Introduction

Welcome back to another episode of The Impact of AI: Explored! If you’ve followed the show for a while, you know it’s usually the two of us — Gerjon and James — trading questions back and forth with our guests. This time is a little different: James was off enjoying a well-deserved holiday, so Gerjon flew solo in the host’s chair for this one. Don’t worry, James is fully caught up on the recording and just as excited about this episode as you’re about to be.
Our guest this episode is Jim Moyle, Principal Product Manager on the Windows Cloud team at Microsoft, working on Azure Virtual Desktop, Windows 365, and — the star of this conversation — Windows 365 for Agents. The theme of the episode is simple to state and genuinely big in scope: what does it actually look like when AI agents stop being chatbots in a browser tab and start operating full desktops, apps, and legacy systems on their own?

Meet the Guest

Jim Moyle has spent more than two decades in the end-user computing (EUC) and desktop virtualization world — a corner of IT he clearly still loves. His path there has been anything but linear: he’s worked as an independent consultant, spent time at Citrix and VMware-adjacent ecosystems, and joined Atlantis Computing, a startup that ultimately didn’t make it. From there he became Chief Technical Evangelist at FSLogix, a startup that did make it — all the way to a successful acquisition by Microsoft.
That acquisition brought Jim into Microsoft’s field organization as a Global Black Belt for what was then called Windows Virtual Desktop, before he moved into engineering and product design. Today he’s a Principal Product Manager for Windows Cloud, helping shape Azure Virtual Desktop and Windows 365 — including the newest addition to that family, Windows 365 for Agents, a platform purpose-built for AI agents to log in and work inside a Cloud PC. Alongside his day job, Jim is a well-known voice in the EUC community: sharing on GitHub, YouTube, and LinkedIn, and speaking regularly at community events and forums.

Setting the Stage

Here’s why this conversation matters right now: most people still picture “AI agents” as a slightly smarter chatbot. Jim’s argument — and the thread that runs through this whole episode — is that the real shift is agents that can operate a full desktop environment: clicking, typing, navigating legacy apps without APIs, and doing it all under the same governance, identity, and compliance controls organizations already use for human employees.
In this post, we’ll walk through the highlights of the conversation, take a deeper look at how Windows 365 for Agents actually works under the hood, share the real-world stories Jim brought to the table (including one seriously satisfying “3,500 lines of code down to 12” moment), and wrap up with the practical takeaways for anyone in IT — especially our fellow EUC professionals — wondering where to even start.

Episode Highlights

Moment one: “Your workflow isn’t broken, it’s waiting for an upgrade.”
Gerjon opened by referencing a recent LinkedIn post from Jim, and Jim didn’t hold back on how much that idea has played out in his own working life. In his words, combining agentic AI with a personal “Second Brain” system over the last three months has completely changed how he works — and he’s seen the same reaction from everyone he’s introduced it to. As he put it: it’s a combination of two things — second brain, and agentic AI — and together they’re a genuinely different way of working, not just a faster version of the old one.

Moment two: the identity question.
One of the sharpest turns in the conversation came when Gerjon asked a deceptively simple question: does an agent running on a Cloud PC get its own identity, or does it borrow yours? Jim’s answer — that each agent gets its own Microsoft Entra ID account, fully governed by existing tenant policy — reframed the entire rest of the discussion. Suddenly this isn’t “AI running loose on a VM,” it’s a new type of managed identity that IT already knows how to govern.

Deep Dive: Inside Windows 365 for Agents
So how does this actually work? Jim walked us through the mechanics in detail, and it’s worth unpacking properly.
A Cloud PC, but for agents, not humans. Windows 365 for Agents is a version of Windows 365 designed specifically for an AI agent — not a person — to log into and do work inside. If you believe agentic AI is coming to the enterprise (Jim clearly does), then you need some kind of compute substrate for those agents to run on. If that agent needs to control a Windows environment, Windows 365 is a natural fit, because — human or agent — you still need the same governance, security, auditability, compliance, and scalability.
Its own identity, its own accountability. Every agent gets its own Entra ID account. That means every permission, restriction, and policy you’d apply to a human user in your tenant can be applied to an agent too — and every action is tracked under that identity, not yours.
Full observability, by design. This is one of the more interesting inversions Jim pointed out: when a human is on a machine, you generally shouldn’t be watching everything they do. When an agent is on the machine, you absolutely should. Windows 365 for Agents ships with a live WebRTC feed so you can watch an agent work in real time — useful for spotting inefficiencies in its first few runs, and for stepping in as a human at any point (at which point, notably, the recording stops, because now a person is on the machine).
Marionetting the desktop. Underneath, there’s an MCP server sitting in front of the Cloud PC. An agent — first-party or third-party — talks to that MCP, which controls mouse, keyboard, and the accessibility tree of the UI. In practice, that means anything a human user could do in an app, browser, or the OS itself, an agent can now do too — including in legacy applications that were never built with an API in mind.
Why not just use an MCP for everything? Jim was refreshingly candid here: he loves MCPs and uses them constantly, but they’re not always the right tool. Sometimes an MCP doesn’t exist for a SaaS platform, sometimes it exists but is open source with unclear provenance, and sometimes you just need one thing, one time, from one website — and installing a whole MCP just to do that eats up your context window for no lasting benefit. His rule of thumb: if you have a reliable, accurate, fast, and observable “computer use” agent, a lot of those occasional MCP use cases simply aren’t necessary anymore.
Consumption-based pricing. For the first time on Windows 365, agent Cloud PCs run on a pure pay-per-hour consumption model — you pay for usage, not for deployment or cleanup time. VMs can be pre-provisioned for faster startup (at a small monthly storage cost) or spun up and torn down entirely on demand.

Real-Life Stories & Examples

The story that best captured all of this in one breath: Jim used to be the PM for App Attach, and years ago he built a PowerShell-based pipeline to automate packaging traditional apps into MSIX format — spinning up a VM, detecting new app versions, finding the silent install, capturing and certifying the package, then tearing the VM down again. Written entirely by hand, before AI coding assistance existed, that pipeline ran to roughly 3,500 lines of PowerShell.
His agentic replacement, demoed on LinkedIn? About 12 lines of plain-language instructions — with the same output quality, and producing both an Intune Win32 app and an MSIX package. And, as Jim pointed out, it’s arguably more resilient than the original script: a 3,500-line script is brittle by nature — any one broken step in a long deterministic chain fails the whole thing. An agent instruction set isn’t deterministic in the same way, so when something changes (a tool update, an API tweak), the agent has a real chance of recovering and adapting on its own, rather than simply breaking.
On the cost side, Jim shared numbers most people never get to see. Even though Microsoft employees have effectively free access to tokens, he went and calculated what it would genuinely cost an average person to live inside a “Second Brain” workflow every working day of the month. His answer: roughly £80–90 a month (about €95–105 or $110–120) — kept low because around 95% of his 216 million monthly tokens are cached, and cached tokens are dramatically cheaper than uncached ones.

Key Takeaways

Agentic AI needs a compute substrate, and for anything touching a Windows UI, Windows 365 for Agents gives IT the same governance, identity, and auditability tooling they already use for human users.
Agents get their own Entra ID identity — meaning existing tenant policy, permissions, and monitoring apply directly, with no special-case security model required.
“Computer use” (mouse, keyboard, accessibility tree) fills the gaps MCPs can’t — especially for legacy apps without reliable APIs, or one-off tasks not worth building an MCP integration for.
Start with your Second Brain, not the agentic part. Getting instructions, context, audience, and constraints right up front dramatically reduces hallucination — and only once you trust that foundation should you let an agent start acting on it in the real world.
Cost is more accessible than it sounds. Thanks to caching, a genuinely heavy daily Second Brain workflow can land around £80–90/month, not the eye-watering figure many assume.
EUC professionals are unusually well positioned for this shift. The skills required to run agents safely at scale — security, compliance, auditability, scalability, app delivery — are the same skills EUC teams have built over the last two decades. As Jim put it, if end-user computing is defined as whatever the end user uses to do their job, then agentic workflows should naturally fall under that same team’s remit.

Closing Thoughts

This was a genuinely energizing conversation to have — even without James in the (virtual) room. Jim’s mix of hands-on technical detail and grounded honesty (about what’s ready today, what’s still on the roadmap, and where MCPs are and aren’t the right tool) is exactly the kind of practical insight we love bringing to this podcast.
If there’s one thing to sit with after this episode, it’s Jim’s closing challenge to IT and EUC teams everywhere: this agentic wave is coming to your organization whether you plan for it or not, and the team best equipped to own it securely is probably already in the building. The only question is whether you raise your hand first.
James will be back behind the mic for the next episode — but we already have a feeling this won’t be the last time Windows 365 for Agents comes up on the show. Got thoughts on this episode, or questions for Jim? Drop them in the comments or reach out to us directly — we’d love to keep the conversation going.

🎧 Listen to the full episode of The Impact of AI: Explored, and don’t forget to follow Jim Moyle on LinkedIn for more on agentic AI and Windows 365.