AI, Cloud PCs and the Future of Work – with Christiaan Brinkhoff
1. Introduction
In this episode of Impact of AI: Explored, we (James O’Regan and Gerjon Kunst) sit down with Christiaan Brinkhoff from Microsoft to talk about how AI and Cloud PCs are reshaping end-user computing and IT. We explore what work could look like by the end of 2026, how agents will change daily operations, and what this means for IT pros, security teams, and everyday users. It is a conversation about real-world change: from early experiments and POCs to production workloads, new permission models, and the mindset shift needed to actually adopt AI in the enterprise.
2. Meet the Guest
Christiaan Brinkhoff is Principal AI Product Manager and Community Director for Windows Cloud at Microsoft. He leads feature strategy for Windows 365 and Windows Cloud, bringing together Windows, AI, Azure Virtual Desktop and Microsoft Intune to build the next generation of cloud-based, AI-powered experiences. Christiaan has been involved with Windows 365 since its earliest incubation stages and has contributed to key features such as Windows 365 Boot, Switch, the Windows 365 app, and new AI integrations on the platform. Beyond his product work, he is a well-known community leader, speaker and author in the virtualization and EUC space, helping customers and partners imagine what’s next in hybrid and remote work.
3. Setting the Stage
We’re in what Christiaan calls the first and second phases of AI adoption at work: experimentation, copilots as assistants, and lots of POCs slowly turning into real workloads. Organizations are learning how to use AI for summarization, content creation, meeting notes, and “human-in-the-loop” assistance, but the next wave is coming fast: agents that actually do work inside Windows and Cloud PCs. This episode looks at why that matters now—because security, governance, culture, and budgets are all being forced to adapt—and what listeners can expect as AI moves from novelty to normal in EUC and IT operations.
4. Episode Highlights
- From assistant to autonomous agent
One big turning point in the discussion is Christiaan’s view of the “next wave” of AI: agents running on Cloud PCs that take over parts of your workload, with humans staying in control and eventually moving toward more autonomous operation. These agents will spin up cloud resources on demand, operate like real users across multiple apps, and then shut down when the task is done, fundamentally changing how we think about 24/7 virtual machines. - Security, RBAC and process-based permissions
Another key moment is when we dive into the “good, bad, and ugly” of agents, especially around prompt injection, over-permissioned systems, and the need for new access models. Christiaan expects a shift from identity-based permissions to process- and workflow-based permissions—temporary, scoped rights that exist only while a specific AI-driven process runs and disappear when it completes.
5. Deep Dive: Agents, Cloud PCs and the New Permission Model
Christiaan describes where we are today as AI phase one and two: copilots embedded in tools like Windows and Microsoft 365, helping with summaries, drafts, and navigation. By 2026 and beyond, he expects a broader shift to agents running inside Cloud PCs in the data center, orchestrated by models like the OpenAI Operator model or similar agentic models, that behave like real users across applications. These agents won’t just live in a browser—they’ll operate at the Windows layer, clicking through Office, line-of-business apps and management tools to complete tasks across multiple layers of the stack.
To make that safe, Christiaan foresees a new era of delegated controls and RBAC: think of an “agentic user” with strictly scoped permissions for a specific process, rather than a broadly permissioned service account. Instead of continuous access, permissions become time-bound and task-bound—granted when a process starts, revoked when it ends—similar in spirit to how UAC added an extra layer between users and admin rights years ago. This will require new tooling, models and a lot of work from sysadmins and security engineers to describe processes, map rights to each stage of a workflow, and make sure agents cannot “go left or right” outside their intended path.
6. Real-Life Stories & Examples
- From POCs to production
We talk about customers who started with small, ring-fenced AI POCs—improving a single process, often with a modest budget and unclear ROI—and are now moving those first use cases into production. The big hurdle is not the tech but getting security and finance on board with innovation where the ROI is not fully known upfront, and shifting from “fix this for 40,000 euros” to “let’s invest and see if this changes how we work.” - Copilot in Intune for real admin workflows
A concrete example Christiaan gives is Copilot for Intune: instead of manually digging through logs and consoles when the helpdesk reports performance problems, admins can ask AI to correlate Cloud PCs and physical devices, check for CPU spikes above a threshold, and summarize issues. This removes the need for manual CSV exports and Excel work, freeing admins to focus on decisions and actions while future agent capabilities will start automating repetitive tasks like license correlation, provisioning or remediations—with humans staying in the loop. - AI as an accessibility and life-enabler
Christiaan shares a personal angle: his father is blind and already relies on assistive technologies like screen readers, but AI combined with devices like smart glasses could describe the world around him, read what’s on screen or even interact with apps on his behalf. We explore how this same model could help people who struggle with traditional input methods—such as James’ daughter, who has dyspraxia—by enabling voice interaction and back-and-forth conversations with AI for schoolwork and daily tasks. - Autonomous cars as a metaphor
The evolution of self-driving taxis in San Francisco becomes a metaphor for AI adoption: at first buggy and even sabotaged by human drivers, now far more stable but still subject to real-world issues like power cuts and complex edge cases. Just like autonomous driving, enterprise AI has to balance innovation speed with safety, handling countless variables while slowly building user trust over years, not months.
7. Key Takeaways
- AI at work is moving from “summarize this meeting” to agents that can operate Cloud PCs, complete tasks across multiple apps, and shut down when done.
- Security and governance will shift toward process-based, time-bound permissions for agents, moving beyond traditional identity-only RBAC models.
- The most successful organizations start small: narrow POCs, ring-fenced use cases, and a frontier mindset that accepts experimentation, failure, and iteration.
- Copilot-style tools in platforms like Intune will first supercharge diagnostics and insights, then gradually automate more admin actions with humans still approving critical steps.
- Adoption is as much cultural as technical: IT and security can no longer be the “department of no” and must become enablers with clear guardrails and education.
- Generational and accessibility angles matter; AI can be a huge leveller for people with disabilities and for younger generations who will grow up with “PhD-level tools” in their pockets.
- Not every sector or organization will enable every AI feature—controls exist to disable or selectively roll out AI capabilities in environments like AI-enabled Cloud PCs.
8. Closing Thoughts
For us, this conversation with Christiaan reinforces that we’re still early—but the direction is clear: Cloud PCs, agents and AI-native experiences will fundamentally change how IT is done and how end users experience work. The challenge and opportunity for all of us in IT is to ride this wave deliberately: experiment, involve security early, focus on “what’s in it for me” for end users, and be bold enough to try new things while staying responsible.
We’ll keep exploring these themes in future episodes—from AI browsers and private AI to new agent patterns and governance models—so if you have questions or real-world stories, we’d love to hear them and maybe feature them next time. Join the conversation, share your experiments, and let’s figure out together what the next phase of AI in EUC and cloud really looks like.
