Copilot Studio is the Microsoft agent-and-chatbot surface, licensed on a per-tenant capacity-purchase basis in messages per month. The Admodum read on the message metering, the M365 Copilot agent overlay, the autonomous-agent model and the audit posture.
Copilot Studio is the Microsoft no-code and low-code surface for authoring conversational agents, chatbots and autonomous agents. The product was launched as Power Virtual Agents in 2019; it was rebranded and re-positioned as Copilot Studio in 2023 alongside the broader Microsoft Copilot announcement.
The product surfaces agents in five principal channels: Microsoft Teams (the internal-user surface), Microsoft 365 Copilot (the agent-extension surface for the per-seat M365 Copilot users), web (a published external bot), Direct Line (the SDK integration for custom hosts), and Voice (the IVR / contact-centre surface). The wider Power Platform licensing framework is the family within which Copilot Studio sits; the wider Copilot seat economics spoke reads the per-seat M365 Copilot surface.
Copilot Studio carries a per-tenant capacity-purchase licensing model. A capacity pack (the base SKU is 25,000 messages per month at the time of writing; the smaller-tenant SKU is a lower-capacity entry; larger tenants stack multiple packs) carries a fixed monthly message allotment; consumption is metered per agent turn (the bot responds to a user; one message consumed) and per autonomous-agent action (the agent performs a non-conversation action; multiple message-equivalents consumed).
The economics shift at scale. The per-message rate within a pre-purchased pack runs at a small fraction of the per-message overage rate; the buyer's planning task is the production-volume forecast against the pack count. A single misjudged pack count carries a multi-x overage penalty; a sized-correctly stack carries the pre-purchased rate.
Agents authored in Copilot Studio and surfaced inside Microsoft 365 Copilot consume against the Copilot Studio capacity pool, not against the per-user M365 Copilot seat licence. The seat licence covers the user's access to the M365 Copilot surface and the agent invocations within the surface; the agent-side consumption (every turn the agent performs, every autonomous-agent action) consumes the Copilot Studio messages.
The Admodum read on the cost stacking is the dual-meter calculus. A buyer with 1,000 M365 Copilot seats and 30 Copilot Studio agents carries two simultaneous cost vectors: the per-seat licence on the 1,000 users and the per-message capacity on the 30 agents; the total is the sum, not the maximum. The wider Copilot pilot methodology reading addresses the pilot-to-production transition where the agent-side cost becomes visible.
The 2024 introduction of autonomous agents (agents that perform actions beyond the conversation turn: scheduling, file operations, third-party-system invocations, multi-step orchestrations) introduced a separate metering surface. The autonomous-agent action consumes a higher per-action message-equivalent rate than the conversation-turn rate.
The economics on the autonomous-agent surface are early; the buyer-side rule is the per-agent pilot with measured action consumption before rollout. A 100-user pilot at low action count generates a sized-correctly pack inference; a 10,000-user rollout at action-rich pattern carries multi-pack consumption that should be forecast pre-rollout. The wider Azure OpenAI economics framework reads the parallel principle on the per-token inference surface.
The channel surfacing (where the agent runs) is a licensing-neutral question on the conversation-turn side: every conversation turn consumes a message, regardless of channel. The channel question becomes licence-sensitive on two specific surfaces: the Voice channel (IVR/contact centre) carries a separate per-minute or per-utterance metering arrangement on the underlying voice platform (Microsoft Dynamics 365 Contact Center, Azure Communication Services, or a third-party voice surface), and the external-web channel carries the public-bot deployment with the same message metering as internal channels.
The wider Teams Phone licensing spoke reads the Microsoft voice surface against which the Voice-channel Copilot Studio agents often integrate; the wider Microsoft BATNA framework reads the alternative agent-platform routes (OpenAI Assistants API, Anthropic Claude, Google Vertex Agents) against which the Copilot Studio lock-in cost is measured.
The buyer-side artefacts to hold against the Copilot Studio estate are: the agent inventory (every agent, every channel, every owner, every business sponsor), the message-consumption record (every agent, every monthly message count, every overage flag), the autonomous-agent inventory (every autonomous agent, every action count, every action type), the capacity-pack register (every purchased pack, every term, every utilisation), the M365 Copilot stacking position (every seat, every agent the seat invokes).
The wider engagement sits in the Microsoft practice and the AI Vendors practice; active renewal moments route to the Renewal Programme.
The per-seat M365 Copilot surface stacked alongside the Studio message pool.
The per-token inference surface against which the message pool is read.
A senior Admodum Microsoft advisor will read your agent inventory, your message-consumption record and your autonomous-agent action profile against your renewal posture on a private call. Active renewal moments route to the Renewal Programme.