This analysis examines why Microsoft Built Copilot Cowork on Anthropic, what evidence supports the argument, and where the conclusion still has limits.
Read the evidence below as a decision trail: what changed, why it matters, which trade-offs shaped the result, and where the conclusion still depends on context.
Microsoft no longer just helps you write files, but helps you "do the whole thing"
On March 9, 2026, Microsoft released the third wave of updates (Wave 3) for Microsoft 365 Copilot. The core product isCopilot Cowork— An AI agent capable of performing long-term, multi-step tasks. All you have to do is say, "Help me prepare for my client meeting tomorrow," and it will automatically assemble the presentation, pull up the financial data, send the letter to the team, and schedule the preparation time. Instead of producing a draft for you to post, you do it from start to finish.
Jared Spataro, Microsoft's chief of AI marketing, said directly: "Copilot is moving from assistance to real execution." This sentence is very important - it means that Microsoft's positioning of the AI assistant has been upgraded from "helping you think" to "helping you do it."
In-depth dismantling: Copilot Cowork’s technical architecture and business layout
The brain of Anthropic, the body of Microsoft
The most noteworthy technical detail about Copilot Cowork: its inference engine isClaude from Anthropic, and using the same set of Claude Coworkagentic harness(A system of frameworks and guardrails that allow AI models to operate external tools). In other words, Microsoft did not build its own Agent architecture from scratch, but directly put Anthropic's Agent technology "in enterprise clothes."
But the key difference is how it's deployed. Anthropic's own Claude Cowork is executed on the local machine; Microsoft's version runs in the cloud, embedded in the customer's Microsoft 365 tenant, and connected to Microsoft's uniqueWork IQ— An intelligence layer that extracts organizational context from your emails, files, meetings, and chat history. Spataro puts it this way: “Models can reason about data, but they don’t understand work.” Work IQ is what fills this gap.
Wave 3 is not just Cowork: the multi-model architecture is fully online
Wave 3 also announces the entry of Microsoft 365 CopilotMulti-model era. Claude can be used in the main conversation (via the Frontier solution), running in parallel with the latest models from OpenAI. Microsoft emphasizes that this is an "open, heterogeneous environment" that is not tied to a single model vendor.
The change this brings to the Copilot chat interface is that chat is not just about Q&A;Entrance to Creation and Execution. You can create documents, spreadsheets, presentations, trigger meetings, send letters and other actions directly from the conversation without switching between tools.
Agent 365: 500,000 Agents require a management platform
During Microsoft's internal testing, more than500,000 Agents, generating more than 65,000 responses per day. When agents spread like wildfire across an enterprise, governance becomes a core issue.
Agent 365is Microsoft’s answer — an AI Agent control panel that allows IT and security teams to observe, govern, manage, and protect Agent activity across the entire organization. It integrates with existing Entra (identity), Defender (security), Purview (compliance), Intune (device management) infrastructure. Launched on May 1, per person per month$15。
IDC predicts that by 2028 there will be1.3 billion Agentscirculate in the market. Microsoft is clearly betting that Agent governance will be the next huge platform business opportunity.
E7 Frontier Suite: All for one price
Microsoft releases simultaneouslyMicrosoft 365 E7, priced per person per month$99, packages E5 ($60) + Copilot ($30) + Agent 365 ($15) + Entra Suite ($12) + Advanced Defender/Intune/Purview. Buy them separately for $117, bundle them and save $18.
The signal from this pricing strategy is clear: Microsoft wants companies to get it right at once, not in batches. As AI Agents begin to erode the value of traditional SaaS (Microsoft stock has fallen 14% since Anthropic launched Claude Cowork in January, as investors worry that AI Agents will reduce reliance on SaaS software), tying everything into one package is both defense and offense.
【My point of view】
The smartest thing Microsoft did this time was not the technology itself, but the recognition of the fact that its own Agent technology was not as good as Anthropic, so it used it directly. Integrating Claude's agentic harness into M365, and then using Work IQ to add organizational context, this strategy is much better than building it from scratch. In the AI competition, "who is closest to the user's workflow" may be more important than "whose model is the strongest" - and Microsoft is almost unbeatable at this point.
But what I care more about is the product Agent 365. 500,000 Agents, 65,000 responses per day, these are just the numbers from Microsoft's own internal testing. When enterprises deploy large-scale deployments, Agent management and security will become as basic a requirement as "IT management of user accounts." Microsoft uses $15 per person per month to hold this position, which is equivalent to grabbing "Active Directory" in the Agent era. If successful, it could be a bigger business than Copilot itself.
As for the fear reflected in the stock price’s 14% drop — will AI Agents replace SaaS? I don't think it will happen in the short term, but there will be pressure in the long term. Microsoft's E7 pricing strategy is actually a response to this problem: Rather than letting Agent erode the value of individual products, it should package everything into an indivisible platform. This is a typical Microsoft play, and one that has generally worked historically.
Practical questions and boundaries
What is the difference between Copilot Cowork and regular Copilot?
Generally, Copilot is a question-and-answer auxiliary tool to help you write a draft or organize information. Copilot Cowork is an AI Agent that can autonomously perform long-term, cross-application multi-step tasks - such as completing complete workflows such as assembling briefings, pulling data, sending letters, and scheduling meetings from a single command.
What’s the difference between Copilot Cowork and Anthropic’s Claude Cowork?
Both share the same Claude inference engine and agentic harness. The difference lies in deployment: Claude Cowork is executed on the local machine, while Copilot Cowork runs in the cloud M365 tenant and is connected to the Work IQ organizational intelligence layer and enterprise security control, making it more suitable for deployment in large organizations.
How much does the E7 Frontier Suite cost? What's included?
$99 per person per month, includes Microsoft 365 E5, Copilot, Agent 365, Entra Suite and advanced security features. The total total is about $117 when purchased separately, and you save $18 on the bundled plan. Officially launched on May 1, 2026.
Original source:Copilot Cowork: A new way of getting work done（Microsoft 365 Blog, 2026-03-09）；Introducing the First Frontier Suite built on Intelligence + Trust（The Official Microsoft Blog, 2026-03-09）
What to take away
The useful decision is not to accept the headline at face value, but to test whether the evidence supports Microsoft Built Copilot Cowork on Anthropic in the reader's own context.