Microsoft Ships Agentic AI to Excel, PowerPoint, Word, and Outlook After Two Years of Research
Key Takeaways
- ▸Microsoft shipped "edit with Copilot" in Excel after two years of research, with agentic capabilities now available in PowerPoint, Word, and Outlook
- ▸The breakthrough was grounded in treating Excel as a low-resource programming language, allowing AI models trained on the internet to better understand spreadsheet logic
- ▸Mukul Singh led the transition from Microsoft Research to product, demonstrating how fundamental AI research on obscure languages can translate into mainstream productivity features
Summary
Microsoft has successfully delivered agentic AI capabilities across its Office product suite, starting with the launch of "edit with Copilot" in Excel (formerly Excel Agent Mode). The breakthrough came from two years of research led by Mukul Singh, who transitioned from Microsoft Research to the Office Product Group to bring the vision to life. Singh's team approached the problem by treating Excel as a low-resource programming language—obscure languages that AI models traditionally struggle with because they're underrepresented in training data.
The research initially focused on how to make AI models proficient at understated programming languages like Kusto Query Language (KQL), a Microsoft-internal telemetry tool. By applying these same principles to Excel, the team discovered that Excel's formula-based structure could be approached as a low-resource programming challenge. This insight became the foundation for enabling AI agents to understand and manipulate spreadsheets autonomously. Following Excel's success, Microsoft rapidly deployed agentic capabilities to PowerPoint, Word, and Outlook, transforming how millions of users interact with productivity tools.
Editorial Opinion
The move to embed agentic AI across Office is a pragmatic validation of the "agentic AI era" narrative. By treating Excel as a low-resource programming language and applying years of research infrastructure, Microsoft shows how deep research can yield practical products that reach billions of users. This approach—grounding product launches in specific research insights rather than chasing hype—could become a differentiator as competitors race to add AI agents to everything.
