Summary
Three months ago, I wrote about Microsoft's commercial pivot from user-centric licensing toward agent-based productivity. I called it the shift from SaaS to Agent-as-a-Service, where Microsoft bundles Microsoft 365 E5, Copilot, and agent tooling into high-growth, high-margin models.
Today, that vision is no longer speculative. It’s here.
Microsoft is shifting from licensing users to licensing AI agents as independent, billable work units.
With the rollout of Entra Agent ID, Copilot Studio tuning, and multi-agent orchestration, Microsoft has launched the technical foundation for what will soon become the most significant licensing overhaul since the original Office 365 subscription model.
We're looking at far more than a simple Copilot upgrade. Microsoft is building a licensing architecture for a world where digital agents are treated like employees. Each one gets their own ID, access controls, usage telemetry, and yes... licensing costs.
What’s Changed Since My Last Post?
In my original article, I asked:
“What happens when agents become independent work units with measurable productivity? Will Microsoft charge per agent, per interaction, or bundle it all under Copilot?”
Microsoft just answered:
Each agent gets an Entra ID – treated like a full-fledged identity.
Managed via Intune, Purview, and Conditional Access – same as human users.
Metered agent usage analytics in the Copilot Control System – enabling billable units.
Embedded knowledge sources decoupled from user data – allowing agents to work independently of users.
In short: agents have evolved beyond simple assistants. They're digital workers. And they'll be licensed as such.
The Emerging Agent Pricing Landscape
Microsoft hasn’t published formal agent pricing yet, but here’s what we’re seeing across early previews, partner conversations, and analogs in Power Platform and Azure OpenAI:
Model | Estimated Pricing | Examples |
---|---|---|
Per-Agent License (Core) | $15–$25 per month per agent | Internal Q&A bots, basic workflow copilots |
Per-Agent License (Advanced) | $25–$40 per agent/month | Embedded knowledge agents, cross-app workflows (e.g. Finance or HR Copilot) |
Autonomous Agent + Consumption | $40+ base + usage fees (tokens/API/calls) | Fully autonomous agents orchestrating across SAP, Dynamics, Salesforce |
Unattended RPA (baseline) | $150/month (Power Automate unattended bot) | Microsoft’s closest precedent |
Consumption (PAYG) | $0.01–$0.05 per 1,000 tokens or API call | Expected for GPT5 powered agents, heavy LLM workloads |
Pack-Based | 10 Agent Pack for $250/month (hypothetical CSP tier) | Targeted at small teams or CSPs |
Scenario Example: Let’s say a multinational deploys 50 Copilot agents across marketing, legal, HR, and customer service.
30 Core agents: $20/month = $600
15 Advanced agents: $35/month = $525
5 Autonomous agents: $50/month base + $200 usage = $450
Monthly Agent Cost: $1,575 Annualized: ~$18,900 – for only 50 digital workers
Agent-based spend could match or exceed human Copilot licensing within 2–3 years.
Scale this across 1,000 agents in a large enterprise, and agent-based spend could rival or exceed human Copilot licensing within 2–3 years.
🖐 Gain insight into emerging agent pricing models. Learn more: Pricing Research and Pricing Metrics.
Why Entra Agent ID Changes the Game
By assigning agents a secure, traceable identity:
Microsoft can track licensable activity per agent.
Customers can govern non-human access across Zero Trust frameworks.
Vendors (and partners) can now bill agents as operational units, creating a viable monetisation model.
The system also enables granular license enforcement:
Want 10 marketing copilots? You’ll need 10 Agent IDs.
Agents without ID won’t be allowed to operate, access data, or be monitored.
🖐 Control and optimise your Microsoft AI and agent costs. Learn more: Microsoft Azure Cloud Cost Optimisation.
Embedded Knowledge = Decoupled Licensing
In the past, Copilot agents were limited to the data that users had permission to access.
Now, with SharePoint Embedded containers, organisations can upload proprietary documents and datasets directly into the agent, removing dependency on user entitlements.
Embedded knowledge lets agents operate independently of user entitlements, creating new licensing revenue streams.
The result:
Agents become products in and of themselves.
ISVs and consulting firms can sell packaged copilots with embedded knowledge (e.g., a “GDPR Legal Copilot” or “Supply Chain Agent”) to customers.
These embedded agents will carry independent licensing costs, separate from Microsoft 365 or E5.
Microsoft’s New Licensing Stack
Here's how I see Microsoft's new licensing pyramid moving to Average Revenue Per AI Agent (ARPAA):
Base Layer (User Tier)
M365 E3/E5 – Traditional Average Revenue Per User (ARPU)
Copilot – $30/user/month
Viva, Intune, Entra ID Suite – $6–$9/user/month add-ons
Mid Layer (Security & Governance Prerequisites)
Purview Premium, Defender, Intune – Required for secure agent deployment
Entra Agent ID – Controls and tracks agent identities
Top Layer (Agent Licensing)
Core / Advanced / Autonomous agent licenses
Consumption-based billing (API calls, token usage, etc.)
Agent Packs or vertical-specific licenses
What's Next? The E7 Debate and Agent Monetisation Path
Many are asking: Will Microsoft finally introduce Microsoft 365 E7 to simplify this complexity?
Short answer: Not yet. The current fragmented licensing landscape makes more revenue.
But if customer pushback intensifies, or Google/Anthropic begin bundling Gemini-style copilots into productivity suites, E7 could become Microsoft's bundling solution.
Until then:
Expect tiered agent SKUs to emerge by early 2026.
Hybrid pricing (per-agent + consumption) to become the standard.
A new licensing economy centered on ARPA as the metric that matters.
Advice to Procurement, IT, and FinOps Leaders
Agents are coming, whether you're ready or not. Start with limited PoCs, tune your licensing, and establish KPIs for automation ROI.
Security equals licensing.
The more agents you deploy, the more critical Microsoft’s security and governance stack becomes. That stack isn’t optional—it’s bundled licensing leverage.
Plan for hybrid costs.
Agent licensing will include fixed and variable costs. Budgeting must be dynamic and closely monitored. 🖐 Control and optimise your Microsoft AI and agent costs. Learn more: Microsoft Azure Cloud Cost Optimisation.
Push for transparency. Demand clear pricing from Microsoft and partners. Don’t let usage-based models creep up without FinOps oversight.
Stay strategic.
Don’t let Microsoft’s evolving SKUs dictate your AI roadmap. Design your Copilot/agent strategy first, and then negotiate the licensing.
My Advice
Microsoft’s commercial model is shifting from licensing people to licensing productivity. In the next two years, organisations will negotiate enterprise agreements that measure both ARPU and ARPAA, where every “worker” (human or digital) carries a price tag and a license key.
Ignore this at your peril. Prepare for it, and you can not only control your spending but also harness it for real advantage.
Agent monetisation has moved beyond theory. Your next negotiation will involve it.
🖐 Prepare for your next EA negotiation in the agent licensing era. Learn more: Microsoft Enterprise Agreement Negotiation.
— Daryl Ullman
Partner & Lead Negotiator, SAMexpert | Microsoft Licensing | Cloud Economics | Copilot Strategy