I've been experimenting with Microsoft's Windows 365 for Agents service — building an autonomous AI system to handle a practical business task. What struck me immediately was the speed: within minutes, the platform automatically provisioned two Cloud PCs running Windows 365 to execute my agent's workflow, and they were enrolled in Intune with standard security policies applied automatically. No manual VM configuration required.
The Problem I Was Solving
The task was checking inventory availability on a UK retailer's website that has no public API — a job that normally means manually looking up each SKU and size variant one by one. Tedious, repetitive, and exactly the kind of work that should be automated.
I created a computer-use agent powered by Claude Sonnet 4.5, designed to navigate websites, read screen content, and perform mouse and keyboard actions autonomously. It received plain-language instructions and got on with it — no brittle scripts, no screen coordinates hard-coded in advance.
What the Infrastructure Actually Looked Like
Rather than running the agent locally, it executed within Windows 365's Cloud PC pool — virtualised Windows desktops managed through Microsoft Entra ID and Intune, matching standard corporate device governance. Two Cloud PCs were automatically deployed within approximately 30 minutes. The machines enrolled in Intune, inherited security policies, and were ready to use. The lifecycle management was entirely handled by the platform.
How It Performed
Real-time visualisation let me watch the agent work. There were moments where human intervention was needed — CAPTCHAs, the occasional authentication challenge — but the agent completed the full workflow autonomously across multiple product checks. The output was a structured report with stock status and pricing for every item. Work that would have taken an hour done in the background while I did something else.
The Four Things That Impressed Me
Security and compliance integration. The Cloud PCs automatically inherited Conditional Access policies, Defender for Endpoint, and DLP controls. The agents operated inside the enterprise security perimeter from day one — not bolted on afterwards.
Scalability. The shared pool model means agents draw resources on-demand rather than requiring dedicated always-on VMs. It scales automatically based on workload, which matters when you're thinking about running this at any kind of meaningful volume.
IT familiarity. The Cloud PCs appear in Intune like standard employee machines. Existing workflows handle software deployments and policy updates. There's no new management toolset to learn — which is important for adoption in real organisations.
Observability and control. Built-in monitoring gives real-time visibility into what the agent is doing. You can take manual control when unexpected conditions arise. That matters when you're trying to build trust in a new kind of workload.
What Comes Next
This is a practical starting point, not a finished product. I'm planning to explore more complex workflows — automated testing, customer support triage, multi-system processes that currently require humans to bridge gaps between applications. Each deployment will go through security reviews and monitored pilot runs, treating agents the same way you'd treat a new hire or a new system: carefully, with clear boundaries, and with appropriate oversight.
The infrastructure is genuinely ready for this kind of work. The main constraint now is figuring out which tasks are worth automating and building the institutional confidence to let agents get on with them.
- Windows 365 for Agents provisions Cloud PCs automatically — no manual VM setup required
- Agents inherit existing Intune policies, Conditional Access, and Defender for Endpoint controls
- The shared pool model scales on-demand; no dedicated always-on infrastructure needed
- Cloud PCs appear in Intune like standard employee machines — familiar tooling, no retraining
- Real-time observability and manual override capability are built in from the start