Copilot and a Copilot Studio agent are not the same thing

Microsoft 365 Copilot is a per-seat assistant that sits over Word, Excel, Outlook and Teams. A Copilot Studio agent is a separately built, separately priced bot aimed at one narrow job, like answering support questions or triaging a specific request type. They share a name and a logo, and that's about it. Buying seats of the first doesn't give you the second, and buying the second doesn't require the first.

What is M365 Copilot?

M365 Copilot is the assistant Microsoft bolted onto Office. Open a Word doc and it can draft or summarise. Open Outlook and it can write a reply or catch you up on a long thread. Open Teams and it can pull the key points out of a meeting you missed. It's useful for that, and it's licensed the way you'd expect: per person, per month, as an add-on to an existing Microsoft 365 plan.

As at 1 October 2026, Microsoft's NZ list price for M365 Copilot on a Business plan is NZ$34 per user per month, paid yearly, on top of the base Microsoft 365 subscription each person already pays for. Microsoft moves this pricing often enough that a specific figure here would go stale within a quarter, so check the current rate before you budget rather than trusting a number from a blog post, including this one. Whether that per-seat money is worth spending at all, seat by seat, is its own question with its own post.

The part that matters for buying decisions: M365 Copilot only knows what's in front of it, plus whatever it's allowed to see across your tenant through Microsoft Graph. It doesn't come with knowledge of your product catalogue, your support history, or the specific way your business handles a refund. It's a general-purpose assistant riding on top of general-purpose apps.

What does a Copilot Studio agent do?

Copilot Studio is a different product entirely: a builder for creating a purpose-built agent aimed at one job. You feed it a specific knowledge source, whether that's a SharePoint library, a set of documents, a database, or a set of actions it's allowed to take, and you shape it to answer one kind of question well rather than every question passably.

The licensing model is usage-based rather than seat-based. Microsoft prices agent interactions in Copilot Credits, a shared pool your organisation draws from as agents get used, sold as a monthly capacity pack on an annual commitment or pay-as-you-go. Not every interaction costs the same: a simple lookup burns fewer credits than a generative answer, and one that searches your tenant's data or takes an autonomous action costs more again. Microsoft renamed the unit from "messages" to "Copilot Credits" in September 2025. Same mechanism, new name.

One detail worth knowing: if your staff already hold M365 Copilot seats, their use of an agent built for them can draw on that licence instead of the shared credit pool. That's a genuine cost saving where it applies, but it's a side effect of already having Copilot seats, not a reason to buy them.

There's no seat count here and no per-person invoice. A five-person business and a fifty-person business can run the same agent on the same credit pool, and the bill tracks how often the agent gets used rather than how many people could theoretically ask it something.

M365 Copilot	Copilot Studio agent
Lives inside Word, Excel, Outlook, Teams	Standalone agent, deployed wherever you point it
Licensed per user, per month	Licensed by usage, via a shared credit pool
General assistant across whatever you're doing	Built and scoped for one specific job, published to whichever channel fits
Knowledge limited to the current doc plus Graph access	Knowledge is whatever data source you connect it to
Cost scales with headcount	Cost scales with how often it's asked something

The same split runs across vendors, incidentally: weighing either of these against OpenAI's per-seat offering is covered further down, and the seat-versus-usage logic carries straight over.

When each one is the right call

A firm of twelve people who spend their day drafting proposals, replying to email and sitting in meetings: M365 Copilot seats are a reasonable buy. Everyone touches Office all day, the assistant sits right there, and the per-seat cost is easy to justify against time saved on drafting.

A support team fielding the same twenty questions about a specific internal system, or a business that wants a chatbot answering questions from its own documentation: that's a Copilot Studio agent. I built one of these for a large internal Power App, aimed only at troubleshooting that system, and it cut the time to find and fix an issue from about half a day to under ten minutes. No amount of M365 Copilot seats would have produced that result, because M365 Copilot was never going to know that system's internals in the first place.

Plenty of businesses need both, and that's fine. They're not competing purchases. One helps your staff do their existing jobs faster inside Office. The other does a specific job on its own.

Why don't Copilot seats give you a support bot?

This is where the wrong purchase gets made. Someone in the business says "we need a bot for the support inbox," and the conversation drifts towards "let's just get everyone Copilot licences." Different product, different problem.

M365 Copilot, even fully licensed across the team, still only knows what's in the documents and emails it can see. It has no concept of your product's failure modes, your refund policy's edge cases, or last month's support tickets, unless someone separately builds and connects that knowledge, which is exactly the Copilot Studio work. Paying for forty seats to get a support bot out the other end buys the wrong product, and leaves you with a bill that scales with headcount instead of usage.

The reverse mistake happens too, just less often: assuming a generic AI assistant will eventually cover the support job on its own. It won't, for the same reason. General tools answer general questions well. A support bot needs to answer specific questions about your specific business, and that's what Copilot Studio is built for.

How to tell which one a problem needs

Ask two questions before anyone reaches for a credit card.

• Is this about helping people do their existing job inside Office faster, drafting, summarising, catching up on threads? That's M365 Copilot, priced per seat, and the decision is mostly about how many of your people would actually use it daily.
• Is this about one specific, repeatable job, answering the same kind of question, checking against the same kind of data, over and over? That's a Copilot Studio agent, and the decision is about how narrow you can make its job and how good its knowledge source is, not how many staff you have.

If you're not sure which bucket a problem falls into, that's normal, the product naming is confusing, and I spend a fair bit of every intro call untangling it. I go through the full Power Platform and Copilot agent picture if you want the mechanics, or if there's already a specific job in mind, what a Copilot Studio agent is good for in practice.

What the seat version can reach in your files, and what it cannot: what Microsoft 365 Copilot can actually see.

Where does ChatGPT Business fit?

ChatGPT Business, which most people still call Team, is the OpenAI product that gets compared with both of these. Its plans are compared side by side in ChatGPT Plus vs Team vs Enterprise. OpenAI renamed Team to Business in August 2025, then cut the price in April 2026: USD 25 a user a month billed monthly, USD 20 billed annually, two-seat minimum. It is a per-seat chat assistant, so the fair comparison is with M365 Copilot, not with Copilot Studio. Copilot Studio agents are billed from pooled credits: USD 200 a month prepaid buys 25,000, or 1c a credit pay-as-you-go. (Prices checked August 2026.)

Between the two seat products, the decision comes down to where your firm's data lives. If "summarise the Henderson job" needs your Outlook, SharePoint and Teams to answer, that tenant grounding is the case for M365 Copilot. If your people want a general assistant for drafting and working through problems, ChatGPT Business is cheaper per seat than it was before the April cut. Listen to how the request is phrased: "can the team get ChatGPT?" is a seat problem, while "can something answer the same twenty questions our staff ask every week?" is a process problem, and buying seats for it means paying per head for one job. The default for a 5 to 20 person firm: trial three to five seats of whichever assistant matches where your data lives, measure daily use, and treat any repeatable question-answering process as an agent candidate priced on consumption.

Questions I get asked about these two