A complete guide to Agent 365

Agent 365 is Microsoft’s new control plane enabling organisations to manage AI agents as governed, auditable and secure entities.

Margarita Redknap

Margarita Redknap

|

Marketing Manager

Posted on: June 10, 2026

|

Last updated: August 6, 2026

|

8 min read

Margarita Redknap

Margarita Redknap

|

Marketing Manager

Posted on: June 10, 2026

|

Last updated: August 6, 2026

Read time: 22 minutes

Agent 365 banner

The looming challenge of agents at scale

We already know that the scale of the expected AI agent population is enormous. In 2025, it was estimated that enterprises had 28.6 million active AI agents, and this number is forecasted to increase to over 2.2 billion by 2030.

Statista 2026 - Number of active artificial intelligence (AI) agents in enterprises worldwide from 2025 to 2030

Source: Statista 2026 – Number of active artificial intelligence (AI) agents in enterprises worldwide from 2025 to 2030

Microsoft is rapidly scaling its AI workforce, with an estimated growth rate of 5,000 agents per week. It was previously speculated that the number of AI agents would exceed its number of employees by April 2026. Last month at the Microsoft 365 Community Conference, Ryan Cunningham, Corporate Vice President of Microsoft Power Platform, announced that Microsoft is currently running half a million agents, while the company has a workforce of approximately 200,000 employees.

While perhaps not as quick as Microsoft, nearly every organisation in the world is going to follow this explosion of agents. Internally, we have discussed the scale of agent adoption, and we think it’s important to consider some big upcoming challenges.

In our experience, organisations often don’t know what agents they have, which limits the ability for them to be used effectively. Furthermore, it limits the ability for the organisation to ensure the agent is fit for purpose and providing value. And most importantly, you can’t secure what you can’t see. How do I know these agents are operating with least privilege? How do I know they’re operating within the guardrails? What data are they accessing, and what are they sharing?

Parallels between human employees and agents

Let’s forget about agents for a second and focus on human employees and what they need to work securely. At the base level, employees have an identity in their Entra tenant (formerly Azure AD). This is their digital representation for security, collaboration, and system access.

Beyond identity, employees interact with massive amounts of data. Organisations use Microsoft Purview to discover, classify, and protect that data. Purview allows us to apply labels and data loss prevention (DLP) to control how data is used.

Finally, we have threat protection through Microsoft Defender to protect against bad actors, and productivity tools like Exchange, SharePoint, and Teams to help them collaborate.

Now, what about our agents? Well, they have the same requirements as human employees.

However, there will be changes in the specific details of how those requirements are met. This is where Agent 365 comes in.

Microsoft Agent 365 provides the ability to observe, govern, and secure the growing number of agents within organisations.

Agent 365 takes the existing ways of meeting requirements (Entra, Purview, Defender, and Microsoft 365) and extends them to your AI agents. Let’s dive into this.

Agent 365 Dashboard

A new type of identity for agents

Let’s start with identity. You might want to use service principals for agents just like for applications, but for agents, they are too limited. Setting up a regular user account, is also not practical because we have things like Multi-Factor Authentication and passkeys.

The solution: Agent ID, where every single instance of an agent will have its own unique identity.

How to set up the Agent ID

Instead of creating every identity from scratch, we create a Blueprint. Think of this as a template for the Agent ID.

  • Roles & permissions: The blueprint defines the roles. Since an agent might work on its own, it needs app-level API permissions. But it also needs delegated permissions so it can act on my behalf as an assistant.
  • The safeguard: There are certain permissions you simply cannot give an agent. For example, you can’t give an agent User.ReadWrite.All. We don’t want an agent going off the rails and deciding to delete every user account in the company for example.
  • Authentication: This is a crucial point, the Blueprint authenticates, the Agent ID does not. The Agent IDs we create don’t have their own credentials. Instead, the platform running the agent (like Copilot Studio or Microsoft Foundry) authenticates via the Blueprint using Federation or Certificates.

Once the blueprint is in place, it has the permission to create actual Agent IDs. These IDs inherit the permissions from the blueprint, but you can also add specific permissions to a single Agent ID.

Note: Every Agent ID must have a Sponsor to avoid the problem where nobody knows who an agent belongs to or if it’s still needed. The Sponsor is the person with accountability. They decide if the agent is still providing value or if it should be retired. You can also add technical owners to manage the settings, but the sponsor ensures we don’t end up with agent sprawl.

While platforms like Copilot Studio handle some of this automatically, enterprise-grade governance requires a strategy. Intelogy can help you design and implement these Blueprints, ensuring your agent ecosystem is secure, compliant, and aligned with your organisation.

By establishing a robust Blueprint architecture, we help you apply a governed and scalable AI workforce.

What is the Agent User?

While a standard Agent ID works for background tasks, it has limitations because it can’t natively hold an Exchange mailbox, join a Teams meeting, or be @mentioned in a SharePoint document.

To bridge this gap, we can extend this identity into an Agent User.

Moving into this collaborative space is where the Agent 365 licence becomes essential. The licence is assigned per user to the people who own, sponsor, or are served by the agent (not to the agents themselves). A single user’s Agent 365 licence covers all the agents acting on their behalf, enabling them to interact with Exchange, Teams, and the broader Microsoft Graph.

Note: While Agent 365 provides the governance framework for these identities today, the specific ‘Agentic User’ functionality, including autonomous mailbox interaction and native Teams presence, is currently available via the Frontier preview programme. We are actively working with clients to prepare their environments for these capabilities, ensuring that as they move from preview to general availability, your human-agent collaboration remains natural, secure, and fully audited.

