Skip to main content

The Switch Connect team are in San Francisco this week for Microsoft Ignite 2026. This year's 'Book of News' runs to 42 pages across Azure, Copilot, security and Windows, and covers all of the key announcements from the event. We've put together the an overview of the announcements that matters to every Australian organisation still running communications and collaboration workloads on-premises - and around it, a clear picture of how Microsoft expects businesses to govern AI in 2027. Here's what we think deserves your attention.

Microsoft has put on-premises servers on notice

The announcement that matters most: Microsoft launched a new on-premises security initiative covering supported Exchange Server, SharePoint Server and Skype for Business Server workloads - designed to help organisations understand their most important risks, move eligible workloads to Microsoft 365, and better protect the critical systems that must remain on-premises. Microsoft's recommendation is explicit: move eligible workloads to Microsoft 365 for cloud security and resilience.

Read between the lines and this is the strongest signal yet. If your organisation still runs Skype for Business Server or Exchange Server, Microsoft now considers that estate a security risk to be managed down, not a platform to be maintained. The practical question is no longer whether to move your calling and collaboration to the cloud - it's how, and in what order.

AI agents get a governance layer

The dominant theme of Ignite was AI agents moving from experiments into production - and the controls organisations need before that happens. The announcements stack up into a full governance layer.

Agent 365, Microsoft's observability platform for AI, gains cost management with spending policies, limits and alerts across usage-based AI services, plus Dynamic Agent Risk - a comprehensive agent risk assessment that fuses identity, access, data, endpoint and network signals, visible across the Microsoft 365 admin centre, Defender, Entra, Intune and Purview.

Microsoft Purview will block Copilot from referencing files containing sensitive information even when those files aren't labelled, closing one of the most common gaps we see in Copilot readiness work. Microsoft Intune adds controls to discover which AI agents are running on managed devices, mark devices with unapproved agents as noncompliant, and block them from company data. And at the operating system level, Windows is introducing process container controls that isolate each running agent with clear limits on what it can reach and how it proves its identity.

The message for IT leaders: Microsoft has now shipped the tooling to answer "what AI is running in our environment, what can it touch, and what is it costing us". Organisations that can't answer those questions today have a defined path to get there.

Security operations go agentic

Microsoft Perception is now generally available - specialised security agents that triage alerts, investigate attacks and prioritise exposures, with the Triage Agent now covering about 80% of alerts in a typical Microsoft Defender queue. Alongside it, the Integrated Security Operations Center reaches general availability, and new vulnerability shielding will autonomously deploy targeted, temporary controls to vulnerable systems during an active attack - buying defenders time to apply a permanent fix.

Microsoft Entra also expands, with Conditional Access policy plans, high assurance credentials, backup and recovery, and strengthened Privileged Identity Management all in preview. Several of these will become standing checks in how we assess Microsoft 365 tenants over the coming year.

The Copilot experience is being rebuilt

For everyday users, the new Copilot brings Home as a single starting point, Office in Copilot now generally available - Word, Excel and PowerPoint inside the Copilot app - and Autopilot, a long-running personal agent that colleagues can @mention in a Teams channel to request information or actions. If your Copilot rollout stalled after licensing, the experience your users see in 2027 will look very different from the one they were trained on.

Copilot is also now openly multi-model. Auto, its smart router, picks the right model and effort for each task, and users can manually select models from OpenAI or Anthropic - with model access governed by organisational policy. That makes "which AI models are our staff allowed to use" a tenant policy decision every IT leader now owns, alongside the agent governance controls above.

What to do with all this

Three practical moves: if you're still running Skype for Business Server or Exchange Server, start scoping the migration now, on your timeline rather than a forced one. If Copilot or AI agents are anywhere in your 2027 plans, get a baseline of your tenant's readiness and governance posture before the agents arrive. And if your security operation is stretched, the agentic SOC tooling is worth a serious look.

Switch Connect works with 400+ organisations across 65+ countries on exactly these moves - Teams calling migrations, Microsoft 365 security, and AI and Copilot readiness. If Ignite raised questions about your environment, talk to us today.