10-09-2026 - Author: Ralph Boccalini
Building an agent in Microsoft Copilot Studio is the easy part. The question I get most often comes right after: “What do we need to license to actually publish it, and what will it cost to run?” Copilot Studio licensing has changed several times in the past year: messages became Copilot Credits, and agents now run on different “harnesses,” each billed differently. This guide explains how it works as of September 2026.
1. Two things to license: the maker and the capacity
Copilot Studio licensing has two parts:
- Who can build and publish: a per-user right for the people who create and manage agents (the Copilot Studio User License, or a Microsoft 365 Copilot license).
- What the agent consumes when it runs: capacity measured in Copilot Credits, pooled at the tenant level and allocated to Power Platform environments.
The people using your published agent don’t need a special license. Anyone with access to the published agent can talk to it, and their usage is billed against your tenant’s credits.
2. What you need to publish an agent
This is where most people get stuck. According to Microsoft’s billing FAQ, the Publish button works if one of the following is true:
- The maker has a Microsoft 365 Copilot license assigned.
- The maker has a Copilot Studio User License and Copilot Credits are allocated to the environment.
- The maker has a Copilot Studio User License and “Draw from tenant pool” is enabled for the environment.
- The maker has only an Office (Microsoft 365) license and the agent uses no generative AI features.
Trial licenses can’t publish. The trial lets you build and test agents in the test chat panel, but publishing is blocked. If you see a billing error on the Publish dialog, the environment has no prepaid or pay-as-you-go capacity. If you see a user license error, the maker is missing one of the licenses above.
There is also a Teams plan, included in some Microsoft 365 subscriptions, for classic-orchestration agents published only to Microsoft Teams. It doesn’t include generative AI features or other channels.
3. Copilot Credits: the common currency
Since September 1, 2025, Copilot Studio measures agent usage in Copilot Credits instead of messages. The quantity per pack and the pay-as-you-go rate stayed the same. There are three ways to buy credits:
| Option | How it works | Typical list price |
|---|---|---|
| Prepaid capacity pack | 25,000 Copilot Credits per month, pooled across the tenant. Unused credits don’t roll over. | €172.56 / pack / month (€0.0069 per credit) |
| Pay-as-you-go | Billed monthly to an Azure subscription through a Power Platform billing policy. No commitment. You pay only for what agents actually use. | €0.0086 per credit |
| Copilot Credit prepurchase plan | One-year prepaid pool of Copilot Credit Commit Units (CCCUs), bought in the Azure portal and usable across eligible Microsoft products. Microsoft advertises savings of up to 20%. | Depends on commitment tier |
You can combine the options. A common setup is a prepaid pack for the baseline, with pay-as-you-go linked to the same environments as overflow, so agents don’t stop when the pack runs out.
4. How many credits does an agent consume?
For agents on the standard harness (the classic Copilot Studio experience), Microsoft publishes these rates:
| Agent feature | Copilot Credits |
|---|---|
| Classic answer (authored, static response) | 1 |
| Generative answer | 2 |
| Agent action (triggers, reasoning steps, topic transitions, computer use) | 5 |
| Tenant graph grounding | 10 |
| Agent flow actions (per 100 actions) | 13 |
| Text & generative AI tools: basic / standard / premium (per 1K tokens) | 0.1 / 1.5 / 10 |
| Content processing (per page) | 8 |
| Voice: classic / GenAI / premium GenAI (per minute) | 10 / 35 / 75 |
One user prompt can trigger several of these at once. For example, a tenant-graph-grounded answer costs 12 credits (10 for grounding + 2 for the generative answer). Reasoning models add the premium tool rate on top of the feature rate.
A quick example: a website support agent gives an average of four classic answers and two generative answers per conversation to 900 customers a day. That’s (4×1 + 2×2) × 900 = 7,200 credits per day, or roughly 216,000 per month, which needs about nine prepaid packs (≈ €1,553 per month) or ≈ €1,864 on pay-as-you-go. Do this math before you publish a public-facing agent.
5. The Microsoft 365 Copilot benefit (and its limits)
If you already pay for Microsoft 365 Copilot (≈ €25.88/user/month), you get a significant benefit. When a Microsoft 365 Copilot–licensed employee uses a standard-harness agent under their own authenticated identity, the classic answers, generative answers, agent actions, graph grounding and AI tools it uses are not charged. This is the internal, business-to-employee scenario, subject to fair-use limits.
Be aware of the exceptions:
- Unlicensed users (colleagues without Microsoft 365 Copilot, customers and external users) consume credits at the standard rates.
- Agent flows are free only when triggered by “When an agent calls the flow” for a licensed user. Scheduled or event-triggered flows consume credits.
- Computer-Using Agents aren’t included in the Microsoft 365 Copilot license.
- Autonomous agents that run without a licensed user behind them consume credits.
6. New in 2026: harnesses change how you’re billed
Copilot Studio now runs everything on one of three harnesses, and each is billed differently:
- Standard harness: rule-based, topic-driven agents and agent flows. Uses the capacity model and rates above, and billing starts after you publish.
- Copilot chat harness: extends Microsoft 365 Copilot Chat with your organization’s knowledge for internal use. It’s billed on consumption or included in the Microsoft 365 Copilot license.
- GitHub Copilot harness: the “new experience” for reasoning-heavy, multi-step processes, with skills, memory, file creation and a sandbox. It uses fully usage-based billing: credits cover LLM tokens, tools (knowledge, MCP) and the harness itself, and billing starts while you build. Test chats, AI-generated instructions and evaluation runs all consume credits.
My advice: prototype new experience agents in a dedicated environment with its own credit allocation, so experiments can’t drain production capacity.
7. What happens when you run out
- Capacity is enforced monthly, and unused prepaid credits expire.
- When an environment uses up its allocation, it can draw from the tenant pool. When the tenant reaches 125% of purchased capacity, enforcement kicks in and custom agents can be disabled.
- For agent flows, new runs are blocked as soon as prepaid capacity is exhausted. Runs already in progress finish, and the agent still answers questions.
- Environments linked to pay-as-you-go aren’t blocked, because the overage is billed to Azure.
Admins can monitor consumption and set monthly limits per agent in the Power Platform admin center under Licensing → Copilot Studio.
Checklist before you publish
✅ The maker has a Copilot Studio User License or a Microsoft 365 Copilot license (not a trial).
✅ The tenant has capacity: a prepaid pack, a prepurchase plan or pay-as-you-go.
✅ Credits are allocated to the target environment, or “Draw from tenant pool” is enabled.
✅ You’ve estimated monthly consumption with the Copilot Studio agent usage estimator.
✅ You know which users are covered by Microsoft 365 Copilot and which will consume credits.
✅ Per-agent limits and alerts are configured in the Power Platform admin center.
Licensing details change often, so always check the latest Microsoft Copilot Studio Licensing Guide before you commit to a purchase.