Copilot Agents Workshop Series
Who This Is For
This series is for people who have gotten as far as individual Copilot use will take them. They prompt fluently, iterate quickly, and use Copilot daily for research, drafting, comparison, and thinking through problems. But every task still starts from a blank chat, the insights that would help their team live in one person's history, and the workflows they've refined can't be handed to anyone else.
Nothing transfers. Nothing scales. The next move is building. Agents package a repeatable workflow, carry organizational context, and scale what one skilled user already knows across a team. This series teaches how to scope, build, and launch them, and how to recognize when an agent isn't the right answer.
The Problem This Solves
Your strongest Copilot users have hit a ceiling. They're getting real value from individual use, but every workflow they've refined lives in their head. Nothing transfers. Nothing scales.
Meanwhile, there's growing interest in agents across the organization, but no shared language for evaluating whether a use case actually fits, no repeatable process for building one well, and no framework for launching one responsibly. Without structure, you get abandoned prototypes, agents that hallucinate because nobody scoped the knowledge layer, and pilot fatigue.
A self-paced course can't solve this. Agent-building requires real use cases, hands-on configuration, live troubleshooting, peer feedback on design decisions, and facilitator judgment on scoping. This is skilled work that benefits from a cohort moving through it together, each building a relevant agent.
What Learners Will Be Able to Do After
From
"I've heard agents might help, but I don't know if my use case fits, how to actually build one, or how to get anyone else to use it."
To
"I can scope an agent against a real problem, build a working v1, and run a disciplined pilot. I know when an agent is the wrong answer."
Specifically, participants will leave able to:
- Evaluate whether a workflow is ready for an agent, or whether a prompt, automation, or knowledge source is the better fit
- Map a workflow into agent behavior, knowledge requirements, and system dependencies
- Build a working agent in Copilot Studio (Lite or Full) with configured knowledge, instructions, capabilities, and suggested prompts
- Design and execute a test plan that covers normal use, edge cases, and boundary conditions
- Translate pilot results into a launch plan that sets accurate expectations with users
- Assess risk, plan for adoption, and set up lightweight support structures
Series Structure
The series runs over six weeks: two in-person morning sessions to scope and build (one day apart), four weeks of office hours while participants iterate and test, and a virtual closing session on launching. Each participant works on their own agent throughout the series, based on a specific problem they bring in. Participants should expect to dedicate meaningful time outside the live sessions to scoping, building, testing, and iterating on their agent.
Workshop Descriptions
Workshop 1: Scoping & Designing
In person, ~3 hours Participants bring a specific problem they want to solve with an agent and work through a structured scoping process. They learn to distinguish agents from prompts, automations, and knowledge sources. They extract the workflow their agent would follow, assess data and systems readiness, identify governance requirements, and leave with a scoped design and a realistic build plan. Homework: Finish the scoping worksheet (started during the session). This means finalizing the workflow mapping, confirming knowledge sources exist and are accessible, and resolving any readiness flags with IT or data owners. Written facilitator feedback is provided on each participant's completed worksheet.
Workshop 2: Building
In person, ~3 hours, next day Participants move from design to build inside Copilot Studio. The session walks through knowledge configuration, writing agent descriptions and instructions, setting up capabilities, and designing suggested prompts. Each configuration step includes a smoke test so participants catch issues in real time. A helper agent converts scoping worksheet inputs into first-draft instructions, giving participants a starting point rather than a blank page. Homework: Finish the test plan worksheet (started during the session), then spend time in Copilot Studio iterating on their agent and running tests against it. Document results in the test results worksheet: what works well, where it falls short, and the one prompt that best showcases what the agent can do.
Office Hours
Virtual, 60 minutes each, weeks 2 through 5 Four drop-in sessions for 1:1 consulting support while learners are building their agents and testing them. Participants come with specific questions, and the goal is to unblock them. These sessions also surface common issues that Workshop 3 can address directly.
