Skip to main content

Agentic Harness Migration in Copilot Studio

Class Details

2 Days, last updated Nov, 2026
Copilot Studio Microsoft Copilot Harness GitHub Copilot Harness Copilot Studio Agents Claude Code Agentic Upgrade Copilot Studio Hooks Copilot Credits Advanced

An agent keeps the harness it was created on, so every move to the Microsoft Copilot harness is a rebuild, paid in Copilot Credits. Microsoft sets the end date of the old UI, which makes a planned migration the better course.

Day 1 introduces the harness and its agent loop, maps every Standard building block by purpose and sets up evaluation as the proof that a rebuilt agent does its job.

Day 2 hands the migration to a coding agent. It clones and reviews your Standard agents, drafts each migration plan for your approval and builds the new agent, proven against the Standard agent’s test set. Pull requests, deployment pipelines and an evaluation gate carry it to cutover.

Audience

  • Enterprise and Solution Architects for Agentic AI
  • Decision Makers who own an estate of Copilot Studio agents
  • Platform and Governance Leads for Copilot Studio

Prerequisites

  • Standard agents running in Copilot Studio
  • A Copilot Studio environment with Copilot Credits
  • GitHub User Account

Curriculum

Module 1: The Microsoft Copilot Harness

  • The Harness as Runtime
  • Two Harnesses and the One-Way Rule
  • Agent Loop: Plan, Call, Check, Replan
  • Skills, Tools, Knowledge and Memory
  • Hooks, Triggers and the Agent Sandbox
  • Build, Preview, Evaluate, Monitor
  • Guardrails You Design Yourself
  • Copilot Credits: Build, Test, Run

Module 2: Migration as Re-Architecture

  • The Business Case per Agent
  • Task Weight and Credit Cost
  • Topics as Skills, Workflows or Hooks
  • Agent Flows Rebuilt as Workflows
  • Deterministic Steps on the New Harness
  • Agents That Stay Standard
  • Three Routes to a Migrated Agent

Module 3: Evaluation as Proof of Migration

  • Test Sets as Evidence
  • Two Evaluation Surfaces
  • Graders and What They Measure
  • Portable Test Sets in CSV
  • Reading a Score Honestly
  • From Low Rating to Fix
  • Checking Passed Cases on Monitor

Module 4: Agentic Assessment of the Agent Estate

  • Copilot Studio Plugins for Coding Agents
  • Agents Cloned as Files
  • Advisor Review with Ranked Findings
  • The Purpose Map per Agent
  • Baseline Evaluation per Agent
  • Migrate, Wait or Stay Standard

Module 5: Agentic Migration

  • Describe, Plan, Approve, Build
  • Steering the Resumable Migration Plan
  • Topics Turned into Skills
  • Actions Marked for Manual Rebuild
  • One Test Set on Both Agents
  • Fix Loop: Change, Push, Evaluate

Module 6: Governance, Release and Cutover

  • Pull Requests for Every Change
  • Schema Validation Before Push
  • Solutions and Deployment Pipelines
  • Evaluation Gate Before Promotion
  • Maker Credentials and Credit Limits
  • Publishing to Microsoft 365 and Teams
  • Cutover with Standard in Service

Class Logistics

Day rate 1.250 EUR · online only, +20% VAT, reverse charge possible
Material Labs, decks · 35 EUR per day per participant
Max Users 16
Lab Requirements GitHub Codespaces / DevContainer
AI Harness: Claude Code or GitHub Copilot CLI