Copilot agents - from idea to action
TLDR
For public sector teams ready to move beyond prompting and start building Copilot agents. Participants identify a real workplace opportunity, then design, build, govern and test an agent using Copilot Agent Builder.
Some public sector teams have got comfortable with Copilot for individual tasks: drafting, summarising, pulling together a first version of something. The next step for such teams is turning a recurring piece of work into something a whole team can rely on: an agent, focused on one clear purpose, with the right knowledge, boundaries and human oversight in place from the start.
Copilot Agents: From Idea to Action gives public sector professionals that next step. It helps you build the business case for an agent, design it with oversight and validation built in, then govern it responsibly once it's live. The programme fully reflects Government Digital Service guidance published in August 2026, which sets out what user-configured AI agents need to be justified, designed and used responsibly.
-
Build a clear understanding of where Copilot agents can add value to your team’s work.
Identify and prioritise suitable agent opportunities from real workplace challenges.
Design and build an agent using Microsoft 365 Copilot Agent Builder.
Define trusted knowledge, clear boundaries, ownership and human oversight.
Test agents systematically and establish practical ways to review and improve them over time.
-
A workplace agent prototype ready for further experimentation.
A practical Agent Canvas that can be reused across future agent opportunities.
Clear guardrails, ownership and human checkpoints.
Greater confidence in deciding when an agent is appropriate and when it is not.
A structured approach to testing, sharing and governing agents responsibly.
-
Discovery and Agent Opportunity Mapping
A structured session to:Understand the team's work, priorities and current Copilot use.
Identify recurring tasks, information needs and potential agent opportunities.
Explore relevant governance, information and organisational constraints.
Agent Design and Build Sprint
An immersive full-day workshop where the team:Identifies and assesses high-value agent opportunities.
Designs an agent using the Agent Canvas.
Builds the agent using Microsoft 365 Copilot Agent Builder.
Defines trusted knowledge, guardrails, human checkpoints and accountability.
Stress-tests the agent against real-world scenarios.
Establishes success measures and a practical plan for responsible use.
Follow-up Agent Clinic
A two-hour virtual session approximately one month later to:Share early results and lessons learned.
Refine agents, knowledge and guardrails based on experience.
Review whether agents should remain private, be shared more widely or require additional governance or technical support.
Plan the next iteration of agent development.
-
09:30 – Welcome and orientation
Understanding the move from everyday Copilot use to reusable Copilot agents.
Clarifying where Microsoft 365 Copilot Agent Builder fits.
10:00 – Find the opportunity
Identifying recurring work and information needs where an agent could create meaningful value.
Starting with the work, before the technology.
10:30 – Agent or not?
Testing potential use cases against prompts, Copilot, agents, more advanced solutions and human-led work.
Exploring where human judgement must remain central.
11:00 – Agent suitability and risk
Assessing value, consequence of error, sensitivity, governance and potential misuse.
Identifying use cases that require additional organisational review.
11:20 – Break
11:35 – Design your agent
Using the Agent Canvas: Aim → Governance → Expertise → Needs and boundaries → Test and transferDefining purpose, users and success.
Identifying trusted knowledge.
Establishing boundaries, guardrails and human checkpoints.
12:30 – Lunch
13:15 – Build your agent
Creating and configuring the agent using Microsoft 365 Copilot Agent Builder.
Developing effective instructions, knowledge and starter prompts.
14:15 – Govern and refine
Defining ownership, accountability and appropriate human oversight.
Deciding what the agent can support and what humans must verify or decide.
15:00 – Break
15:15 – Break your agent
Stress-testing the agent against ambiguity, missing information, edge cases and inappropriate requests.
Refining the agent based on what fails.
16:00 – From prototype to action
Reviewing readiness for wider experimentation.
Defining success measures, ownership, sharing and next steps.
16:30 – Close
