UiPath AAAv1 Practice Questions with Explanations
Free UiPath AAAv1 practice questions. 68 of them, each with the correct answer, a full explanation, and the reason every other option is wrong.
AAAv1 practice questions
Question 1
A developer is implementing a few-shot structured prompt for an email classification task. The prompt includes examples of email subjects labeled with their respective classifications, such as “Spam” or “Work”.
What is the most important aspect to consider when selecting examples for the prompt?
- A. Use random and unrelated examples to test the prompt’s robustness.
- B. Always use more than 10 examples, regardless of task complexity.
- C. Choose examples that are diverse, relevant, and typical of the task’s expected input.
- D. Include examples with intentionally incorrect labels to improve training.
Show answer and explanation ▾
Correct answer: C
In a few-shot prompt, examples should be diverse, relevant and typical of the real inputs so the model learns the task, not quirks.
Why the other options are wrong:
- A. Random and unrelated examples teach the model nothing about the task.
- B. There is no rule to always use more than 10 examples.
- D. Incorrect labels would teach the model wrong behavior.
Question 2
An agent uses Web Search, Slack integration, and a custom process to resolve IT support tickets. The agent must: a. Retrieve relevant troubleshooting steps from the web. b. Notify the user via Slack if a solution is found. c. Escalate unresolved tickets via a custom process.
Which evaluation strategy ensures comprehensive coverage while avoiding redundancy?
- A. Use random input sampling across all tools and rely on the default “LLM-as-a-Judge” assertion.
- B. Create 30 evaluations for Slack notifications, 30 for web searches, and 30 for escalation processes.
- C. Group evaluations into sets: Valid web results triggering Slack notifications, Invalid web results triggering escalations, Edge cases.
- D. Create more than 30 evaluations for Slack notifications, more than 30 for web searches, and more than 30 for escalation processes.
Show answer and explanation ▾
Correct answer: C
Grouping evaluations into sets by scenario (valid results, invalid results, edge cases) covers the whole flow without repeating tests for each tool separately.
Why the other options are wrong:
- A. Random sampling with only the default judge does not guarantee coverage of each path.
- B. Separate sets of 30 per tool repeat tests and ignore how the tools interact.
- D. More than 30 per tool adds more redundancy.
Question 3
What are the primary benefits of Context Grounding when querying data across multiple documents?
- A. Context Grounding is limited to querying within a single document at a time.
- B. Context Grounding only extracts random sentences without contextual understanding.
- C. Context Grounding requires manual intervention for identifying connections between data points across documents.
- D. Context Grounding understands relationships between data points across documents, enabling tasks like summarization, data comparison, and retrieval of highly relevant information.
Show answer and explanation ▾
Correct answer: D
Context Grounding understands relationships between data points across documents, which enables summarization, comparison and retrieval of relevant information.
Why the other options are wrong:
- A. It is not limited to a single document.
- B. It does not extract random sentences without context.
- C. It does not need manual work to connect data points across documents.
Question 4
An agent is built to extract customer feedback sentiment. You want to show the LLM how to classify it as ‘Positive’, ‘Neutral’, or ‘Negative’.
Which few-shot design is most helpful?
- A. Input: “The app is okay I guess.” → Output: “Text”
- B. Input: “I love the new design, very intuitive!” Output: “Positive” Input: “Nothing special, just works.” Output: “Neutral” Input: “Terrible experience, won’t use again.” Output: “Negative”
- C. List words like: “great, okay, bad” and map them to tone.
- D. Use a multiple-choice table with numerical ratings from 1-5.
Show answer and explanation ▾
Correct answer: B
A good few-shot design shows labelled input and output pairs for each class, here Positive, Neutral and Negative.
Why the other options are wrong:
- A. The example uses the wrong output (Text) instead of one of the three classes.
- C. A word list does not show the model input and output pairs.
- D. A ratings table does not show the required labels.
Question 5
Which configuration area defines what the agent should do after a human resolves the escalation?
- A. Inputs description fields
- B. Assignment recipient list
- C. Agent Memory toggle
- D. Outcome behavior section
Show answer and explanation ▾
Correct answer: D
The Outcome behavior section is where you define what the agent does after a human resolves the escalation.
Why the other options are wrong:
- A. Input description fields describe the data passed to the human.
- B. The assignment recipient list says who gets the escalation.
- C. The Agent Memory toggle controls memory, not what happens after resolution.
Question 6
A developer is working on fine-tuning an LLM for generating step-by-step automation guides. After providing a detailed example prompt, they notice inconsistencies in the way the LLM interprets certain technical terms.
What could be the reason for this behavior?
- A. The LLM’s tokenization process may have split complex technical terms into multiple tokens, causing slight variations in how the model interprets and weights their relationships within the context of the prompt.
