Training / Copilot Studio Practitioner — Standard Harness
Copilot Studio Practitioner — Standard Harness
Move a working agent to generative orchestration, extend it with your own connector, bound its sensitive actions, and prove it with a regression suite.
By the end of eighteen hours, every participant has migrated a working rule-based agent to generative orchestration, extended it with a custom connector they built themselves, placed explicit control boundaries around its sensitive actions, and proved its reliability with a regression suite. The shift is a change of craft, not of feature: instead of trigger phrases and hand-drawn branches, participants write descriptions, shape inputs and outputs, and set boundaries — and an AI planner decides the route.
Makers running agents in production who need to move to generative orchestration without degrading a working service.
Integration developers who need custom connectors and MCP servers that still work after publication.
Technical leads and reviewers who must decide which actions an AI planner may take unsupervised.
Consultants and delivery partners who need to explain, with evidence, why an agent chose what it chose.
First-time builders with no hands-on agent experience. Our Foundations — Standard Harness course is one route in.
The GitHub Copilot harness, agent skills or autonomous reasoning agents. Our Foundations — GitHub Copilot Harness course covers those.
Multi-agent topology or evaluation across an agent estate. Our Practitioner — GitHub Copilot Harness course covers those.
Building MCP servers. This course connects to them; it does not write them.
ALM, data loss prevention, tenant authentication or estate-wide licensing. Our Copilot Studio Enterprise course covers those.
- Duration
- 3 days or 9 sessions × 2 hours
- Level
- Practitioner · Standard Harness
- Delivery
- Remote or on-site
- Group size
- Up to 10
- Price
- Priced per team — ask us
We reply within one working day. No sales sequence.
What your team will be able to do
Outcomes, not topics covered.
By the end, each of these is something they have done rather than watched.
Explain how generative orchestration selects topics, tools, connected agents and knowledge, and what weight each signal carries.
Write topic, tool and parameter descriptions that steer a planner reliably, and remove the overlap that makes selection unpredictable.
Restructure topics to return outputs rather than send messages, and use answered-state to stop duplicate responses.
Design input parameters so the orchestrator asks sensible follow-up questions unaided.
Read an activity map to diagnose why a planner selected what it did.
Configure connectors and connection references, and anticipate the authentication failures that appear only after publication.
Build, authenticate and register a custom connector as an agent tool.
Connect an MCP server and explain why it depends on generative orchestration.
Tune knowledge grounding and trace a hallucination back to its cause.
Place each action in the right control layer (fixed path, human approval, or planner) and defend the choice.
Build a regression suite and improve an agent one change at a time.
Read fallback rate, tool usage and transcripts to drive an improvement backlog.
What's covered
16 modules, a little under half of it building.
Labs run on your own tenant, and one agent is migrated across all three days, so the before-and-after is yours to show. Day one makes it worse on purpose: watching the planner misfire is what the rest of the course fixes. We publish what each day produces rather than a minute-by-minute timetable.
Day 1
Orchestrate
Leaves with: A toolkit that steers the planner, with every misfire caught and evidenced
-
01
From trigger phrases to descriptions
What changes when generative orchestration is switched on, what becomes possible, and what quietly stops working.
-
02
How the orchestrator selects
The signals a planner reads and their weight, multi-step planning, and a first reading of the activity map.
-
03
Lab 1: Switch it on and watch it break
Baseline a working agent, enable generative orchestration, and catalogue every misfire from activity-map evidence.
Hands-on -
04
Writing descriptions that steer
The writing rules a planner obeys, the failure patterns behind wrong selections, and a rewrite workshop.
-
05
Lab 2: Rewrite the toolkit
Audit, merge and rewrite descriptions, rename tools and parameters, and measure how many misfires are resolved.
Hands-on
Day 2
Compose and integrate
Leaves with: Topics that compose, and a custom connector the planner picks unprompted
-
06
Inputs, outputs and the answered-state pattern
Topics that return rather than answer, answered-state against duplicate replies, and inputs that drive follow-up questions.
-
07
Lab 3: Restructure topics to compose
Convert topics to return outputs, add answered-state, and force an answer drawn from two sources.
Hands-on -
08
Connectors, connections and authentication
Connection references, per-connector authentication, and the failures that appear only after publication.
-
09
Lab 4: Build a custom connector
Define, authenticate, register and describe a connector, then confirm the planner selects it unprompted.
Hands-on -
10
MCP servers
What a server exposes, dynamic discovery, why it needs generative orchestration, and who is responsible for what it does.
Day 3
Harden and prove
Leaves with: An agent that knows what it may decide alone, and a regression suite to prove it
-
11
Knowledge depth and grounding quality
Proactive search, intercepting retrieval, and tracing a hallucination back to what the agent was given.
-
12
Control layers and safe boundaries
Fixed path, human approval or planner: placing each action, with a workshop on the contested cases.
-
13
Lab 5: Grounding and guardrails
Fix a grounding failure, filter a knowledge search, lock a sensitive action to a fixed path, and add confirmation and redaction.
Hands-on -
14
Testing, analytics and the improvement loop
Single-change iteration, transcripts, and the metrics that turn usage into a backlog.
-
15
Lab 6: Prove it
A regression suite, a documented single-change fix, and the before-and-after from the Day 1 baseline.
Hands-on -
16
Solutions, responsible AI and next steps
Moving agents between environments and telling users content is AI-generated, then the capstone.
Upcoming batches
No public batch scheduled yet.
Teams book this course on their own dates. Joining as an individual? Tell us and we'll let you know when the next batch opens.
Before the first day
What we need from you.
Prerequisites
- Working experience building agents on the standard harness: topics, nodes, variables, knowledge and publishing. Our Foundations course is one route, not a requirement.
- No coding. The custom connector is built in the connector designer, though reading an API specification helps.
- Copilot Studio access with permission to create and publish agents, and to enable generative orchestration.
- Permission to create a custom connector. This is a separate permission from creating an agent, and the one most often missing.
- Network access to the supplied Northwind inventory API and MCP server.
- Every participant completes a five-minute access check 48 hours before day one.
We send a checklist a fortnight ahead and will talk to your platform team directly if access needs arranging.
Next step
Tell us about your team.
Roughly how many people, what they build today, and when you'd want to run it. We come back with dates and a quote.
Check availability