlook, Teams, and SharePoint, then carries the work forward autonomously, SiliconAngle reported.
Picture a monthly budget review. That normally eats a morning. Open the spreadsheet, pull last month's numbers, switch to Outlook for the updates, copy them back. Copilot Cowork handles the rest. Pulls the numbers, checks in with colleagues by email, assembles a report. Progress checkpoints let users review, redirect, or stop the process entirely.
"This isn't about generating content or answers," said Barton Warner, SVP of Enterprise Technology at Capital Group, one of the first organizations to test the feature. "It's about taking real action, connecting steps, coordinating tasks, and following through across everyday workflows."
One critical distinction separates the Microsoft and Anthropic implementations. Claude Cowork runs locally on a user's device. Copilot Cowork runs in the cloud within a customer's Microsoft 365 tenant, wrapped in the company's enterprise data protection and grounded in what Microsoft calls Work IQ, a context layer drawn from emails, files, meetings, and organizational relationships.
"We actually don't work locally, and that's a feature, not a bug," Spataro told Fortune when the product was first announced in early March.
Claude checks GPT's homework
Alongside Cowork, Microsoft introduced a "Critique" feature for its Researcher agent that pairs GPT and Claude in a generative-then-evaluative loop. One model drafts a research response. The other reviews it for factual accuracy and citation quality, a process Microsoft compared to "academic and professional research settings," Engadget reported.
Microsoft says adding the Critique layer pushed its Researcher score to 57.4 on the DRACO benchmark for deep research quality, a 13.8% gain. The roles can be reversed, with Claude drafting and GPT reviewing.
Get Implicator.ai in your inbox
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
Microsoft also shipped Model Council, which feeds the same question to both models and lays the results next to each other. When two AIs agree on a claim, it probably holds. When they don't, you know where to look.
But the multi-model approach reflects a broader strategic pivot. Microsoft built the original Copilot entirely on OpenAI's models. Now the company pitches model diversity as the selling point. "Every 60 days at least, there's a new king of the hill," Spataro told Fortune. "There's so much demand for a platform that doesn't feel like, 'I have to skip over to the next vendor.'"
The $99-per-seat bet
Copilot Cowork is headed for Microsoft 365 E7, the company's new premium tier launching May 1 at $99 per user per month. That bundle combines E5, Copilot, Agent 365, and the Entra security suite, a package that would cost $117 if purchased separately, CRN reported. It marks Microsoft's first new enterprise license tier in roughly a decade.
The financial pressure is visible. Microsoft's share price had dropped more than 14% since Anthropic debuted Claude Cowork in mid-January, Fortune reported in early March, fueling investor anxiety that autonomous AI agents could reduce dependence on traditional productivity suites. Microsoft bets the opposite is true. Embedding agentic capabilities inside the apps enterprises already pay for makes its platform stickier, not more expendable.
Copilot paid seats have grown more than 160% year over year, with daily active usage up tenfold, according to Microsoft. Agent 365, the company's control plane for managing AI agents across an organization, registered tens of millions of agents during just two months of preview. Internally, Microsoft says it now tracks more than 500,000 agents across the company.
For organizations already deep in the Microsoft 365 stack, the question isn't whether AI agents show up at work. It's whether they'll answer to Microsoft, Anthropic, or both.
Frequently Asked Questions
What is Copilot Cowork and how does it differ from regular Copilot?
Copilot Cowork handles long-running, multi-step tasks autonomously across Microsoft 365 apps. Regular Copilot focuses on single-shot tasks like drafting emails or summarizing documents. Cowork creates a plan, executes steps across Excel, Outlook, Teams, and SharePoint, and provides checkpoints for human oversight.
How does the new Critique feature work in Microsoft 365 Copilot Researcher?
Critique pairs two AI models in a review loop. One model (GPT or Claude) drafts a research response, then the other reviews it for accuracy and citation quality. Microsoft says this architecture improved the Researcher's DRACO benchmark score by 13.8%, reaching 57.4.
What is the Microsoft 365 Frontier program?
Frontier is Microsoft's early-access channel where enterprise customers can test experimental AI features before general availability. Copilot Cowork and the new Researcher features are currently available through Frontier ahead of a broader rollout.
How much does Microsoft 365 E7 cost and what does it include?
E7 launches May 1 at $99 per user per month. It bundles Microsoft 365 E5, Copilot, Agent 365, and the Entra security suite. Purchased separately, those components cost $117, making the bundle an $18-per-user discount.
What is the difference between Copilot Cowork and Claude Cowork?
Both use the same underlying technology platform from Anthropic. Claude Cowork runs locally on a user's device. Copilot Cowork runs in the cloud within a customer's Microsoft 365 tenant, with enterprise data protection and access to organizational context through Microsoft's Work IQ layer.
Harkaram Grewal
Maps the India–Germany–U.S. AI triangle from New Delhi. Background in cross-market operations and business development. Writes about supply chains, enterprise adoption, and talent—the unsexy forces that actually move global AI.