- B. The LLM’s interpretation is solely based on the frequency of terms within the training dataset, rendering technical nuances irrelevant during generation.
- C. The inconsistency is related to the token limit defined for the prompt’s length, which affects the LLM’s ability to complete a response rather than its understanding of technical terms.
- D. The LLM does not rely on tokenization for understanding prompts; instead, misinterpretation arises from inadequate pre-programmed definitions of technical terms.
Show answer and explanation ▾
Correct answer: A
An LLM reads a prompt as tokens, not whole words. A long or rare technical term is often split into several tokens, and the model weighs those pieces in context, so the same term can be interpreted slightly differently from one place to another.
Why the other options are wrong:
- B. The model uses the context of the prompt as well as learned patterns, so technical nuance is not irrelevant.
- C. A token limit affects how much text fits or gets completed, not how a technical term is interpreted.
- D. LLMs do rely on tokenization and have no pre-programmed dictionary of term definitions.
Question 7
What is the main purpose of using a context grounding strategy with an ECS Index in Agents designer canvas in Studio Web?
- A. To retrieve data based on the user’s current session or inputs
- B. To define static rules for retrieving data from the index
- C. To limit the number of results retrieved from the ECS Index
- D. To keep the ECS Index stored in a shared Orchestrator folder
Show answer and explanation ▾
Correct answer: A
A context grounding strategy lets the agent query the ECS index at runtime with the current inputs, so the retrieved passages are the ones relevant to the request being handled in that session.
Why the other options are wrong:
- B. Grounding retrieves by relevance to the current query, not through static retrieval rules.
- C. Limiting the number of results is a separate setting, not the purpose of the strategy.
- D. Where the index is stored in Orchestrator is a deployment detail, not the purpose of grounding.
Question 8
What is the key difference between a system prompt and a user prompt when configuring an agent?
- A. A system prompt is used for input formatting and passing dynamic arguments, while a user prompt guides the agent’s behavior and planning over time.
- B. A system prompt defines the agent’s role, goals, rules, and when to use tools or escalate, while a user prompt structures how input arguments are passed to the agent at runtime.
- C. A system prompt and a user prompt both serve the same purpose but are written in different parts of the agent.
- D. System prompts exist solely to keep agents constantly adapting in real time, while user prompts are meant for agents that never change their behavior.
Show answer and explanation ▾
Correct answer: B
The system prompt sets the agent's role, goals, rules and guidance on when to use tools or escalate. The user prompt structures the runtime request and carries the input arguments into each run.
Why the other options are wrong:
- A. This reverses the two prompts: the user prompt passes the dynamic inputs and the system prompt guides behavior.
- C. The two prompts have different purposes, not the same purpose in different places.
- D. Neither prompt is defined by whether an agent adapts in real time or never changes.
Question 9
What is the purpose of grouping evaluations into evaluation sets?
- A. Evaluation sets help organize evaluations to address distinct testing needs.
- B. Evaluation sets automatically apply evaluators to all inputs without needing manual assignment.
- C. Evaluation sets are used to calculate and report evaluation scores for individual tests.
- D. Evaluation sets are predefined configurations that ensure evaluations target only root-level outputs.
Show answer and explanation ▾
Correct answer: A
An evaluation set is a logical group of evaluations and evaluators. Grouping them lets a team organize tests around distinct needs, such as a full regression set and a smaller set for a specific scenario.
Why the other options are wrong:
- B. Evaluators are assigned to a set by the designer, not applied automatically to all inputs.
- C. Scores are produced by evaluators for each evaluation, while the set is the grouping.
- D. Evaluation sets are created by the user and are not predefined configurations limited to root-level outputs.
Question 10
Which of the following is an essential aspect of crafting a comprehensive agent story during the validation stage?
- A. Starting immediately with agent behavior prototyping using tools like the Agents designer canvas in Studio Web without assessing mapped automations or impacted systems.
- B. Brainstorming automation use cases without validating personas or critically evaluating existing processes, focusing purely on agent capabilities.
- C. Understanding the daily pain points and inefficiencies of the selected role to identify tasks that consume unnecessary time and potential gains from agent intervention.
- D. Generalizing automation opportunities across all processes and roles without tailoring solutions based on specific personas or organizational contexts.
Show answer and explanation ▾
Correct answer: C
A solid agent story starts from the person in the selected role. Understanding their daily pain points and inefficiencies shows which tasks waste time and what an agent could realistically gain for them.
Why the other options are wrong:
- A. Prototyping before assessing mapped automations and impacted systems skips validation.
- B. Use cases must be validated against personas and existing processes, not built on agent capabilities alone.
- D. Generalizing across all roles ignores the specific persona and organizational context a story needs.
Question 11
An analyst opens Autopilot for Everyone inside Assistant Web in a browser and types, “Run the InvoiceReconciler process and compile the latest vendor invoices”, all without installing any desktop software.
Which distinctive feature enables this workflow?
- A. The ability to run any UiPath automation from any device with absolutely no prerequisites
- B. Natural-language execution of cross-platform automations on a serverless machine, directly in the browser, removing the need for Assistant or Robot installation
- C. Built-in document-understanding models that silently install the Robot service in the background
- D. Browser access that lets users view automation logs but still requires a locally installed Robot to execute workflows
Show answer and explanation ▾
Correct answer: B
In Assistant Web, Autopilot for Everyone accepts a natural-language request and runs cross-platform automations on a serverless machine from the browser, so the user does not need Assistant or a Robot installed locally.
Why the other options are wrong:
- A. Only cross-platform automations run this way, and prerequisites such as access and licensing still apply.
- C. No Robot service is silently installed on the user's machine.
- D. The automation executes without a locally installed Robot, and the browser is not limited to viewing logs.
Question 12
You need to pass a DateTime to an agent tool.
What is the correct way to handle this?
- A. Pass the date directly as a DateTime object, as it is natively supported.
- B. Send the date as a CRON expression for easier scheduling interpretation.
- C. Convert the DateTime to String and parse it inside the agent tool.
- D. Convert the date to an integer representing the number of days since 01/01/0001.
Show answer and explanation ▾
Correct answer: C
Agent arguments support the String, Number, Integer, Boolean, Object and Array types, and DateTime is not one of them. A date and time value is therefore converted to a String before it is passed and parsed back into a DateTime inside the tool.
Why the other options are wrong:
- A. DateTime is not one of the supported argument types, so the object is not passed directly.
- B. A CRON expression describes a recurring schedule, not a single date and time value.
- D. A day count integer loses the time portion and is not the way to pass a date.
Question 13
A solution architect is tasked with building a structured prompt for an agent that extracts key phrases from legal documents. Upon testing, they find that the agent frequently misses extraction patterns.
How can the architect enhance the effectiveness of the few-shot prompt structure?
- A. Add clearly labeled examples demonstrating correct extraction for both simple and complex patterns.
- B. Remove extraction examples from the prompt to test the agent without guidance.
- C. Increase the word count of the legal document examples included in the prompt.
- D. Replace detailed phrases with shorter, random content to improve performance.
Show answer and explanation ▾
Correct answer: A
Few-shot prompts work by showing the model the pattern to follow. Clearly labeled examples of correct extractions, covering both simple and complex cases, teach the agent the patterns it has been missing.
Why the other options are wrong:
- B. Removing the examples turns the prompt into zero-shot and gives the agent less guidance.
- C. Longer sample documents add length, not clearer extraction patterns.
- D. Random content gives the model nothing relevant to imitate.
Question 14
A project manager drafts a vague prompt for an LLM to analyze customer feedback and generate insights. While the response seems generic, some key details appear unrelated to the feedback provided.
What could explain this behavior, and how should the manager proceed?
- A. The LLM failed to generate accurate insights because it requires external validation from another AI model before producing relevant outputs, suggesting the manager should integrate multiple AI systems for better results.
- B. The LLM response depends entirely on its learned patterns and token analysis, and adding more detailed customer feedback won’t significantly alter the result.
- C. The lack of specificity in the prompt caused the LLM to fall back on general knowledge or assumptions, requiring the manager to refine the prompt by including precise context or related references for accuracy.
- D. The LLM breaks the input into tokens but misinterprets individual words, meaning the issue arises from tokenization granularity rather than the structure of the prompt itself.
Show answer and explanation ▾
Correct answer: C
A vague prompt leaves gaps that the model fills with general knowledge and assumptions, which produces generic or unrelated details. Adding precise context, the feedback itself and clear instructions anchors the response to the real data.
Why the other options are wrong:
- A. An LLM does not need validation from a second AI model to produce relevant output.
- B. Adding detailed feedback and context does change the result significantly.
- D. The cause is the lack of specificity in the prompt, not tokenization granularity.
Question 15
Which characteristic primarily differentiates zero-shot chain-of-thought prompting from basic zero-shot prompting?
- A. It supplies a single worked example plus a step-by-step explanation so the model can imitate both format and logic.
- B. It links the output of one prompt to the input of the next, creating a multi-stage workflow for complex tasks.
- C. The prompt explicitly instructs the model to reason step-by-step, causing it to generate intermediate reasoning before the final answer.
- D. It prevents the model from giving any final answer, requiring only the reasoning steps to be produced.
Show answer and explanation ▾
Correct answer: C
Zero-shot chain-of-thought adds an instruction to reason step by step, with no examples. The model then writes out intermediate reasoning before its final answer, which basic zero-shot prompting does not ask for.
Why the other options are wrong:
- A. Supplying a worked example makes it one-shot prompting, not zero-shot.
- B. