Agent 365 is per-seat. Copilot Studio runs on credits. Azure Foundry bills per token. Three cost lines, two models, one governance gap.
May 2026 | 6 min read
At a Glance
| $15 | $99 | $200 | 40% |
|---|---|---|---|
| Per user/month, Agent 365 at GA (Microsoft, May 2026) | Microsoft 365 E7 Frontier Suite per user/month (Microsoft) | Copilot Studio prepaid pack: 25,000 credits/month (Microsoft) | Enterprises on consumption AI expected to exceed budget 2x by 2027 (Gartner) |
1. Three Cost Lines, One Billing Mess: Agent 365, Copilot Studio, and Azure Foundry
Microsoft’s agent platform is not one product with one price. It is a stack with three distinct billing mechanics, each arriving at the finance team at different times, in different formats, under different governance regimes.
The first line is Agent 365. This is Microsoft’s governance and security layer for AI agents. It will be $15 per user per month at general availability on 1 May 2026, and it is bundled into the new Microsoft 365 E7 suite at $99 per user per month. Per-seat, predictable, easy to budget.
The second line is Copilot Studio. This is where agents are built and run. It is billed in Copilot Credits at $200 per pack of 25,000 credits per month, or via Azure pay-as-you-go at an equivalent unit rate. Credits are consumed based on what an agent actually does. Simple replies cost less. Workflows, connector calls, retrieval, and custom skills cost more.
The third line is Azure AI Foundry, where agents built on custom models consume tokens directly. GPT-5.4 runs at $2.50 per million input tokens and $15 per million output tokens at standard tier. The Pro tier is $30 input and $180 output per million. Token use scales with content length, grounding context, and output verbosity.
Three lines. Two billing models. One finance team that wants a single monthly number.
2. Per-Seat vs Consumption: Why Microsoft’s Two Billing Models Create Budget Blind Spots
Microsoft 365 licensing was built around per-seat pricing for a reason. It is predictable. Finance can model it, IT can enforce it, procurement can negotiate it. Agent 365 fits this mould neatly.
Copilot Studio and Foundry do not. They are consumption products that happen to be sold through the same commercial motion. Unit rates are published, but actual spend depends on behaviour, and behaviour is difficult to predict in advance.
The difference matters because finance governance assumes the first model. Budget holders, approvers, and cost-centre owners sign off on a per-user rate and assume the cost is locked. With consumption, the cost is not locked. An agent that retrieves ten files per prompt consumes more than one that retrieves three. An agent that runs 500 times a day consumes more than one that runs fifty.
When per-seat governance meets consumption billing without a control layer in between, the bill always wins.
Gartner’s 2026 forecast is that by 2027, 40% of enterprises using consumption-priced AI tooling will see unplanned costs exceeding twice their budget. That is not a wild estimate. It is what happens when consumption meters run against usage patterns no one modelled in advance.
3. Why the Microsoft 365 E7 Bundle Does Not Solve Your Agent Spend Problem
Microsoft 365 E7, Microsoft’s first Frontier Suite, launches on 1 May 2026 at $99 per user per month. It bundles E5, Microsoft 365 Copilot, Entra Suite, and Agent 365. According to SAMexpert’s breakdown, buying the four products separately would cost $117, so E7 saves around 15% at list price.
It is a reasonable deal for organisations that would have bought all four anyway. It is not a solution to the cost problem.
E7 covers the per-seat governance layer. It does nothing for the consumption layer underneath. Every agent built in Copilot Studio and every model call routed through Foundry generates a separate bill. Organisations that read “Agent 365 included” and assume agent execution is covered will find out at month-end that it is not.
The same pattern appeared when Microsoft 365 Copilot launched. Finance teams budgeted the $30 per user per month licence and treated it as done. Then the first agent-prompt usage reports came in, and suddenly there were credit packs, pay-as-you-go meters, and a governance conversation nobody had prepared for.
4. What Gartner and Microsoft Both Predict for AI Agent Costs in 2026 and 2027
Two data points from 2026 tell the same story from different angles.
