AI Agents • • 8 min read

Microsoft Scout: I cloned my work PC and started truly delegating

Microsoft Scout on Windows 365 with GitHub Copilot: prerequisites, how it differs from Copilot Cowork, custom MCP servers and the process-to-skill method.

Last updated on

Microsoft Scout running on a Windows 365 Cloud PC as an enterprise AI agent
Table of Contents

Right now Microsoft Scout is one of the AI tools giving me the most satisfaction, full stop. It was announced at Build this year and is available to Frontier program participants. In short, Microsoft took an idea that already existed, agents like OpenClaw or Hermes Agent, and is turning it into an enterprise solution. Here’s how I use it every day on a Windows 365 Cloud PC, and why I think it genuinely changes the way you work.

What Microsoft Scout is

Scout is essentially Microsoft’s OpenClaw, or Microsoft’s Hermes Agent. The way I see it, the goal is to create a real second operator: one that works exactly like me and can reach everything I give it access to on my work computer.

For me that means proposals, documents, research and advanced troubleshooting, all done with the tools already installed on my PC.

My setup: a 1:1 copy of my workstation on Windows 365

If you’ve seen my previous videos, you know that for personal use I built an always-on agent on Azure Virtual Desktop: a Windows 11 machine running the GitHub Copilot CLI and Hermes Agent, with GitHub as the model router. That approach actually came to me from looking at how Scout works.

For work, I replicated the same pattern the enterprise way:

  • I have a Windows 365 license, so a Cloud PC assigned to me;
  • on that Windows 11 machine I installed my entire work suite and company tools;
  • I installed Microsoft Scout and, as the documentation requires, connected it to my GitHub Copilot Business license.

The result is a one-to-one copy of my workstation, reachable at any time of day.

Processes first, agent second

Before giving Scout any instructions, I thought about processes: what do I do most often that adds little value? Writing documents, checking the CRM, checking Power BI dashboards, monitoring a specific mailbox. I described each of these activities, turned it into a skill, a recipe the model can reproduce, and assigned it to Scout.

Microsoft Scout prerequisites

According to Microsoft, you need:

  • Windows 11 or macOS Monterey;
  • a Microsoft 365 Work or School account (Personal accounts don’t work today, which makes sense to me);
  • a Microsoft 365 Copilot license assigned to your account;
  • the latest Microsoft Visual C++ Redistributable (Windows only);
  • admin rights to install Scout, plus an Intune configuration (the documentation covers all of this in depth);
  • a GitHub Copilot Business or Enterprise license.
Microsoft Scout prerequisites page on Microsoft Learn
Figure 1: Microsoft Scout prerequisites in the official Microsoft Learn documentation

Licensing-wise, the bar is high. But I’m not selling you anything: this tool can really change how you work. Once the prerequisites are in place and Intune enablement is done, find the download on the web, install it and launch it.

Scout vs Copilot Cowork: models and custom MCP servers

What Scout suggests at startup, Copilot Cowork can do too. So why pick Scout? For me the answer is simple, and it comes down to two things.

1. I choose the model

Scout uses GitHub Copilot, just like OpenClaw and Hermes Agent do, to reach the models. So I get everything GitHub Copilot serves: from Claude Opus 4.7 to Claude Opus 5, which I consider one of the best models today in terms of performance and token cost; the whole GPT-5.6 family, including GPT-5.6 Luna, very handy for tasks like document review; GPT-6 Astra, released a week ago and genuinely impressive; plus Grok, Gemini, DeepSeek and MAI, Microsoft’s own models. I pick the model, not the platform on my behalf.

Microsoft Scout model picker with Claude Opus 5, Claude Sonnet 5 and GPT-5.5
Figure 2: Scout’s model picker: I decide which model to use for each task

2. MCP servers, including custom ones

This is the real game changer. The usual objection is that Cowork, being part of the Copilot suite, pulls user context through WorkIQ. True. But in Scout, besides the skills I’ve built for my day-to-day, I can add MCP servers, including custom ones. With Cowork I’m forced through plugins, meaning a store. That’s tight corporate control, sure, but I want the freedom to integrate with different tools.

MCP is a driver to an external source: it gives the model context and visibility so its answers are more precise, the famous harness everyone talks about. I want an MCP for SAP, SQL Server, Azure, Foundry, Fabric (which Cowork actually already has), or for SharePoint, OneDrive and Teams to rebuild that WorkIQ-style context. Then I can ask: “I’ve had 10 meetings with this customer, there are transcripts, 10 emails and 10 Teams messages: draft a note for the next meeting and email it to me.”

Microsoft Scout Extensions section with the configured MCP servers
Figure 3: Scout’s Extensions section with the configured MCP servers (account details blurred)

Scout vs Copilot Cowork at a glance

Microsoft ScoutCopilot Cowork
Model choiceEvery model served by GitHub Copilot, my choiceThe platform decides
ExtensionsSkills and MCP servers, including custom onesStore plugins
User contextVia MCP (SharePoint, OneDrive, Teams…)Built in through WorkIQ
ApproachMore integration freedomMore corporate control

Demo: Scout browsing like a human

I can’t show my real chats in the video: it’s a production instance with data better kept private. So here’s an example instead. I pick a lighter, fast model, GPT-5.6 Luna, and ask Scout to open an online newspaper (ilfattoquotidiano.it) and give me a press review in the chat, 30 lines max, grouped by topic. I asked for Chrome, it opened Edge, probably because it’s the default browser; it went through the news and returned the review.

Topic-sorted news digest generated by Microsoft Scout with GPT-5.6 Luna
Figure 4: The topic-sorted news digest, returned by Scout right in the chat

The point isn’t the press review: it’s that Scout can operate like a human. If I lose fifteen minutes every morning reading data scattered across two or three systems that don’t talk to each other, I can tell it: open the browser, grab these ten metrics, these ten KPIs, and email me a summary plus a static HTML dashboard I can browse directly. The principle is simple: delegate to Scout the mechanical work you don’t feel like doing.

And it all runs on my virtual machine: I can kick off tasks from a Teams chat or from Windows 365, on my phone, PC, Mac or iPad. For me, that’s the real game changer: an agent stacking solution supported and organized directly by Microsoft, with all the enterprise policies.

Results and method: from processes to skills

Today we can genuinely clone our PC, and with it a part of ourselves, to delegate work. In my experience I got a measurable 3x in performance, and I track my results very closely. As a hypothetical presales example: if I used to deliver 5 documents a week, with this tool I can do 15. Why? Because we defined the processes, the processes became skills, and the skills became recipes for the various LLMs to get the job done.

The three-step method: start from processes, turn them into skills, assign them to Scout.

Today a freelancer, a small business or a sole trader can compete on productivity, and on quality too, with much larger companies, simply by using these tools the right way.

Key takeaways

  • Microsoft Scout is an enterprise agent in the OpenClaw/Hermes Agent mold, available to Frontier program participants.
  • Prerequisites are demanding: Microsoft 365 Copilot, GitHub Copilot Business/Enterprise, a Work or School account, Intune and admin rights.
  • On Windows 365 you get a 1:1 copy of your workstation, always reachable, including from Teams and mobile.
  • Compared with Copilot Cowork: you choose the model and can use custom MCP servers instead of store-only plugins.
  • The method: start from processes, turn them into skills, assign them to Scout.

Watch the video and let’s talk

The full demo is in the YouTube video (in Italian), and you’ll find more on my YouTube page. If you want to talk about it, reach me on LinkedIn or check my talks on Sessionize; there’s more about me here. No fluff: jump on this train.

Luca Torresi

Written by

Luca Torresi Microsoft Azure MVP

Cloud Solution Architect and Microsoft Azure MVP: landing zones, networking, security and hybrid cloud from the field.