Microsoft pushes multi‑model AI
Microsoft updated 365 Copilot so GPT and Claude can check each other’s work, signalling enterprise AI is moving toward orchestrated multi‑model stacks rather than single models. The shift comes with reliability caveats — Microsoft has a ‘Copilot code red’ response for UX issues and Claude experienced outages this week — highlighting uptime risk for desks that integrate multi‑model workflows. (geekwire.com, techcommunity.microsoft.com, moneycontrol.com)
Microsoft just put two rival chatbots in the same office suite and told one to check the other’s homework inside Microsoft 365 Copilot. In Microsoft’s new setup, OpenAI’s GPT can draft and Anthropic’s Claude can critique inside the Researcher agent and the broader Copilot workflow. (techcommunity.microsoft.com) (geekwire.com) That is a break from the first wave of workplace artificial intelligence, where one model usually did everything from writing to summarizing to answering questions. GeekWire described Microsoft’s move as a sign that the “single-model era” in enterprise artificial intelligence may be ending. (geekwire.com) The logic is simple: one model is no longer being treated like a lone genius, but like a first drafter whose work can be reviewed by a second system. Microsoft says this multi-model design lets users access Anthropic and OpenAI models directly in Copilot for longer, multi-step jobs such as building briefing documents, presentations, and Excel files from one prompt. (techcommunity.microsoft.com) Microsoft is also pushing a product called Copilot Cowork, which it says can pull from email, calendar, and SharePoint while a task is running and accept new instructions midstream. That turns Copilot from a chatbot that answers one question into more of a project runner that keeps working across several company apps. (techcommunity.microsoft.com) This matters because office software is where bad artificial intelligence answers become expensive. A wrong joke in a chatbot is annoying, but a wrong number in a finance deck or a bad citation in a strategy memo can spread through a company before a human catches it. (geekwire.com) Microsoft’s own design chief, Steve Gustavson, told GeekWire the company has a “Copilot code red” path for user-experience failures, which shows how seriously it treats trust problems inside workplace tools. That phrase matters because it suggests the weak point is no longer just model intelligence, but the full product experience around the model. (geekwire.com) The catch is that adding a second model can reduce one kind of risk while adding another. If your workflow depends on both GPT and Claude, then accuracy may improve, but uptime now depends on more vendors, more connectors, and more places where a task can stall. (geekwire.com) (techcommunity.microsoft.com) That tradeoff was visible this week. Anthropic’s status page logged an April 8, 2026 incident affecting access to Claude.ai on desktop and web, and on April 9 it also showed elevated errors for Sonnet 4.6 and higher error rates for certain connectors. (status.claude.com) Other reports described the April disruptions as Claude’s second outage of the week, with users reporting login problems, lag, and access failures. For a solo chatbot, that is frustrating; for a multi-model office workflow chained across documents and approvals, it can freeze real work in the middle. (moneycontrol.com) (techrepublic.com) So Microsoft is not really betting that one model will finally become perfect. It is betting that enterprise buyers will prefer an artificial intelligence stack that behaves more like a team of specialists, even if that means they now have to manage the reliability of the whole team instead of just one star employee. (geekwire.com) (techcommunity.microsoft.com)
