One question decides most of these
Does the agent need to call something outside its knowledge sources? If yes, you need Copilot Studio. If no, start in Agent Builder today and stop reading comparison tables.
That is not a preference. Agent Builder is the no-code builder inside the Microsoft 365 Copilot app and Teams, and it has no Actions and no external service integration. Microsoft routes that requirement to Copilot Studio explicitly. So an agent that writes a record into a ticketing system, reads a line-of-business API, or triggers anything beyond answering from the knowledge you pointed it at is a Copilot Studio job. An agent that answers questions over a set of SharePoint sites, files and web sources is not.
Two other constraints on Agent Builder are worth knowing before you plan a pilot around it. It runs at microsoft365.com/chat, office.com/chat, and Teams on desktop and web, but not on mobile. It is included in the Microsoft 365 Copilot licence, with agents grounded on web knowledge only being free.
Everything else in the usual comparison is secondary to that first question, and most of it only matters once you have answered it.
Starting in the cheap tool costs you nothing
The fact that changes how you sequence this work: agents can be copied from Agent Builder into Copilot Studio without starting over.
That works because the thing you are building is not tool-specific. A declarative agent is the artefact, and Agent Builder, Copilot Studio and the Microsoft 365 Agents Toolkit all produce one. The instructions, the knowledge sources, the tone, the scope: that is the part you actually sweat over, and it carries.
So the default sequence is: build it in Agent Builder, with the people who own the process sitting there while you do it, and find out within a fortnight whether anyone wants it. Most do not survive that. The ones that do can move when they need something Studio has, and the move does not cost you the design work.
The failure mode I see most often is the reverse. An organisation decides agents are strategic, provisions Copilot Studio environments, sets up a governance board, and six weeks later has an approval process and no agents. Building the process before you know what people ask for produces a very tidy queue of nothing.
Where Copilot Studio earns the extra work
Copilot Studio is a low-code standalone portal, and it buys you the things a real deployment needs: broader audiences beyond Microsoft 365 Copilot, multi-step workflows, custom integrations, application lifecycle management, environments, and DLP. If you need any of those, the extra effort is not overhead, it is the point.
The part people miss in 2026 is that Copilot Studio is no longer one runtime. It now exposes three harnesses, and the harness determines billing:
| Harness | What it is for | Billing |
|---|---|---|
| GitHub Copilot | Reasoning-heavy multi-step work; natively creates and edits Word, Excel, PowerPoint and PDF; skills and memory; sandboxed | Copilot Credits |
| Standard | Rule-based agents and agent flows | Copilot Credits |
| Copilot chat | Extends Microsoft 365 Copilot Chat with enterprise knowledge | Consumption-based, or included in the Microsoft 365 Copilot user subscription licence |
Pick the harness before you build, not after. A design that assumes rule-based flow behaviour and then gets rebuilt on the reasoning harness has a different cost profile, and the business case you wrote no longer describes the thing you shipped.
On cost, the headline is better than most people assume and the exceptions matter. Employee-facing agent usage is no charge for Microsoft 365 Copilot licensed users, subject to fair-use limits, when the agent runs under the licensed user’s identity. Two documented exceptions: Computer-Using Agents are not included in that licence, and agent flows are only free through the “When an agent calls the flow” trigger. Get either wrong and you are on consumption without having decided to be. Overage enforcement fires at 125% of prepaid capacity and disables custom agents, which is a poor way to discover your billing model.
How I would choose
Work down this list and stop at the first yes.
- The agent must call an external system, or write anything anywhere. Copilot Studio. Agent Builder has no Actions, and no amount of prompt work gets around it.
- The audience is outside Microsoft 365 Copilot, or on mobile, or external to the organisation. Copilot Studio.
- You need environments, DLP, ALM or a promotion path because this thing is going to be supported by someone other than the person who built it. Copilot Studio.
- It is scoped to one site, library or set of files and does nothing else. Use a SharePoint agent and do not overthink it.
- Your developers want source control, and want new capabilities first. The Microsoft 365 Agents Toolkit, with one caveat: publishing is not supported in Microsoft 365 Government tenants.
- None of the above. Agent Builder, this week, with the person who does the job.
The uncomfortable version of this advice is that most organisations are choosing between tools when they have not yet found a task worth automating. The tool question takes ten minutes once you have a real one. Without a real one, it takes six months and produces a policy.