If you’re landing here after hearing about Agent 365 at Ignite 2025 and you’re wondering “How is this licensed?”, you’re in the right place. I’ve been gathering information about this new product initiative for months already, so that I’m ready to dissect the official MS narrative once it is published. Reading between the lines is a crucial skill to survive in the domain of Microsoft licensing and product roadmaps.
In my previous post I covered why the original per-user Copilot model was always going to hit a wall. High AI compute costs and low adoption made “$30 per person per month” an awkward fit. Now Satya Nadella has basically confirmed the next step: AI is pushing Microsoft from per user to per user and per agent.
We can hear this and more about Satya’s current thinking around the business of AI from the Dwarkesh Podcast episode that came out three days ago:
For anyone who wishes to understand how massively different the business model of Microsoft is when compared to their partner and the leading “lab” in the field, OpenAI, I recommend watching the full 1.5h talk. For this post, I’ll dive into the implications for agentic AI product strategy and how that shapes Microsoft’s commercial offering. Manifesting in the new Agent 365 product that can be considered a “wave 2” after the initial push of Microsoft 365 Copilot.
1. Satya’s new mental model: subscriptions as AI entitlements, not just seats
Classic SaaS assumed near-zero marginal cost per extra user. AI has blown that up. In Satya’s words, the COGS of AI is so high it “breaks” old SaaS models.
He reframes subscriptions as entitlements to a certain amount of AI consumption. Tiers aren’t just “Standard vs. Pro features” anymore; they’re “how much model usage you’re allowed to burn every month.”
Then comes the crucial line for licensing people:
The way to think about the per-user business is not just per user – it’s per user and per agent.
In other words:
- Humans still have Microsoft 365 user licenses.
- On top of that, AI agents become their own billable population, each with its own subscription and consumption profile.
Agent 365 is the Microsoft 365 flavour of that idea.
2. What an Agent 365 (A365) license likely includes
Microsoft hasn’t published a full spec sheet yet, but between Satya’s comments and the Message Center post about Agentic Users, the contours are clear.
2.1 A full “AI user” identity
Agentic Users are described as autonomous virtual colleagues that:
- exist as Entra ID objects (Azure AD),
- have email addresses and Teams accounts,
- appear in the org chart under a human sponsor,
- can participate in chats and meetings and work with enterprise data.
Licensing-wise, that’s more than an API key. It’s a user-like account that needs rights to:
- Exchange Online (mailbox),
- Teams,
- OneDrive/SharePoint access,
- directory features and security policies.
So an A365 license is almost certainly a bundle that includes a “user shell” plus AI capabilities, but tied to a non-human identity.
2.2 Entitlements to AI consumption
Satya’s subscription definition matters here. An A365 license is not just “turn this thing on”; it’s “give this agent the right to consume some AI capacity every month.”
You should expect:
- a baseline pool of AI usage bundled per agent (likely Copilot-style credits),
- tiers (Standard / Pro / Industry) differentiated by consumption rights and maybe model selection,
- overage billing when agents exceed their included quota, using the same metering infrastructure already in Copilot Studio and Azure AI Foundry.
In plain English: A365 gives each agent an allowance of “thinking time”. Heavy agents will push you into consumption.
2.3 A computer for the agent
Satya also called out something concrete: autonomous Office-style agents that want to provision Windows 365 so they have their own Cloud PC. He says Microsoft is already seeing strong demand for this pattern.
From a licensing standpoint, there are two likely options:
- A365 includes some right to a basic Windows 365 instance for the agent, or
- A365 plugs neatly into a separate Windows 365 subscription or usage plan dedicated to agents.
Either way, the underlying story is:
For every serious agent, you license identity + collaboration + AI + some form of compute.
That’s very different from a helpful Copilot button inside Word to summarize your document.
3. From seats to digital headcount: the sprawl problem
Once you license agents this way, your estate no longer maps cleanly to human headcount.
Today you roughly track:
- N employees → N M365 licenses (+ maybe N Copilot add-ons).
With Agent 365, the numbers can diverge:
- 500 employees,
- 60 frontline workers,
- 40 human managers,
- plus 120 agents (procurement agents, onboarding agents, channel agents, reporting agents, etc.).
Nothing stops that 120 from becoming 1,200 over a few years if templates in the Teams Agent Store make it trivial to “hire” a new digital colleague whenever somebody has an idea.
From a licensing perspective:
- A365 is a new licence population you’ll need to track separately from human users.
- Agents can be spun up and forgotten faster than employees, so the usual “joiner/mover/leaver” processes aren’t enough.
- And because A365 also entitles AI consumption, the cost impact of a rogue or overly busy agent is higher than an unused mailbox.
Satya explicitly expects the end-user compute business (Windows 365, device-like capacity for agents) to grow faster than the human user base. Translate that: Microsoft is planning for a world where you pay them for more agents than people.
4. Governance as part of the SKU, not an optional extra
Agent 365 is being positioned primarily as a security and governance story: “Confidently activate your AI agent estate”, as one Ignite 2025 session description puts it.
That lines up neatly with Satya’s list of what has to be “baked into” every agent:
- identity,
- security,
- observability,
- management layers.
And with Microsoft’s broader Entra Agent ID initiative, which promises unified lifecycle, permissions, and auditing for non-human identities.
Expect A365 to monetise three things at once:
- The right to run an agent at all (licence per Agentic User).
- The AI compute that the agent burns.
- The governance surface around that agent: logging, policy, DLP, eDiscovery, and sprawl visibility across tenants.
That last one is important. If you create agents outside of this framework – DIY bots on top of generic LLM APIs – you may save on licensing in the short term, but you also lose the built-in governance that your security team and regulators will start expecting. Agent 365 turns “doing this properly” into a licensable feature.
5. Where Agent 365 leaves you as a licensing owner
Strip away the branding and you get a simple framework:
- Subscriptions are now AI entitlements, not just software access.
- The Microsoft 365 business becomes per user and per agent – both are first-class, licensed identities.
- Each useful agent needs: an A365 licence, some AI headroom, and likely a slice of Windows 365 or similar compute.
- The number of agents can easily outpace human headcount, creating a new class of “digital labour sprawl”.
- The governance you actually need to stay compliant is itself bundled into the Agent 365 product.
For customers, that means:
- You’re no longer just counting people; you’re counting people + digital coworkers.
- Budget conversations have to include both seat licences and agent licences plus consumption.
- And understanding Agent 365 is now part of the core Microsoft licensing literacy, not a niche AI topic.
Once the dust from Ignite settles and the SKUs are public, we’ll have to go into the details: exact entitlements, price points, how A365 interacts with existing E-plans and Windows 365. But the direction is already clear.
If Copilot was Microsoft testing what you’d pay to make humans a bit more productive, Agent 365 is Microsoft testing what you’ll pay to put actual work onto autonomous software workers – and to keep those workers under control.
Read more of my analysis on the A365 licensing story from this earlier post:
