Skip to content

Copilot Agents

We build agents that finish the work

Agents carry out tasks on their own. We advise, develop and implement them — from ready-made to fully bespoke.

Advice first

You already have these agents

Before we build: this is already in place. That is where our advice begins.

Researcher

Runs its own research across your work data and the web — with sources and a model of your choice.

Analyst

Analyses data with visible Python code. You never need to read it.

Facilitator

Takes live minutes in Teams: agenda, notes and actions for everyone.

Sales Agent

Brings live CRM data from Dynamics 365 or Salesforce into Outlook, Teams and Copilot Chat.

Word, Excel & PowerPoint agents

Documents, calculation models and presentations from a single brief — included with Copilot Chat, just like the Prompt, Writing and Idea Coach.

Copilot Cowork

Delegate multi-step work that Copilot completes on its own — billed in Copilot Credits.

See the complete agents overview

Want your team up to speed quickly? That runs through our sister platform Copilotcursus.nl.

Our conviction

Standard agents where you can. Bespoke where it counts.

Microsoft’s growth model — assistant, agent colleague, agent orchestration — we build it with you phase by phase.

Our approach

From idea to working agent

No months-long project: short iterations towards an agent that runs.

  1. 1

    Intake

    We pick the process with the biggest impact.

  2. 2

    Design

    Architecture, data access and security — fixed before any build.

  3. 3

    Build

    We develop the agent and connect your systems.

  4. 4

    Iterate

    Measure live, adjust and extend to new processes.

Real-world figures

Proven in practice

67%
of Eneco’s customer conversations the agent resolves on its own

Microsoft customer case, Eneco

27,000
conversations a day handled by Rabobank assistants

Microsoft customer case, Rabobank

±2 hours
a week saved by a KPN salesperson with Copilot

Microsoft customer case, KPN

100,000
invoices a year checked by Dow’s agent

Microsoft WorkLab, Dow

Results as published in Microsoft customer cases; no guarantees for individual situations.

Governance

Keep control as you scale

We set up your agent management: register, owner, clear ground rules.

Agent register

One overview of every agent in Agent 365 — including those from other platforms.

Ownership

Every agent has an owner and its own Entra Agent ID. Ownerless agents are flagged.

Security templates

Fixed security bundles that apply automatically on activation.

Shadow AI detection

Visibility on agents and AI tools running outside IT.

FAQ

Frequently asked questions

Frequently asked questions about AI agents in Microsoft 365.

What is the difference between Copilot and an agent?

Copilot helps you, inside the app you are in right now. An agent works on its own — often a whole process, like handling customer queries. Copilot lifts personal productivity; agents transform processes.

What are declarative and custom engine agents?

Declarative agents run on Copilot’s orchestration and models: quick to build and automatically inside Microsoft 365 security. Custom engine agents use your own models — maximum freedom, but security becomes your responsibility. We advise on that trade-off.

Do I need developers to build agents?

Not for simple knowledge agents: you build those with Agent Builder in Copilot Chat, on a SharePoint site for example, and once your admin approves them you share them through the Agent Store. Agents that update systems or run multi-step workflows need technical expertise — exactly the bespoke work we develop.

How do you keep a grip on every agent in your organisation?

With central governance: a register of every agent in Agent 365, its own Entra identity and owner per agent, security templates on activation and shadow AI detection. We set up this control on every agent project.

What does an agent cost to run?

It depends on the type. Agents on your work data are usually covered by the Copilot licence for licensed users. Cowork, Autopilot and agents on the GitHub Copilot harness in Copilot Studio consume Copilot Credits — the latter already while you build and test. We are happy to work it out for your scenario — get in touch.

What is Copilot Autopilot?

Autopilot, previously known as Scout, is an agent that stays: with its own name, role, identity and workspace in your tenant. It follows up on threads, keeps an eye on channels and picks up recurring work; you address it with an @-mention in Teams or Outlook. In private preview since late September 2026, billed on consumption.

What is MCP and why does it matter for agents?

The Model Context Protocol (MCP) is an open standard that lets agents use data and actions from other systems securely. Microsoft is going all in on it: federated connectors use MCP to fetch real-time data without storing it in Microsoft 365, and Copilot Studio agents use MCP servers as tools. We connect your systems through MCP and record which servers are allowed.

Which process will you hand over first?

In a single working session we map your processes and prioritise agent opportunities by impact.