Workshop 3: Launching
Virtual, 90 minutes, week 6 Participants arrive with completed test results. The session covers how to introduce an agent to users (leading with the best demo, naming what it can and can't do), how to assess risk and decide whether the agent is ready to launch or needs more iteration, how to plan for adoption and behavior change, and how to set up support and monitoring. A launch kit of communication templates goes home with participants. Homework: Build out the launch plan using the templates and frameworks from the session.
What's Included
Live Sessions (7)
- Two in-person workshops (Scoping & Designing + Building), each ~3 hours, one day apart
- One virtual workshop (Launching), 90 minutes
- Four optional virtual office hours sessions (60 min each, weeks 2 through 5) for 1:1 consulting support while learners are building their agents and testing them. Additional sessions can be scheduled based on enrollment.
Worksheets (3)
- Scoping worksheet for mapping use cases, workflows, readiness, and build plans (started in Workshop 1, completed as homework)
- Test plan worksheet for defining test cases, success criteria, and evaluation approach (started in Workshop 2, completed as homework)
- Test results worksheet for documenting pilot findings, limitations, and the best demo prompt (completed as homework after Workshop 2)
Quick Reference Guides (4)
- Copilot Lite vs. Full Studio comparison (capabilities, tools, evaluation, channels, analytics)
- Instructions best practices QRG (structural rules, common pitfalls, iteration process)
- Knowledge pro tips and limitations (what's searchable, what to exclude, grounding behavior)
- Agent Testing reference card (exact match, keyword, semantic similarity, tool verification, with "when to use" guidance)
Launch Kit (3 Templates)
- Launch email template
- Team-meeting announcement template
- Usage document template
Tools and Feedback
- Helper agent that converts scoping worksheet inputs into first-draft agent descriptions and instructions
- Written facilitator feedback on each participant's scoping worksheet
Why This Format
This cannot be compressed into a single 90-minute session. Here's why.
Scoping must precede building
Most agent projects fail at the scoping stage. People jump to building without determining whether the workflow is defined enough, whether the data is accessible, or whether an agent is even the right solution. Workshop 1 exists to prevent that. Workshop 2 builds directly from its output.
The content is high cognitive load
Scoping and building are tightly coupled, but cramming both into a single day creates fatigue that undermines quality. one day apart keeps the scope fresh while giving participants time to absorb design decisions before they start configuring.
The build-and-test period gathers real data to inform governance
Workshop 3 (launching) only works when participants have real pilot data and real adoption friction to plan against. Theory about risk assessment and change management is abstract without it. The four weeks between building and launching give participants enough time to iterate on their agent, run meaningful tests, and surface real issues while balancing their other responsibilities.
Office hours maintain forward progress
Most people who give up on an agent build do so when they hit a wall alone. Four weekly touchpoints during the build-and-test period keep momentum and surface issues early.
Prerequisites
You will need to provide:
- Copilot Studio access: All participants need full Copilot Studio access (Lite at minimum; Full preferred for participants whose scoped agents require skills, tools, or agent evaluation). Access must be confirmed and tested before Workshop 1.
- In-person space: A room for two consecutive mornings (Workshops 1 and 2) with reliable internet, power for laptops, and enough space for partner work.
- Participants with a specific problem to solve: Each participant must arrive at Workshop 1 with a workflow or business problem they want to build an agent for. They will work on this throughout the series. This is not a theoretical exercise, and participants who arrive without a use case will not get full value from the program.
- Use case validation prior to enrollment: At the time of enrollment (at least four weeks before Workshop 1), each participant submits one or more agent ideas. Mellonhead and an IT leader reviews these together to confirm that at least one is buildable in full or as a proof of concept, given current systems and data access. If none are feasible, there's time to discuss alternatives before the series begins. This prevents participants from arriving with a use case that's blocked by dependencies they can't resolve.
- IT contacts for dependencies: Some agents will require data access, integrations, or permissions that participants can't grant themselves. Having a named IT contact available during the series (not necessarily in the room, but reachable) prevents builds from stalling on access issues.
- Travel reimbursement: You provide travel reimbursement for 1-2 facilitators for the in-person workshops.
Mellonhead Labs