AB-100 ALM and components
Each Microsoft platform has its own unit of deployment and its own way to promote to production. This page lines them up, then covers Copilot Studio's components and the Power Platform ALM toolchain.
1. ALM by Platform (Master Table)
Each Microsoft platform has its own deployment unit, source control story, and promotion mechanism. The exam tests whether you can match the right ALM pattern to the right platform.
| Platform | Deployment Unit | Source Control | Promotion Method | IaC Support |
|---|---|---|---|---|
| Copilot Studio | Dataverse solutions | SolutionPackager to Git | Power Platform pipelines | No |
| Microsoft Foundry | Git artifacts | Native Git (repos) | Bicep / ARM templates | Yes |
| D365 CE (Service, Sales) | Dataverse solutions | Solution export/import | Power Platform pipelines | No |
| D365 F&O (Finance, SCM) | LCS packages | Lifecycle Services (LCS) | LCS deployment | No |
| D365 F&O AI features | Feature toggles | Feature management workspace | N/A (toggle on/off) | No |
Critical distinctions
- Copilot Studio is NOT Foundry. Copilot Studio uses Dataverse solutions; Foundry uses native Git artifacts and supports IaC (Bicep/ARM).
- CE is NOT F&O. Customer Engagement apps (Service, Sales) use Dataverse solutions and Power Platform pipelines. Finance & Operations apps use LCS packages and LCS deployment -- completely different toolchain.
- F&O AI features are just toggles. They do not go through LCS or pipelines. You enable/disable them in the Feature management workspace.
- Only Foundry supports IaC. Every other platform: No.
Decision Rule: "How do I promote this to production?"
Is it a Copilot Studio agent or D365 CE customization?
--> Power Platform pipelines (Dataverse solutions)
Is it a Foundry project?
--> Bicep/ARM deployment (Git-native)
Is it D365 Finance & Operations?
--> LCS packages and LCS deployment
Is it an F&O AI feature?
--> Feature management workspace toggle2. Copilot Studio Component Taxonomy
When building a Copilot Studio agent, every capability maps to exactly one component type. The exam gives you a scenario and expects you to name the component.
| Need / Scenario | Correct Component |
|---|---|
| Structured conversation flow (step-by-step dialog) | Topics |
| Inline AI task (summarize, classify, extract) | Prompt actions |
| External system call (REST API, connector) | Agent flows |
| Collect required info before proceeding | Slot filling |
| Route user utterance to the correct topic | Generative orchestration |
Decision Rule: "Which component do I use?"
User says something --> Generative orchestration (routing)
|
v
Need structured steps? --> Topic
Need to call an API? --> Agent flow
Need AI to summarize/classify? --> Prompt action
Need to gather required fields? --> Slot filling (within a topic)Testing Flows -- Where You Test What
| What you are testing | Where you test it |
|---|---|
| Agent orchestration, topics, prompt actions | "Test your agent" panel in Copilot Studio |
| Power Automate desktop flows (RPA) in isolation | Power Automate desktop console |
The "Test your agent" panel tests the full orchestration stack. The Power Automate desktop console tests individual desktop flows outside the agent context.
3. Declarative Agents vs. Copilot Studio Agents
Two different agent paradigms. The exam tests when to use which.
| Dimension | Declarative Agents (M365 Copilot) | Copilot Studio Agents |
|---|---|---|
| Ecosystem | M365 Copilot | Power Platform / standalone |
| Build method | JSON manifest + instructions | Low-code canvas (topics, actions, flows) |
| Best for | Extending M365 Copilot with custom knowledge/actions | Building standalone conversational agents |
| Channels | M365 apps only (Teams, Outlook, Word) | Any channel (web, Teams, D365, custom) |
| Governance | M365 admin center | Power Platform admin center + managed environments |
Decision Rule
"Extend M365 Copilot with custom grounding data" --> Declarative agent
"Build a standalone agent for customers on a website" --> Copilot Studio agent
"Agent needs to run in Outlook/Word natively" --> Declarative agent
"Agent needs to run on any channel" --> Copilot Studio agent4. Solution Components vs. Environment Variables
This is the single most important ALM distinction on the exam. The two concepts answer different questions.
| Concept | What It Is | What It Answers | Example |
|---|---|---|---|
| Solution components | The unit of portability across environments | "How do I move X between dev, test, and prod?" | Connectors (types 371/372), flows, agents, custom tables |
| Environment variables | Parameters WITHIN a component that change per environment | "How do I configure X differently in each environment?" | Host URL, Client ID, feature flags, thresholds, SharePoint site URLs |
Exam Pattern Recognition
| Exam phrasing | Answer |
|---|---|
| "Manage X across environments" | Solution components + Power Platform pipelines |
| "Different config per environment" | Environment variables |
| "Move agent from dev to prod" | Solution export (managed) + pipeline |
| "API endpoint changes per environment" | Environment variable |
What Counts as an Environment Variable?
