What it does

Copilot Studio is the low-code tool for building custom agents: conversational bots, generative-answer assistants and autonomous agents that act on triggers. It's the place you build when the out-of-the-box M365 Copilot won't do what you need.

Key facts

• Built on the Power Platform. Agents live in Dataverse and environments, and follow the same solution/ALM model.
• Two authoring styles: classic topic-based flows and generative orchestration, where the model picks actions and knowledge for you.
• Consumption is metered in messages. You buy capacity packs or pay-as-you-go via Azure; different actions burn different message counts.
• Publishes to Teams, web, Dynamics, Slack, SMS and more from one agent definition.

When to use / skip

Use it when you need a bespoke agent grounded in your own data and connected to your systems. Skip it if standard M365 Copilot or an agent from the agent store already covers the job — don't rebuild what you get for the licence.

Configuration decisions

• Which environment agents live in, and whether makers get their own dev environment.
• Message capacity model: prepaid packs vs pay-as-you-go.
• Classic topics vs generative orchestration as the default build pattern.

Gotchas

• "Low-code" still means real design work. Ungoverned makers produce sprawl fast.
• Message consumption surprises people. Model it before you scale.

Consultant notes

• Treat this as a Power Platform workload, not a chatbot toy — CoE, environments and DLP apply from day one.
• Build vs buy: check the agent store and M365 Copilot agents before committing to a custom build.
• Capacity licensing is the commercial risk. Get finance comfortable with the message model early.

---

Revisit each Copilot Studio release wave.