We're at an interesting inflection point with agents.
On one hand, "assistant agents" are already changing how work gets done. A sophisticated assistant that helps you write, summarise, analyse, plan, and reason — right in the flow of work — is genuinely transformative. It amplifies human capability, and it should be celebrated. Most organisations haven't even fully absorbed what's possible there yet.
On the other hand, there's a different class of agent emerging — one that doesn't just help you do the work, but can be delegated work and then reliably complete it under governance. That's the difference between "AI as a feature" and "AI as labour". And it's exactly where Agent 365 becomes strategically important — especially the part people find odd at first: licensing an agent.
The scenario: every client team has a digital employee
Imagine every client team has a digital employee called "Meeting Agent". The team can invite it to their client meetings. It listens, and produces very specific notes — sentiment, a sentiment timeline, products and funds mentioned, the sentiment associated with each, and a proposed next best action. Then — crucially — upon explicit confirmation from the meeting organiser, it performs the Salesforce updates automatically.
Now you have structured data, consistently captured, immediately usable for follow-up and analysis — without turning every meeting into manual admin work. That's not just "meeting summary". That's operationalised workflow. And it's exactly the kind of scenario where a basic "bot running under a user identity" starts to break down.
The hard truth: most agents are still "user-shaped"
Most agents today run under a human user's identity (impersonation) or behind a generic application identity (service principal / managed identity). Both can work. Both can be governed to an extent. But they hit limits quickly when the agent needs to be persistent across time, delegated work by multiple humans, clearly accountable, bounded by access that doesn't inherit one person's role, and able to participate naturally in the workplace — with enterprise-grade monitoring and policy.
Digital employee ≠ assistant
A digital employee is not "better AI". It's a different operating model. It's what you create when you want an agent to have continuity, own tasks, be addressable as an actor in the organisation, and be governed like one. It's less about intelligence and more about trust, accountability, and enterprise integration.
Blueprint → Instance → Agentic User Identity
In the Agent 365 identity model you don't just have "an agent". You have an agent blueprint (the contract), one or more agent instances (the deployed units), and an agentic user identity (the runtime identity — the "digital employee"). The runtime identity is what lets the agent be invited to meetings, respond to email and Teams, answer @mentions, be governed through identity and security controls, and obtain the right access tokens to use tools and data through a governed path.
So why license a digital employee?
Here's the cleanest way I think about it: you license the agent when it stops being "a capability" and starts being "a colleague". Not because humans are being replaced, and not because assistants aren't valuable — but because organisational actors create organisational risk, and require organisational controls: identity lifecycle, explicit access boundaries, consistent auditability, predictable workflow integration, and a standardised path to tools and data.
In regulated industries especially, this is the difference between "we tried some agent stuff" and "we can safely operationalise digital labour at scale." The point is not that everything must be a digital employee — it's that when you do want one, you need the identity and governance model to match. That's what I believe Microsoft is making real with Agent 365.