Copilot shifting strategy
Reports say Microsoft’s Copilot strategy is evolving toward paid tiers and multi‑model orchestration — teams are testing ‘Council’ setups that use multiple models for verification and integrating other vendors’ models. (x.com) That matters for product teams because pricing changes and a multi‑model architecture affect vendor lock‑in, auditability, and per‑query costs. (x.com)
Microsoft is changing what “Copilot” means. For the past two years, the pitch was simple: pay for Microsoft’s AI assistant, get OpenAI-grade intelligence woven through Word, Excel, Outlook, and Teams. That is no longer the whole story. In March, Microsoft began rolling out a new version of its work AI that is explicitly multi-model, not single-model, and increasingly split across pricing tiers that separate free chat, paid app integration, and a new premium enterprise bundle (microsoft.com 1) (microsoft.com 2). The clearest sign came from Microsoft’s own Researcher product inside Microsoft 365 Copilot. On March 30, the company introduced two new systems, Critique and Council. Critique has one model generate a report and a second model review and refine it. Council goes further and runs multiple models on the same task so users can compare where they agree and where they diverge (techcommunity.microsoft.com) (support.microsoft.com). That is not a cosmetic feature. It is Microsoft admitting that one model is not enough for the kind of work it now wants Copilot to do. That shift matters because Microsoft is no longer pretending Copilot is just “OpenAI in Office.” The company says outright that Copilot should “choose the right model for the job regardless of who built it.” In practice, that now includes Anthropic. Microsoft first added Anthropic models to Copilot Studio in September 2025, and by January 2026 those models were available by default in most geographies. Admins can enable or restrict them, and if Anthropic is turned off, Microsoft falls back automatically to OpenAI GPT-4o (microsoft.com). The technical architecture is starting to look less like a single assistant and more like a traffic controller. Once Microsoft made that move, the pricing story changed too. Copilot Chat remains available at no additional cost for eligible Microsoft 365 customers, but the full Microsoft 365 Copilot license is still a separate paid product at $30 per user per month for enterprise customers. Microsoft’s business pricing page also shows a discounted business tier starting at $18 per user per month when paid yearly, while deeper agent features require Azure capacity on top (microsoft.com 1) (microsoft.com 2). In other words, Microsoft is not flattening AI into one simple subscription. It is building a ladder. The top rung of that ladder is now visible. Microsoft announced Microsoft 365 E7, a new “Frontier Suite” that bundles E5, Microsoft 365 Copilot, Entra Suite, and Agent 365 for $99 per user per month, with general availability set for May 1, 2026 (microsoftpartners.microsoft.com). That bundle tells you how Microsoft thinks this market will work. Basic chat can be cheap or free. Real orchestration, governance, and action-taking will be sold as premium enterprise infrastructure. The product changes point in the same direction. Copilot Cowork, announced on March 9 and opened to the Frontier program on March 30, is built for long-running, multi-step tasks that move beyond one prompt and one answer. Microsoft says it uses “skills from Claude and Microsoft built in,” reasons across files and tools, and keeps work observable with checkpoints and approvals (microsoft.com 1) (microsoft.com 2). At the same time, Copilot Studio is expanding multi-agent orchestration and governance controls so companies can connect many agents across systems instead of running isolated bots (microsoft.com). That is the real shift. Microsoft is moving Copilot away from being a branded chatbot and toward being a managed AI stack. The surprising part is how openly it is now embracing model plurality after spending so long defining its AI future through OpenAI. The concrete detail is sitting in Microsoft’s own support page: if you want to use the new Auto and Model Council features in Researcher, you must be in the Frontier program, and your admin must allow access to Anthropic models first (support.microsoft.com).
