Following Microsoft Build earlier in June, you’ve probably heard someone say Copilot, Cowork and Scout all in the same sentence. And honestly, who can keep up. Microsoft seems to be releasing new AI tools on a monthly basis and I am loving it.

By the end of this post, you’ll know exactly what Microsoft Scout is, how it’s different from other Microsoft AI tools, what it can do, the governance decisions required before turning it on and the settings your IT team need to lock down before testing it.

Microsoft Scout is the first of the Microsoft autopilots. It’s an agent that can do things for you without you having to interact.

It runs continuously, carries its own identity, and acts across your Microsoft 365 stack, watching the same email, chat, calendar, and files that you work in. You can talk to Scout through a chat window or through Teams, but the key here is autonomous. This agent can act on its own to clear up the small, repetitive tasks that eat away at your day without you having to ask for it. It’s like a personal assistant that never sleeps.

Scout runs as a desktop application installed directly on your Windows PC or Mac. So it’s not a web app like Microsoft 365 Copilot in Word or Teams. Because it operates locally, Scout doesn’t just read cloud data but can interact with local files on your machine, launch your local web browser, and execute commands in your device’s shell. This is exactly what gives Scout its power, but it’s also why it requires a whole new layer of device-level control.

Microsoft Scout is currently only available through Microsoft’s Frontier preview programme.

Microsoft Scout examples:

• Keep track of my inbox and Microsoft Teams for any outstanding decisions that are waiting on me.
• Keep track of my Outlook calendar and help me prepare for every upcoming meeting.

Let me try and help you wrap your head around all the different AI layers that Microsoft has released.

Layer 1: Microsoft 365 Copilot

• Driver: You. You provide the prompt, and it responds right inside the apps you already use.
• Characteristic: Copilot does nothing until you ask it.
• Best for: Things you need help with right now in an instant, on-demand way (e.g. summarise, draft or prepare).

Layer 2: Copilot Cowork

• Driver: You delegate, then Cowork executes.
• Characteristic: The work moves away from you. Unlike Copilot, you don’t prompt and wait for a response. You delegate a whole multi-step task, walk away, and it returns when the work is complete.
• Best for: Multi-step business processes where you don’t just want a single reply.

Layer 3: Agents

• Driver: Human-triggered or event-triggered structured processes.
• Characteristic: M365 Copilot and Copilot Cowork act for one person across one task. An agent works fundamentally different. Once you create an agent you can scope it to a job or domain and your whole team can reuse it.
• Best for: Automating specific portions of a process to provide consistent, structured outputs.

Layer 4: Microsoft Scout

• Driver: The AI.
• Characteristic: Always-on and running. It works proactively rather than reactively off a prompt. It can even hand off work to sub-agents.
• Best for: Background autonomy, proactive assistance, and automated execution.

Summary of Microsoft’s 4 AI layers:

As I mentioned earlier, Microsoft Scout is currently only available through Microsoft’s Frontier preview programme. Microsoft is targeting General Availability around October this year, presumably to coincide with their annual Ignite conference, but this may slip into early 2027.

At that stage, before you can even consider enabling Microsoft Scout, you have to ask the following qualifying questions:

Even if the answer to all these questions are yes, it’s important to note that by enabling Scout, you are signing off on organisations data leaving the standard Microsoft governance boundary to interact with Microsoft Scout and GitHub services. If your organisation is not comfortable with this data leaving your governance boundary, you shouldn’t enable Scout, at least not for now.

In short, organisations should only enable Scout if all three of the following conditions are met:

1. Your organisation is already enrolled in the Microsoft Frontier program.
2. You hold meet the required license requirements and have budget controls in place to prevent unexpected costs.
3. You have a dedicated governance plan already staffed (a named individual responsible for monitoring the agent, managing autonomy settings, and reviewing audit logs).

For most organisations, the answer today is to wait. However, if you are interested in Scout, you should familiarise yourself with all these prerequisites, licensing models, and governance guardrails now.

Microsoft Scout can work proactively in the background on your behalf so it’s critical to establish the right governance and security controls. These can be managed through Microsoft Intune, Entra and Purview. Treat this as a governance project first.

Control what Scout can do autonomously

Ask yourself how much autonomy you’re comfortable giving. We would recommend that you start with a human-in-the-loop approach, ensuring actions are reviewed and approved before they are carried out by the agent.

Define where Scout can work

Scout can access information across Microsoft 365 and other connected systems. We recommend starting a pilot with clearly defined boundaries, so the agent only operates within approved workspaces.

Protect your data

As you begin to pilot autonomous agents, you want to make sure you can control how those agents interact with your (sensitive) business data and applications. You need to understand exactly what information Scout can access and which actions (e.g. read versus write) it can perform.

Implement ownership and accountability

Autonomous agents require ongoing monitoring. You should identify who is responsible for governance, reviewing activity, managing permissions and ensuring Scout continues to operate within agreed policies.

Start by running a small pilot. Don’t run it across the whole company and maybe not even your whole IT team. Use one or two test users and one or two test devices with people who can provide direct feedback so you can fine-tune your policies.

Make sure you consider the 4 governance controls mentioned above.

Currently, we are testing Microsoft Scout in our own environment. Setting up Intune policies, configuring Entra ID identities, and establishing controls isn’t just about tweaking settings. You need to protect your organisation’s data boundary while unlocking enough for Scout to actually bring value.

Here is how we can guide you through every stage of agent readiness and deployment:

• Tenant & security readiness assessments: We can assess your existing Microsoft 365 permissions, Intune policies, and SharePoint architecture before enabling autonomous agents.
• Intune & guardrails: We can help you setting up exact ADMX templates, workspace containment, and permission restrictions tailored to your risk model.
• Pilot run: We can support a small-scale, controlled pilot with clear human-approval workflows and oversight.
• Governance: We can set up ongoing identity enforcement, licensing optimisation, budget controls and agent monitoring through Microsoft Purview and Agent 365.

If you’re preparing to pilot Microsoft Scout, AI agents or want help with your security guardrails ahead of any AI deployments please reach out to our team to discuss a review of your environment.

As Microsoft has not confirmed a final commercial licensing model for Scout, we don’t yet know whether it will require a separate licence. Our current view is that like Copilot Cowork, Microsoft will opt for a consumptive model for Scout with the costs coming from the combination of baseline licensing requirements (e.g. Microsoft 365 Copilot + GitHub Copilot) combined with metered AI credit costs.