Agent governance pricing: seats vs capacity

Per-seat control planes

Microsoft Agent 365 is generally available at USD 15 per user per month standalone, or included in Microsoft 365 E7 (per Microsoft's GA announcement (May 2026); verify current packaging). Under this model the bill scales with the number of licensed users, and Microsoft's GA material notes Agent 365 "works best" with Microsoft E5-class prerequisites. The control plane's effective cost can include the suite around it.

Two cautions on that framing. First, the E7 suite price is not the incremental cost of Agent 365: an already licensed, eligible E7 tenant may have no incremental Agent 365 license charge under the described packaging, and this page does not price E7: the only suite figure available to this review was unattributed secondary reporting, so it has been removed. Second, list-price multiplication is not contractual billing: enterprise agreements, discounts, prerequisites, and implementation costs decide the real invoice.

Capacity-metered governance

AAES's internal rate card (v2.0) sets the charging model as a design decision: humans are never billed, there is no per-agent license charge or licensed-agent-count limit, and governed-decision capacity is the only meter. The people who approve are not seats. The agents that act are not seats; capacity and deployment constraints still exist, so this is a statement about license structure, not a scalability promise. Buyers purchase capacity to govern decisions: the evaluated, journaled, recorded decisions the gateway makes on actions routed through it. This is a stated pricing policy from a pre-launch company with no public rate card; Rates have not been published and would be agreed for a future evaluation. Nothing on this page is an AAES quote.

The meter is published here as a versioned definition, without rates. One governed decision is one admission decision the gateway evaluates and journals for an action routed through AAES. Permits and refusals both count: a refusal is a decision the gateway evaluated and journaled, not a non-event.

The current definition does not yet settle, and any evaluation agreement would need to pin down:

• whether retries and polling requests count as separate governed decisions;
• what happens when purchased capacity is exhausted (additional refusals versus queuing);
• whether support, setup, record retention, storage, or export are separately charged.

AAES runs on the customer's infrastructure; infrastructure costs are the customer's either way.

This table is arithmetic on Microsoft's published list price, not an AAES quote and not a price comparison. The Agent 365 column multiplies the published USD 15 per user per month. The AAES column contains no amount because no public AAES rate card exists. The table shows how a per-seat list price scales with headcount; list-price multiplication is not contractual billing, and it excludes discounts, prerequisites, and implementation costs. Capacity pricing has no per-user number to put here.

Two conclusions. First, per-seat pricing is predictable and, at USD 15 per user, cheap to start. A bundled control plane an eligible E7 tenant already pays for is an easy default, and this page does not argue otherwise: AAES may supplement rather than replace those licenses, and no relative-cost or savings conclusion is possible from this table: no public AAES rate card exists, and list-price multiplication is not contractual billing. Second, the meter, not the amount, is the durable difference: the sections below cover what each meter pays for and what it incentivizes.

Agent adoption and seat count do not grow together for long. The first agents arrive under existing staff; then agents outnumber people. Under a per-seat meter, the governance bill grows with human headcount (or whatever proxy for headcount the vendor defines), even as the work shifts to agents. Under a per-agent meter, the bill grows with every agent added, and capping agents becomes a budgeting act rather than a governance decision. AAES's design removes both per-head charges: the estate can add approvers and agents without a licensing event, and the bill moves only with the governed-decision capacity actually purchased.

The limit of this argument is equally plain: capacity pricing is a policy today, not a proven market price. No customer has paid it, no design partner has validated it, and acceptance by regulated buyers who cannot benchmark it against per-seat pricing remains unobserved. The policy is stated here because buyers comparing charging models deserve to know what the AAES meter is and what it is not yet.

A governance vendor's meter is a conflict-of-interest statement. Consider the three candidates:

• Paid per seat. Revenue grows as the customer's organization and adoption grow, whether or not the governance controls are exercised. The vendor is paid for presence, not for governed decisions.
• Paid per approval or per intervention. Revenue grows when more human approvals are required, creating a direct incentive to make autonomy expensive and keep the human gate busy.
• Paid per agent. Revenue grows as autonomy spreads, and every cap the customer places on its own agents costs the vendor money.

The supportable policy is narrower than "no conflict": AAES does not propose separate charges for approvers, agents, or approval events. Required approval rules are intended to remain independent of billing. Capacity pricing still creates incentives around volume and purchased capacity: revenue correlated with routed throughput can rise as automation increases, a vendor paid for governed volume still prefers more routed traffic, and routing more of your estate through one vendor has its own concentration risk, which the evaluation section below is meant to test rather than hide.

