Microsoft Just Made Copilot Use GPT and Claude at the Same Time — Here's Why That Matters

March 30, 2026  ·  7 min read  ·  Reuters, AllWork.Space, TechPortal

What Microsoft Actually Announced Today

Reuters broke the story at 1:04 PM UTC on March 30, 2026. Microsoft announced three new developments for its Copilot AI platform, all part of what the company is calling "Wave 3" of Microsoft 365 Copilot.

Critique: The Copilot Researcher agent now uses both GPT-5.4 and Claude simultaneously. GPT generates the initial response. Claude then reads GPT's output and checks it for accuracy, coherence, and quality before the answer is delivered to the user. Microsoft says this reduces hallucinations. Future versions will make the flow bidirectional — Claude drafts, GPT reviews.

Model Council: Instead of a single answer, Copilot now optionally presents responses from GPT and Claude side-by-side. You choose which output fits your needs. Microsoft says this is particularly useful for complex analysis and strategic thinking where a single answer may not capture the full picture.

Copilot Cowork expansion: Microsoft is expanding early access to Copilot Cowork — an autonomous, multi-step task agent — through the "Frontier" early-access program. Cowork is based on Anthropic's Claude Cowork product and handles longer-horizon tasks with minimal human intervention.

Nicole Herskowitz, Corporate Vice President of Microsoft 365 and Copilot, said the goal is to make models "work together, not in isolation," to reduce errors and accelerate user workflows.

Why Does Microsoft Need Two AIs to Check Each Other's Work?

The Critique feature is not just a product upgrade. It is a public acknowledgment of a fundamental problem: frontier AI models still hallucinate frequently enough that Microsoft cannot ship Copilot Researcher without a second model reviewing the output.

GPT-5.4 and Claude are currently ranked among the top two models on the LMSYS Chatbot Arena leaderboard. Even so, Microsoft's answer to hallucinations is not to wait for a better model — it is to use both simultaneously, with one acting as a fact-checker for the other.

The Model Council feature solves a different problem. Power users already know that different AI models produce meaningfully different outputs for the same prompt. Instead of making users juggle multiple browser tabs, Copilot surfaces both answers at once. This is an explicit UX concession that model diversity is now a feature, not a bug.

Microsoft Copilot vs Happycapy: Multi-Model AI Compared

Both platforms now use multiple AI models. But the approach, cost, and flexibility are very different.

What the Copilot Cowork Expansion Means

Copilot Cowork is Microsoft's answer to the agentic AI wave. It handles multi-step tasks with minimal human intervention: research a topic, draft a report, schedule a follow-up, update a spreadsheet — in sequence, autonomously. Microsoft is now rolling it out beyond its original pilot cohort through the "Frontier" program.

Critically, Cowork is built on Anthropic's Claude Cowork product. This means Microsoft is using Claude both as a model in Critique (to review GPT's outputs) and as the foundational architecture for its autonomous agent product. Anthropic is now deeply embedded in Microsoft's AI stack at two distinct layers simultaneously.

For IT buyers, the Frontier expansion means enterprise teams can now request early access to Cowork through their Microsoft account manager. General availability is expected in Q3 2026 alongside the broader Wave 3 rollout.

The Bigger Picture: Multi-Model Is Now the Industry Default

Microsoft is not alone. Google Workspace already routes tasks between Gemini 3 Ultra, Gemini 3.1 Flash, and third-party models depending on complexity. Apple's redesigned Siri runs Google's Gemini with ChatGPT as an opt-in fallback. Amazon's Bedrock lets developers mix Claude, Titan, and Llama in the same application.

The era of platform loyalty to a single AI model is ending. The new question is not "which AI do you use?" but "how well does your platform orchestrate AI models together?" Happycapy was designed around this principle from the start — giving users direct access to every major frontier model and the ability to run them side-by-side or in sequence.

Microsoft's Critique and Model Council are powerful features. But they represent a $30/month subscription being retrofitted for multi-model AI in 2026. Dedicated AI-first platforms built this way from the beginning.