Microsoft Agent 365 Launched. Here's the Three-Invoice Trap Nobody Warned You About.

We counted three separate line items before a single agent in Microsoft’s new stack does anything useful. Agent 365 launched May 1 at $15/user/month — and if that’s the only number your finance team is watching, they are already wrong.

Every tech roundup leads with the headline governance price and stops there. Agent 365’s $15 standalone seat fee is honest. But it’s not the bill. It’s the first bill of three, and the other two are where the math gets ugly.

Invoice One: Agent 365 Governance ($15/user/mo)

The $15 per user buys you Agent 365 governance — the control plane. Registry. Lifecycle management. Permissions. Audit logs. Role-based oversight across your agent fleet. It’s what IT admins and security leaders buy.

If you already license the Microsoft Frontier suite (which bundles Agent 365 alongside Microsoft 365 Copilot and security tooling), the agent governance is included as part of the suite at $99/user/month. But the governance itself — the Agent 365 product line — is $15 standalone.

According to Microsoft Learn, as of May 1, 2026, Agent 365 is generally available for the Commercial segment on a per-user basis. It does not require product prerequisites to enable, though Entra P1 or Purview DLP are recommended to make full use of the security benefits.

That’s the easy invoice. Now you actually build and run an agent.

Invoice Two: Copilot Studio Execution Credits

Here’s where every procurement team gets ambushed. Building the agent happens in Copilot Studio, and Copilot Studio is billed separately on a consumption model.

You prepay for Copilot Credits in 25,000-credit blocks. Microsoft prices these blocks at $200 per 25,000 credits — that’s $0.008 per credit in bulk. Or you can pay per-use at $0.01 per credit with no upfront commitment.

But here’s the trap: credits don’t roll over. Each month’s allocation burns or disappears. If you over-provision, you lose the difference. If you under-provision, you hit the overage fence at 125% and agents stop answering until you buy more capacity.

The credit rates themselves are granular:

• Classic answer (static, pre-written): 1 credit
• Generative answer (AI-generated): 2 credits
• Agent action (API call, trigger, topic transition): 5 credits
• Tenant graph grounding (retrieval-augmented generation over Microsoft Graph): 10 credits per response
• Premium AI tools (advanced reasoning): 100 credits per 10 responses

A production agent mixing generative answers and graph grounding hits 12 credits per user interaction. A 500-person organization with moderate daily usage (say, 1 interaction per user per day on average) runs 6,000 credits per day. That’s 180,000 credits per month — roughly $1,440/month just for execution.

Invoice Three: Azure AI Foundry Token Costs

The third invoice hits separately from Azure. When Copilot Studio exhausts its prepaid credits or when you bring your own language models via Azure AI Foundry, you pay for tokens directly.

Azure AI Foundry pricing for GPT-4o runs $2.50 per million input tokens and $10.00 per million output tokens. Depending on model tier and regional deployment, the blended rate across input and output can climb to $30 per million tokens for premium configurations.

For a conversational agent handling 500,000 interactions per month with an average of 500 input tokens and 200 output tokens per interaction — that’s roughly 350 million tokens monthly, or about $1,750 at blended GPT-4o rates (~$5 per million tokens). Larger deployments with multiple agents or higher token volume can easily push this to $5,000–$15,000 per month per organization.

Worked Example: 500-Seat Enterprise

Let’s run the actual numbers for a mid-market IT company deploying a service desk agent plus an onboarding bot.

Month 1:

• Agent 365 governance: 500 users × $15 = $7,500
• Copilot Studio credits: 180,000 credits/month ÷ 25,000 × $200 = $1,440 (assuming moderate use)
• Azure AI Foundry tokens: 350M tokens × $5/M (blended) = $1,750

Total: $10,690/month. That’s not $7,500. That’s not even $9,000.

Scale to annual, and you’ve just committed to $128,280 before you pay for ongoing optimization, refactoring, or scaling agents to additional departments.

Worked Example: 50-Seat Team Testing

Smaller teams get hit harder per seat because the credit block doesn’t scale down. A 50-person ops team experimenting with a single procurement agent:

Month 1:

• Agent 365 governance: 50 × $15 = $750
• Copilot Studio (minimum block): 25,000 credits × $200 ÷ 25,000 = $200 (you’re forced to buy the whole block even for 1 agent)
• Azure tokens (light use): 50M tokens × $5/M (blended) = $250

Total: $1,200/month. For a team of 50 running one experimental agent, that’s $24 per person per month on top of governance. A year commitment is $14,400 for what looks like a “$750/month” implementation.

The Credit Expiry Bomb

According to SAMexpert’s independent analysis, Copilot Studio credits expire monthly with no carryover. If you forecast 150,000 credits for the month but only use 120,000, the 30,000-credit difference burns. There’s no “save it for next month” safety net.

This is a forcing mechanism: you must forecast demand accurately, or you waste money. Most organizations under-forecast in month one, over-provision in month two (to avoid the shame of hitting overage), and then either perpetually over-buy or face agents going dark on high-traffic days.

What Microsoft Won’t Tell You in One Place

The $15 seat price is honest pricing for governance. But Microsoft’s marketing doesn’t front-load the fact that Agent 365 is a control plane — not an execution environment. You buy the right to manage agents you build elsewhere. Building and running them costs separately, billed from different teams’ budgets (IT sees $15, engineering sees $1,500+), and in different invoice cycles.

If you’re evaluating Agent 365 right now, insist your finance and engineering teams talk to each other before signing the contract. The governance cost is the tip. The execution bill is the iceberg.

Related reads: How we rated other AI agent pricing models — Replit’s agent cost breakdown — Devin AI’s real ACU spend.