AB-100 flashcards
AB-100 · Build recall with a focused review session.
32 cards. Read the term and the definition here. The deck at the top of this page is optional.
- AI Center of Excellence
- A cross-functional team that owns AI standards, reusable patterns, governance reviews, and skilling so individual business units do not rebuild the same agent capabilities or repeat compliance mistakes.
- CAF AI adoption phases
- Microsoft's Cloud Adoption Framework for AI sequences adoption as Strategy, Plan, Ready, Adopt, Govern, and Manage; the Plan phase produces the prioritized use case portfolio, roadmap, and skills plan.
- Use case qualification criteria
- A viable agent use case has a named process owner, a measurable baseline KPI, permission-trimmed source content, and a bounded scope that a single agent can complete end to end.
- Agent ROI metric
- The business measure an agent is expected to move, captured as a baseline before launch and re-measured after, such as handle time, first-contact resolution, or processing cost per case.
- Autonomous agent trigger
- An event or schedule that starts an agent run with no user present, such as a new Dataverse row, a SharePoint item, or a recurrence, so the agent must be designed for unattended execution.
- Human-in-the-loop approval
- A design control that pauses an agent action and routes it to a person for confirmation before execution, used when the operation is irreversible or financially material.
- Responsible AI six principles
- Microsoft's Responsible AI Standard names fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability; each use case impact assessment must show how these are addressed.
- Copilot Studio consumption metering
- Copilot Studio agent usage is metered separately from Microsoft 365 Copilot licenses; capacity is purchased as prepaid packs or enabled as pay-as-you-go billing in the Power Platform admin center.
- Microsoft 365 Copilot license scope
- A per-user Microsoft 365 Copilot license enables in-product Copilot experiences in apps such as Word, Outlook, and Teams for that user; it does not by itself cover Copilot Studio agent consumption.
- Data readiness gate
- Before build, confirm each knowledge source exists, is permission-trimmed to the intended audience, refreshes on a known schedule, and has a named owner accountable for its accuracy.
- Agent versus chatbot
- A chatbot answers within predefined flows and menus, while an agent plans, selects tools, and executes multi-step tasks toward a goal, optionally with no user in the conversation.
- Generative orchestration
- An agent mode where a model reads the descriptions of topics, tools, and knowledge sources and chooses at runtime which to invoke, instead of matching a fixed trigger phrase.
- Knowledge source
- A configured content location the agent can search at answer time, such as a SharePoint site, Dataverse table, uploaded file, public website, or Azure AI Search index.
- Topic trigger phrases
- Example utterances that route a user message to a specific topic under classic orchestration; overlapping phrases across topics cause misrouting, so each topic needs distinct, tested phrases.
- System instructions
- The persistent prompt defining the agent's role, scope, tone, allowed and forbidden actions, refusal behavior, and citation expectations; generation and orchestration decisions are conditioned on it.
- Connector action
- A declared tool, built from a Power Platform connector, Power Automate flow, REST API, or MCP server, that the agent can call to read or change data outside the conversation.
- Retrieval-augmented generation
- A pattern that embeds and indexes source content, retrieves the top matching chunks for a question, and inserts them into the prompt so the model answers from grounded evidence.
- Connected agent
- A child agent called by a parent agent and scoped to one domain; the parent routes the request and the child answers using its own instructions, knowledge, and tools.
- Fallback topic
- The topic invoked when no other topic or tool matches or confidence is low; it should explain limits, offer escalation, and log the unmatched question for later tuning.
- Human escalation handoff
- A configured path that transfers a conversation to a live representative, typically through Dynamics 365 Customer Service or a handoff connector, preserving transcript context and reason codes.
- Declarative agent
- An agent defined only by instructions, knowledge, and actions that runs on the Microsoft 365 Copilot orchestrator, requiring no custom hosting or model and deployable through Teams or the agent store.
- Custom engine agent
- An agent whose orchestration and model are owned by the maker, hosted in Copilot Studio, Azure AI Foundry, or a service built with the Microsoft 365 Agents SDK, then surfaced in Microsoft 365 Copilot channels.
- Solution ALM pipeline
- A Power Platform deployment pipeline that promotes an agent as a solution from development to test and production, exporting unmanaged in development and importing managed downstream so customization is locked.
- Managed environment
- A Power Platform environment setting that adds sharing limits, solution checker enforcement, usage insights, and maker welcome content, typically applied to test and production environments hosting agents.
- Data loss prevention policy
- An environment-scoped Power Platform policy that classifies connectors as business, non-business, or blocked, preventing an agent from combining connectors that would move data across those boundaries.
- Application Insights telemetry
- An Azure monitoring target an agent can emit events to, capturing conversation transcripts, tool calls, latency, and errors so failures can be diagnosed beyond built-in analytics.
- Evaluation test set
- A fixed collection of representative questions with expected answers, run against an agent to compare versions and score groundedness, completeness, and correctness before publishing.
- Purview sensitivity label
- A Microsoft Purview classification applied to content and, where supported, to agent output, driving encryption, retention, and DLP so agent-generated material inherits the source's protection.
- Teams app manifest
- The JSON package definition, with icons and permissions, that must accompany an agent published to Teams or Microsoft 365 Copilot and that an administrator approves for org-wide distribution.
- Resolution rate
- The share of agent sessions in which the user's request is handled without escalation or abandonment, used as the primary post-deployment health metric for an agent.
- Publishing channel
- The surface where a published agent is reachable, such as Teams, Microsoft 365 Copilot, a website, Direct Line, or a custom app; available capabilities and authentication vary by channel.
- Environment strategy
- The separation of development, test, and production Power Platform environments so that agent changes, connector policies, and data connections are validated before reaching business users.