Copilot Studio — bespoke development
Building agents in Copilot Studio
Copilot Studio is Microsoft’s low-code environment for building your own AI agents: on your knowledge sources, connected to your systems and with your own rules, in Teams, Microsoft 365 or on your website. We design, build and test those agents for your organisation, set up management and cost control, and keep improving them after go-live.
Who it is for
Who is Copilot Studio for?
As soon as standard Copilot is not enough and you want an assistant that knows your organisation or process. Want to know what an agent is first? Read what Copilot agents are.
Organisations with a recurring process
The same customer questions, invoices or HR requests over and over? That is where an agent pays off fastest.
Teams that want more than standard Copilot
You need an assistant with its own knowledge, its own connections and its own rules — for a team or for your customers.
IT and Microsoft 365 administrators
You want agents within your existing permissions, labels and management — with oversight of every agent and of usage.
Three routes
Buy, build or bring your own
There is no single path to a working agent. We advise the fastest route per use case.
Buy
Ready-made agents from the Agent Store. Live instantly — the fastest route to value.
Build
We develop in Copilot Studio: from no-code knowledge agent to autonomous workflows.
Bring your own
Connect existing AI from Azure, AWS or elsewhere — in one register and under one governance through Agent 365.
What you get
Four concrete deliverables
Not a loose prototype, but an agent running in your environment — one you can manage.
An agent design
Knowledge sources, actions, channels and rules — set before we build.
A working agent
Built in Copilot Studio and tested on real conversations, in the channels you choose.
Connections to your systems
Connectors and MCP servers to your CRM, ERP or own APIs.
Control over management and cost
The right harness, separate development environments and per-agent limits on Copilot Credits.
Approach
How we develop
From idea to working agent, in four steps.
- 1
Intake & data check
We test the process, data and feasibility. Honest advice.
- 2
Design
Knowledge sources, actions and rules — set before we build.
- 3
Build & test
We build and test on real conversations.
- 4
Live & ongoing
Rollout, measure and smarter every sprint.
Standard Copilot you buy. Bespoke we build.
Applications
What we build
Six applications already running today — with publicly reported results.
Customer service agents
Chatbots on your own knowledge base. Eneco’s agent resolves 67% on its own.
Internal knowledge assistants
One assistant that knows your handbooks and policies — with a source for every answer.
Document and invoice processing
Agents that read and check invoices. Dow checks around 100,000 a year.
HR self-service & onboarding
Answers to leave and expense questions. Onboarding ready for new colleagues.
Connections to your systems
Connectors and MCP servers (Model Context Protocol, the open integration standard) link Microsoft 365 to your CRM, ERP or own APIs.
Analysis & reporting
Agents gather data and prepare a readable report — fresh every week.
Lite or full
Two flavours of Copilot Studio
Lite for team knowledge, full for process automation. Read how to keep that manageable under governance & AI Council, and how to bring teams along under adoption & change management. Want to learn it yourself? You can via Copilotcursus.nl.
| Agent Builder (in Copilot Chat) | Studio full (Power Platform) | |
|---|---|---|
| For whom | Information workers | Makers & developers |
| Typical agent | Knowledge Q&A on SharePoint or documents | Multi-step workflows and autonomous actions |
| Channels | Within Microsoft 365 | Also external: website, apps, telephony |
| Connections | Work data and simple sources | Hundreds of connectors, CRM/ERP, own APIs |
| Management | Microsoft 365 admin centre | Power Platform admin centre, with Agent 365 as the overarching register |
| Time to first version | Minutes to hours | Weeks (MVP), then iterate |
| Billing | Copilot licence or pay-as-you-go | Licence or Copilot Credits — always credits on the GitHub Copilot harness |
Harnesses
Three engines under your agent
Since September 2026 you choose a harness for each agent in Copilot Studio: the runtime between the model and your agent. That choice determines what the agent can do — and how you are billed.
Standard harness
For agents with fixed topics and agent flows. For licensed users in Microsoft 365 channels, covered by the Copilot licence.
GitHub Copilot harness
For goal-driven multi-step agents with skills, memory and files. Always consumes Copilot Credits, even while you build and test.
Copilot Chat harness
For agents that extend Microsoft 365 Copilot Chat, in the environment where your team already works.
FAQ
Frequently asked questions
Frequently asked questions about Copilot Studio.
What exactly is Copilot Studio?
Microsoft’s low-code environment for building and managing AI agents, workflows and, more recently, apps. You set the knowledge sources, the model, the permitted actions and the channels — from Teams to your website.
What is the difference between lite and full?
Lite — called Agent Builder in Microsoft 365 — sits in Copilot Chat: staff quickly build a knowledge agent with it, on a SharePoint site for example. Full is for makers and developers: multi-step workflows, autonomous actions, external channels and lifecycle management via the Power Platform.
When do you choose Studio over standard Copilot?
As soon as you want an assistant specific to your organisation or process. Own knowledge, own connections, own rules — available to a whole team or all your customers.
How quickly is a working agent ready?
Often within a few weeks. Holland America Line built its MVP in three months, Amgen its first agent in six weeks. We then improve on real conversations until a broad rollout is responsible.
Does an agent respect our security and compliance?
On the Microsoft platform an agent inherits your security: permissions, sensitivity labels and DLP still apply, and every new agent automatically gets its own Entra Agent ID. Choose an Anthropic model and that processing falls outside the EU Data Boundary — a deliberate choice. Build outside the platform with your own models and the responsibility is yours; we advise on that trade-off.
How does billing for agents in Copilot Studio work?
Agents on the standard harness are covered by the Copilot licence for licensed users in Microsoft 365 channels. Agents on the GitHub Copilot harness, computer use and apps consume Copilot Credits; on the GitHub harness that starts with the very first build or test. That is why we set up separate development environments, with limits per agent.
Which AI models can I choose in Copilot Studio?
By default an agent runs on a GPT model from OpenAI. You can also choose Claude models from Anthropic, and Mistral on an experimental basis. Note: Anthropic models process data outside the EU Data Boundary and must be allowed by your admin. For each agent we pick the model that fits the task, quality and compliance.
Further reading
More on agents and Copilot
- Copilot agents What agents are, what they do and when to use one.
- All Copilot agents (2026) Which agents come with Microsoft 365 Copilot out of the box.
- Copilot implementation Live securely and compliantly in your existing Microsoft 365 environment.
- Copilot Credits & cost management How usage works and how to keep it under control.
- Governance & AI Council Ground rules and oversight of every agent in your organisation.
- Blog: build an agent without code From idea to working agent in seven steps.
An agent that knows your processes?
Tell us which process you want to improve. In a demo you see what is possible — with an honest estimate of the build time.