🧠 Food for thought

Implications, context, and why it matters.

Microsoft’s move to multi-model Copilot features follows performance results, not only partnerships

• Microsoft is leaning into a multi-model approach after internal results found that pairing AI models produces better output.
• On a benchmark for deep research quality, Microsoft’s “Critique” feature uses OpenAI’s GPT for drafting plus Anthropic’s Claude for review. It earned 57.4, above Claude Opus 4.6 running alone at 42.7 1.
• Microsoft is also stepping back from exclusive reliance on OpenAI for Copilot. Leaders found Anthropic models beat OpenAI models, including the newly released GPT-5, on work tasks such as building PowerPoint decks plus running financial functions in Excel 2.
• These higher-end capabilities aim to lift slow adoption. As of January, paid Copilot seats reached 15 million, about 3.3% of Microsoft’s 450 million commercial Microsoft 365 users 3.

AI is shifting from assistant to long-running agent, raising the stakes for enterprise software

• Copilot Cowork suggests a jump from quick help to an agent that can run multi-step work, such as building a product launch plan, while keeping the user in the loop 4.
• Microsoft’s focus is an enterprise “trust layer,” meaning software controls that pick, supervise, plus govern multiple AI models so Copilot can coordinate them 1.
• That setup lets companies run agentic tools, AI systems built to take actions across several steps, using Anthropic technology inside Microsoft 365. Identity, permissions, plus compliance rules apply by default, with actions and outputs available for audit 4.
• Microsoft is packaging this into a new paid tier called Microsoft 365 E7: The Frontier Suite, priced at $99 per user per month 5.