Microsoft launches Agent 365 and Work IQ to command the bot workforce

Microsoft Ignite 2025 just kicked off, and the company is making it crystal clear where enterprise software is heading: less about chatting with a generic bot, and more about managing an army of autonomous AI workers.

At the center of the announcements is Microsoft 365 Copilot getting infused with new model options from Anthropic, a unified intelligence layer dubbed Work IQ, and a full governance dashboard called Agent 365.

Smarter Office agents and deeper context

Microsoft introduced Work IQ, described as an intelligence layer that reads not just your corporate org chart, but your dynamic work chart. By combining knowledge across your emails, meetings, chats, and personal work patterns, it aims to anticipate next steps and route tasks to the right agent automatically.

In the productivity suite, Word, Excel, and PowerPoint are getting dedicated agents right inside Copilot chat. Instead of juggling complex menus, you give a simple prompt, the agent asks targeted follow-up questions to understand your intent, and then spits out an interactive preview. Users can iteratively refine slides, spreadsheets, or docs via chat or jump straight into the native Office apps.

Model flexibility is expanding too. Excel users in Agent Mode can now toggle between OpenAI reasoning models and Anthropic models. Meanwhile, Word Agent Mode is generally available, PowerPoint Agent Mode is hitting the Frontier early access program, and OpenAI's Sora 2 video generation is integrated into Copilot Create for short-form content.

Beyond docs, Copilot Voice is now open to all commercial users, letting people triage inboxes on mobile or schedule meetings with a single chat prompt.

Agent 365: A control plane for 1.3 billion bots

Citing an IDC prediction that there will be 1.3 billion AI agents by 2028, Microsoft unveiled Agent 365 to stop bot sprawl before it starts.

Agent 365 essentially treats AI agents like digital employees. Powered by Microsoft Entra, it assigns unique Agent IDs, maintains an enterprise-wide registry, and even quarantines unsanctioned shadow agents. IT admins get a centralized dashboard to track performance, monitor compliance, and enforce least-privilege access rules.

Interoperability also took a big step forward. In Teams Channels, agents can now connect to third-party tools like Jira, GitHub, and Asana via Model Context Protocol (MCP) servers. An agent can pull engineering blockers directly from Jira and immediately book a sync meeting in Outlook.

Why it matters

For everyday knowledge workers, this shift shrinks the tedious gap between rough ideas and polished deliverables. Being able to draft a presentation or crunch financial formulas purely through iterative chat removes massive friction, though many of the flashiest features remain locked behind the early access Frontier program for now.

For the industry, Microsoft's embrace of Anthropic and open standards like MCP shows it knows enterprise customers won't tolerate a single-vendor lock-in. By letting teams build with both OpenAI and Claude, Microsoft keeps Copilot as the dominant interface regardless of which lab wins the frontier model race.

We think Microsoft's real moat here isn't raw model benchmark performance, but enterprise governance. Companies are terrified of rogue bots leaking sensitive data. By building identity, auditing, and permissions directly into Agent 365, Microsoft is offering the safety rails that CIOs desperately need before letting agents run wild in their business processes.