Free NCA-AMAD Agentic AI Practice Questions

Correct answer: C. The agent is goal-directed and plans and executes multi-step tasks using tools

C is correct because traditional generative AI apps follow a request-and-respond pattern, while agents coordinate models with external tools to plan and carry out multi-step work toward a goal. A is incorrect because agents are defined by their use of tools, APIs and data sources, the opposite of answering from training data alone. B is incorrect because a fixed decision tree describes a rule-based system; agents reason and choose their own steps. D is incorrect because model size does not define an agent; agents can run on small open models or frontier models alike.

Correct answer: D. An agent harness

D is correct because NVIDIA describes the harness as the scaffolding that gives the LLM the ability to act: it connects short- and long-term memory, knows which tools are available and can create new skills if allowed. A is incorrect because the secure runtime is the policy-enforced sandbox in which the agent executes, and the stem says the prototype already has one. B is incorrect because the LLM already reasons well; a bigger model still could not reach memory or tools without the scaffolding around it. C is incorrect because a planning module breaks complex tasks into steps; it does not connect the LLM to memory or tools.

Correct answer: D. A long-running agent that operates continuously on tool-driven workflows

D is correct because long-running agents are always-on coworkers that keep context across sessions and execute multi-step workflows, such as an AIOps copilot that monitors systems and triggers fixes. A is incorrect because a single call keeps no context between alerts and cannot carry out a multi-step remediation workflow. B is incorrect because self-evolving agents add continuous self-improvement of prompts, tools and memory, which the team leaves to its engineers. C is incorrect because a simple reflex agent has no memory, so it cannot keep context across sessions or plan multi-step fixes.

Correct answer: A. The ability to interpret varied, unstructured claim documents without new rules

A is correct because LLM agents can process unstructured data from diverse document formats without explicit programming, which is exactly where rule-based RPA breaks. B is incorrect because the problem is rigidity, not speed; running the same rules faster still fails on new layouts. C is incorrect because LLM steps are probabilistic; agents add flexibility, not a guarantee of matching the old rules. D is incorrect because agent workflows still keep the overall flow defined by business logic; the LLM handles ambiguity inside steps.

Correct answer: D. The agent core

D is correct because the agent core is the central coordination module where goals, the tool "user manual", guidance on planning modules, relevant memory and an optional persona are assembled. A is incorrect because tools are the executable workflows the agent calls; the core only describes them. B is incorrect because the memory module stores past actions and interactions; the core only pulls in the relevant items from it. C is incorrect because the planning module performs task decomposition and reflection; the core explains when to use it.

Correct answer: A. Nothing in it plans, calls tools or acts on results; it still only responds

A is correct because agentic AI is a shift from systems that respond to systems that act, using models, tools, memory and guardrails; a bigger model in the same chat window is not agentic. B is incorrect because many agents use off-the-shelf models with prompts and tools; fine-tuning is optional. C is incorrect because an agent can be powered by a single model; using tools and acting on results is what matters. D is incorrect because latency does not define whether a system is agentic; many agents tolerate higher latency.

Correct answer: A. Move the agent onto an agent framework or harness that provides those features

A is correct because plain Python suits simple workflows and experiments, while frameworks and harnesses matter for complex production workflows because they provide telemetry, logging and evaluation without building them from scratch. B is incorrect because fine-tuning changes model behavior but provides none of the required telemetry, logging or evaluation. C is incorrect because longer prompts do not add telemetry, logging or evaluation, and hand-building those features strains a small team. D is incorrect because rule-based pipelines lose the flexibility the agent was built for, and the requirement is observability, not rigidity.

Correct answer: B. A deterministic rule-based program that applies the published formula

B is correct because the logic is fixed and the data is structured, so rule-based automation gives exactly reproducible, auditable results; LLM agents add value where steps are ambiguous or data is unstructured. A is incorrect because a self-critique loop reduces some errors but still cannot guarantee identical, auditable outputs. C is incorrect because LLM arithmetic can vary between runs, which conflicts with the requirement for identical results. D is incorrect because planning adds nondeterminism and cost to a task with no ambiguity, which works against the reproducibility requirement.

Correct answer: C. The overall order of tasks, set by business logic; the LLM works inside each step

C is correct because in a workflow, a complex task is broken into definite, constrained paths dictated by business logic; LLMs resolve ambiguity within each subtask, but the larger flow is predetermined. A is incorrect because responses are still generated per request; only the order of tasks is predetermined. B is incorrect because letting the model choose the order of every step describes an exploratory agent, not a workflow. D is incorrect because weights are frozen at inference in any LLM application; that is not what distinguishes a workflow.

Correct answer: C. Tool outputs are added to the context, and their size depends on query and tool

C is correct because tool responses go directly into the context window, and their size varies with the specific query and the tool's design, so input length is no longer predictable. A is incorrect because the system prompt stays in place; tools add content rather than removing it. B is incorrect because the input side grows with tool results; output length is not randomized by tool use. D is incorrect because the tokenizer vocabulary is fixed by the model and does not change with tool use.