Not just API URLs. The exam tests breadth:
- API host URLs
- Client IDs / secrets (via Key Vault reference)
- Feature flags (on/off per environment)
- Thresholds (e.g., approval limits)
- SharePoint site URLs
- Connection references
Trap: "Only API URLs are environment variables" is WRONG. Feature flags, thresholds, and SharePoint URLs also change per environment.
5. Power Platform Pipelines and Quality Gates
Pipelines
Power Platform pipelines move Dataverse solutions between environments with governance controls. They are the promotion mechanism for both Copilot Studio and D365 CE.
Dev environment --> [Pipeline stage] --> Test environment --> [Pipeline stage] --> Prod environment
| |
(optional gate) (optional gate)Quality Gates vs. Version Control
| Mechanism | Purpose | Blocks deployment? |
|---|---|---|
| Quality gates (pipeline approvals) | Enforce that a compliance review HAPPENS before promotion | Yes -- deployment is blocked until approved |
| Version control | Track change history, enable rollback | No -- does not block deployment |
Decision Rule
"Ensure compliance review before production" --> Quality gate (pipeline approval)
"Track changes and enable rollback" --> Version control
"Block deployment until reviewer approves" --> Quality gate
"See who changed what and when" --> Version control6. Solution Awareness
| Type | Meaning | ALM Impact |
|---|---|---|
| Solution-aware | Component can be added to a Dataverse solution | Portable across environments; participates in ALM |
| Non-solution-aware | Component is stuck in one environment | Cannot be promoted via pipelines; must be manually recreated |
If a component is non-solution-aware, it cannot participate in ALM pipelines. This matters for planning: you must account for manual steps or workarounds for non-solution components.
7. Managed Environments and ALM Governance
Managed environments are a Power Platform admin center feature that adds governance controls on top of standard environments.
| Capability | What It Does for ALM |
|---|---|
| Sharing limits | Control who can share apps/flows |
| Usage insights | Track adoption and usage |
| Data policies | Enforce DLP at the environment level |
| Pipeline integration | Required for production pipeline stages in some configurations |
| Maker welcome content | Onboard makers with governance guidance |
Managed environments are the governance layer for Copilot Studio agents. Declarative agents (M365 Copilot) are governed through the M365 admin center instead.
8. Analytics vs. Evaluations
Two different measurement systems for two different questions.
| System | Platform | What It Measures | When to Use |
|---|---|---|---|
| Copilot Studio Analytics | Copilot Studio | Usage, adoption, session metrics, CSAT | "How are users interacting with the agent?" |
| Foundry Evaluations | Microsoft Foundry | Model quality, accuracy, groundedness, relevance | "How good are the AI model's outputs?" |
Decision Rule
"Track how many users are engaging with the agent" --> Copilot Studio Analytics
"Measure whether the model's answers are accurate" --> Foundry Evaluations
"Monitor adoption trends over time" --> Copilot Studio Analytics
"Evaluate groundedness and relevance of responses" --> Foundry Evaluations9. Common Exam Traps (Yes/No)
| Statement | Correct? | Why |
|---|---|---|
| "Copilot Studio uses Git repos for source control" | No | Copilot Studio uses Dataverse solutions. SolutionPackager CAN export to Git, but the native unit is a Dataverse solution, not a Git repo. |
| "Only API URLs are environment variables" | No | Feature flags, thresholds, SharePoint URLs, connection references -- all can be environment variables. |
| "Quality gates and version control both block deployment" | No | Quality gates block deployment (require approval). Version control tracks history but does not block. |
| "Foundry and Copilot Studio use the same ALM" | No | Foundry = Git + Bicep/ARM + IaC. Copilot Studio = Dataverse solutions + PP pipelines. Completely different. |
| "D365 CE and D365 F&O use the same deployment mechanism" | No | CE = Dataverse solutions + PP pipelines. F&O = LCS packages + LCS deployment. |
| "Non-solution-aware components can be promoted via pipelines" | No | Only solution-aware components participate in pipeline-based ALM. |
10. Quick-Fire Recall Table
| If the exam says... | The answer is... |
|---|---|
| Move Copilot Studio agent to prod | Dataverse solution + Power Platform pipeline |
| Deploy Foundry project | Bicep/ARM template (IaC) |
| Promote D365 Sales customization | Dataverse solution + Power Platform pipeline |
| Deploy D365 Finance update | LCS package |
| Enable F&O AI feature | Feature management workspace toggle |
| API URL changes per environment | Environment variable |
| Ensure review before prod deployment | Quality gate (pipeline approval) |
| Track change history | Version control |
| Structured conversation in Copilot Studio | Topic |
| AI summarization in Copilot Studio | Prompt action |
| Call external API in Copilot Studio | Agent flow |
| Gather required info from user | Slot filling |
| Route utterance to correct topic | Generative orchestration |
| Extend M365 Copilot | Declarative agent |
| Build standalone web agent | Copilot Studio agent |
| Measure user adoption | Copilot Studio Analytics |
| Measure model accuracy | Foundry Evaluations |