Pricing decides what you pay; it does not decide what works. Whichever charging model you prefer, run one bounded workflow in a sandbox or a controlled, reversible equivalent (do not make your first test an irreversible production effect), and test the control path directly:

• Attempt the action without approval and expect a recorded refusal.
• Approve the exact request, alter it, and confirm the approval cannot be reused.
• Confirm an agent cannot approve itself.
• Attempt the same action directly against the provider with the agent's own credential (a bypass attempt) and confirm whether anything stops or records it.
• After a refusal, confirm at the downstream provider that the refused action had no effect.
• Present the same approval or grant twice, and concurrently, and confirm it cannot be replayed or double-counted.
• Make the decision journal unavailable and confirm new requests fail closed; check how long previously issued grants remain usable.
• Drive two concurrent requests against a nearly exhausted budget and confirm the shared accounting refuses the overflow.
• Export the records and verify them offline with the network disabled, using keys trusted through a separate channel.
• Require independence on the default export and confirm the verifier rejects it, since no external witness is wired by default.

A platform that passes these tests in a sandbox has cleared a first gate, not an expansion decision: passing smoke tests does not rule out bypassable credentials, replayable grants, broken shared accounting, or fabricated but consistently signed history. A presentation or price sheet alone does not demonstrate these action-layer controls. Enforcement requires control of the agent's credential path. Work that bypasses AAES is invisible. Observation is not enforcement.

Review the architecture Discuss a design partnership

Does AAES publish pricing?

No. AAES is pre-launch and has no public rate card. What is published is the charging model, decided in the internal rate card: humans are never billed, there is no per-agent license charge or licensed-agent-count limit, and governed-decision capacity is the only meter. Rates have not been published and would be agreed for a future evaluation. This page states a pricing policy, not a quote.

What does Microsoft Agent 365 cost?

Per Microsoft's general-availability announcement (May 2026): USD 15 per user per month standalone, or included in Microsoft 365 E7. The E7 suite price is not the incremental cost of Agent 365: an already licensed, eligible E7 tenant may have no incremental Agent 365 license charge under the described packaging. Packaging and pricing change; verify Microsoft's current pages before budgeting.

Is the cost table on this page an AAES quote?

No. The table is arithmetic on Microsoft's published USD 15 per user per month price at 10, 100, and 1,000 users. It contains no AAES amount, because no public AAES rate card exists. It answers one question only: how a per-seat list price scales with headcount. List-price multiplication is not contractual billing: discounts, prerequisites, and implementation costs are not included, and no relative-cost or savings conclusion is possible from it.

Why does a governance vendor's charging model matter?

Because the meter is an incentive. A vendor paid per seat grows its revenue as your adoption grows, whether or not its controls are used. A vendor paid per approval grows its revenue when more approvals are required, and a vendor paid per agent grows its revenue as autonomy spreads. AAES does not propose separate charges for approvers, agents, or approval events. Required approval rules are intended to remain independent of billing. Capacity pricing still creates incentives around volume and purchased capacity.

The Microsoft figure was reviewed against the sources below on September 20, 2026. The USD 15 standalone price and E7 inclusion come from Microsoft's GA announcement. An earlier draft quoted an M365 E7 suite price from unattributed secondary reporting; it has been removed, and no suite arithmetic is shown. The AAES charging model is an internal design decision (rate card v2.0), stated as policy; no public rate card exists and no customer has been billed under it. The table is arithmetic on the sourced figure, not a quote from either vendor, and list-price multiplication is not contractual billing. Corrections are welcome at hello@aaes.ai and will be listed here.

• September 20, 2026: First publication.
• September 21, 2026: Corrected after external adversarial review. Removed the unsourced M365 E7 suite figure and its column; stated that the E7 suite price is not the incremental cost of Agent 365 and that an eligible E7 tenant may have no incremental Agent 365 license charge; stated that AAES may supplement rather than replace those licenses and that no relative-cost or savings conclusion is possible; replaced "never capped" with "no per-agent license charge or licensed-agent-count limit"; published the versioned meter definition; replaced the design-partner-terms sentence with the rates status; narrowed the incentive claim to the billing-independence policy; extended the buyer test list.