Free AB-100 Practice Questions With Full Rationale

Correct: C. This matches the domain objective directly: evaluate whether to build, buy, or extend using ROI criteria including total cost of ownership, before committing to either path.

• Why not A: Jumps to a six-month build without comparing it against extending the existing connector — skips the ROI step entirely.
• Why not B: Accepts a known 30% gap without evaluating whether extending the connector could close it more cheaply than a full build.
• Why not D: Delays the business need indefinitely instead of making a decision — avoids the analysis rather than doing it.

Correct: B. The source is topically correct and online — the problem is purely that a monthly refresh cadence can't keep up with weekly price changes. Tightening the sync interval or moving to a near-real-time feed addresses this directly.

• Why not A: The pricing data is exactly the right topic — relevance is not the issue, staleness is.
• Why not C: Nothing in the scenario indicates downtime; the source is reachable, just refreshed too infrequently.
• Why not D: No formatting or malformed-record issue is described — this is a timing problem, not a data-quality-format problem.

Correct: C. This is exactly the use case for a customized small language model: narrow task, strict latency, and a private-network deployment constraint that a small model can meet without external calls.

• Why not A: A third-party hosted API is an external call by definition — violates the no-external-calls constraint outright.
• Why not B: General-purpose generative summarization isn't built for a strict, low-latency, near-deterministic redaction task, and doesn't address network isolation.
• Why not D: A shared central model serving all teams adds queueing and contention that risks breaking the 200ms budget under load.

Correct: B. Topics with conversational language understanding preserve exact, auditable responses for defined intents, while a generative fallback topic — the fallback design pattern Copilot Studio is built for — catches the phrasing gaps.

• Why not A: Generative-only orchestration sacrifices the exactness and auditability the defined intents specifically require.
• Why not C: No fallback means any phrasing outside the exact keyword set goes completely unhandled — the gap the scenario describes.
• Why not D: Skips Copilot Studio's own fallback-topic capability and over-relies on escalation for a problem the platform is designed to handle.

Correct: B. MCP is the standard for exposing tools and data sources to an agent — reusing the warehouse system's existing MCP-compatible interface avoids custom connector work entirely.

• Why not A: Ignores the already-standardized interface and duplicates work that MCP extensibility would avoid.
• Why not C: A2A is for coordination between autonomous peer agents, not for consuming a tool/data interface — the wrong standard for this job even though it sounds plausible.
• Why not D: Manual re-entry defeats the point of a live integration and ignores the standardized interface that already exists.

Correct: A. Computer Use automates UI interaction on systems with no API, which fits the no-API portal. Because the sequence is fixed and repeatable with no independent decision-making, it belongs to a task agent, not an autonomous one.

• Why not B: Labeling a fixed, no-decision sequence as autonomous over-scopes the design — autonomous agents are for tasks requiring independent judgment.
• Why not C: A custom-trained model to browse a UI is significant overkill when Computer Use already exists for exactly this scenario.
• Why not D: Unnecessarily blocks a solvable problem on an external dependency instead of using an available Copilot Studio capability.

Correct: B. This is the ALM process objective directly: a defined environment structure and promotion path for Copilot Studio agents, connectors, and actions, not ad hoc production edits.

• Why not A: A backup reduces blast radius but doesn't create an actual test environment — production is still the first place changes run.
• Why not C: Verbal peer review is not a substitute for having a test environment at all.
• Why not D: Overly restrictive and doesn't fix the missing environment structure — it just stops iteration instead of enabling it safely.

Correct: B. The telemetry directly ties the slowdown to a tripling of grounding-data lookups per conversation — that is the signal to chase first, before broader or more disruptive changes.

• Why not A: Possible, but not what the given telemetry actually points to — investigating it first ignores the specific signal provided.
• Why not C: Not supported by any data in the scenario — a distraction from the lookup-volume signal that was actually measured.
• Why not D: Disproportionate before root-causing a specific, already-identified telemetry signal.

Correct: B. This directly satisfies the data residency and movement compliance requirement — processing must stay within the required jurisdiction, which is an architecture decision, not an afterthought.

• Why not A: Model version currency has nothing to do with where data is processed — doesn't address residency at all.
• Why not C: Anonymization is not a reliable substitute for residency compliance under most frameworks this objective refers to — a risky assumption to design around.
• Why not D: Contradicts the objective directly — data residency and movement compliance explicitly covers AI grounding data, not just structured records.

Correct: B. This matches the vulnerability-mitigation and audit-trail objectives directly: detect and mitigate prompt manipulation from untrusted input, and maintain an audit trail of changes to instructions and data.

• Why not A: Rate-limiting addresses volume or abuse, not the prompt-injection vulnerability itself.
• Why not C: A larger context window has no relationship to detecting or blocking manipulated instructions.
• Why not D: Removing functionality entirely is a disproportionate fix that breaks legitimate use instead of architecting a real mitigation.