Microsoft Entra for agents

Agent 365 allows us to take the security capabilities we have for human employees and apply them to our agents.

Because agents now have an identity, you can manage them with the same precision as a human user:

  • Role-based access control: Assign specific API permissions and access to Azure resources.
  • Entitlement management: Grant access packages and manage ownership.
  • Conditional access: You can now target ‘All Agents’ or specific Agent IDs with Conditional access policies.

Entra also understands that agent behaviour is different from humans. For example, if a human is active 24 hours a day, Entra flags it as a risk (likely account sharing). However, if an agent works 24 hours a day, Entra knows that’s expected.

It does still watch for true anomalies such as a risky user interacting with an agent or abnormal sign-in frequencies and can automatically trigger blocks or additional requirements based on those signals.

Through Entra you also get a full audit of agent activity like authentication logs.

If you need help configuring your agents so that they are secure by design don’t hesitate to speak to us.

Microsoft Purview for agents

As agents interact with your data, Microsoft Purview acts as the lens through which every interaction is inspected, governed, and protected. Just as it does for your employees, Purview ensures that your most sensitive information remains secure.

Within Agent 365, Purview has been specifically adapted to understand how agents work.

  • Context monitoring: If a human employee accesses 500 documents in five minutes, it’s often a sign of data exfiltration. If an AI agent scans 500 documents in five minutes, it’s likely just performing a standard information retrieval task.
  • Identifying intent: Purview can identify malicious intent vs high-speed productivity, ensuring your security alerts stay relevant.

Purview also actively enforces your organisational guardrails. For example, if data is labelled as “Confidential” or “Do Not Use for AI Grounding,” Purview ensures your agents respect those boundaries. If an agent attempts to access data outside of its permitted scope or triggers a data loss prevention policy, Purview can block the interaction.

We work with clients to audit their existing Purview setup ensuring they are AI-ready. By configuring Purview to specifically monitor agentic behaviour, you can confidently use agents without compromising compliance or data privacy. We ensure that your agents only know what they are supposed to know and nothing more.

Microsoft Defender for agents

Just as it protects your human workforce from malware and phishing, Microsoft Defender now provides specialised AI Security Posture Management and runtime protection for your agents.

Defender works proactively to ensure your agents aren’t created with security gaps. It helps you:

  • Discover vulnerabilities: Automatically identify misconfigured agents or those with excessive permissions that could be exploited.
  • Real-time blocking: Through AI Runtime Protection, Defender monitors agent actions across different environments and platforms. It uses observability logs and webhooks to detect and block unintended or malicious actions as they happen.

If a threat is detected, Defender doesn’t just block it, it helps you understand the full attack chain. Through the Defender portal, you can trace exactly what an agent was doing, including its chain of thought and you can see how different agents interacted leading up to an incident.

Work IQ for agents

The goal of Agent 365 is not just to restrict or govern. It takes the entire suite of enterprise capabilities we provide for human employees (security, identity, and data protection), and tailors them specifically for the unique behaviours of agents.

By establishing this framework, your agents are safely cleared to access the most powerful layer of the Microsoft ecosystem: Work IQ.

Just as a human employee builds a mental map of their workplace over time, Work IQ provides your agents with an advanced organisational context. It gives them:

  • A digital memory: The ability to understand the complex relationships between data, projects, and people.
  • Contextual awareness: Instead of looking at documents in isolation, the agent analyses messages, meetings, transcripts, and emails to build a rich mental graph.
  • Proactive capabilities: By understanding how your organisation functions, the agent can stop being reactive and start anticipating your business needs.

Registry in Agent 365

As your AI ecosystem expands, keeping track of every bot, copilot, and autonomous workflow becomes a massive challenge. To solve this, Agent 365 introduces a Registry.

If Microsoft Entra is the foundation of identity, the Registry is your directory service for the AI workforce.

Note: The Registry handles discovery, not access control.

Just like the human Global Address List helps you find a colleague’s email but doesn’t dictate what files you can share with them, the Registry allows humans and agents to find each other. Access control and guardrails remain strictly the job of Entra and Purview.

Within the Registry, every agent has an Agent Card. This card outlines:

  • Inventory: A complete view of your entire agent estate so you know exactly what is running in your environment.
  • Capabilities: A clear description of what the agent does, its specific skills, and how it can interact.
  • Agent-to-agent collaboration: Armed with this information, autonomous agents can query the registry to discover other agents and collaborate to achieve complex, multi-step tasks.

One of the most powerful features of the Registry is its open nature. Any agent can be listed in the Registry, regardless of the platform it was built on.

Note: Only agents with an official Agent ID have the permission to actively query and discover other agents from the Registry.

Plans and pricing for Agent 365

Agent 365 is included with a Microsoft 365 E7 licence (inc. Teams) for £81.60/user per month and Microsoft 365 E7 (No Teams) at £75.00. You can also get a standalone licence for Agent 365 for £11.50/user per month.

Agent 365 Pricing UK

Closing

Agent 365 is Microsoft’s new control plane enabling organisations to manage AI agents as governed, auditable and secure entities. If you have any questions for us that we haven’t answered in this blog or need help with Agent 365, please don’t hesitate to contact us directly or read more about our Agent 365 Consultancy Services.

Last updated 6 Aug 2026

About the Author: Margarita Redknap

Margarita Redknap
I am a forward thinking, motivated, passionate and driven marketing manager. I always strive to do things smarter and better as I oversee execution of marketing strategies.

Table of contents

Go to Top