Microsoft reports that 80% of Fortune 500 companies already run active AI agents as of February 2026. That is not an aspiration number. It is a deployment number, and most of those deployments sit on consumption billing.
Gartner’s 2026 forecast is that 40% of enterprise applications will embed task-specific AI agents by the end of the year, up from under 5% in 2025. Gartner also predicts that more than 40% of agentic AI projects will be cancelled by 2027, citing escalating costs, governance gaps, and unclear ROI as the main reasons.
Those are not contradictory forecasts. They describe the same phenomenon. Adoption is outrunning governance, and governance will catch up through project cancellations and budget blowouts before it catches up through better controls.
Consumption billing risk
Agents running in loops, over-aggressive retrieval patterns, and orphaned test bots are the three most common drivers of unexpected monthly spend. A single misconfigured agent can consume a multi-thousand-dollar credit budget in days without anyone noticing until the invoice lands.
5. The Microsoft 365 Governance Parallel: Licence Sprawl Meets Agent Sprawl
Every organisation running Microsoft 365 already knows what licence sprawl looks like. Unused E5 assignments. Duplicate Business Premium seats. Guest accounts from a 2022 project nobody shut down. Storage growth nobody reviewed. Permissions assigned by someone who has since left.
Agents will produce the same sprawl, faster, and with sharper cost consequences.
The parallel is exact. A licence bought and not used still bills at month-end. An agent built and left running still consumes credits at month-end. A guest account with orphaned access is a compliance exposure. An agent with inherited permissions is a compliance exposure. The mechanics are the same. The governance discipline is the same.
What changes is the velocity. M365 licence sprawl accumulates across quarters. Agent consumption sprawl accumulates across days. A misbehaving agent running in a loop does not wait for the next licence review.
The organisations that budget agents the way they budgeted Exchange servers in 2005 are about to repeat 2005’s bill shock at cloud speed.
6. How to Govern Agent Costs Before the Bill Arrives
The governance habits that contain M365 costs are the same habits that will contain agent costs. There is no separate playbook for agents, just a compressed one.
Visibility comes first. Every agent, every model, every consumption meter, mapped to an owner, a cost centre, and a business justification. If IT cannot answer “who owns this agent and what does it cost this month”, the budget conversation is already lost.
Lifecycle management comes second. Agents need creation, review, and decommission gates, the same way user accounts and licences do. Without a lifecycle, agents accumulate. Retired projects leave behind running services. Trial builds become production dependencies. The storage and permissions attached to each one outlive the agent that created them.
Policy comes third. Spending caps per agent, per team, and per workload type. Alerting on consumption deviation. Segmentation between experimentation budgets and production budgets. The same practices Azure governance teams have run for a decade, applied one abstraction layer up.
None of this requires a dedicated agent governance product to start. Most of it can be built on the discipline already running against M365 licences, storage, and permissions.
Conclusion
Microsoft Agent 365 is a legitimate improvement. It gives IT administrators a first-party control plane for agent visibility, identity, and security. It does not, and was not built to, govern consumption spend.
The gap between what Agent 365 controls and what actually shows up on the Azure invoice is where the 2026 surprises are going to land. Organisations that budget agents as a purely per-seat cost will be corrected by reality in their first full quarter of deployment.
The fix is not another product purchase. It is applying the governance principles that already work in Microsoft 365 to a workload class that now bills by the token, the credit, and the seat all at once.
Organisations that treat agent governance as an extension of their existing M365 discipline, rather than a new domain, will be the ones in control when agents become standard infrastructure. The rest will spend 2026 explaining consumption bills to their CFOs.
About TeamsFox
TeamsFox is the Microsoft 365 management and optimisation platform that gives IT teams tenant-wide visibility, evidence-based governance, and automated licence and storage right-sizing. Headquartered in Düsseldorf and trusted in 20+ countries, TeamsFox helps organisations reduce licence spend by up to 30%, cut storage costs by 40%, and free up 60% of administrative time.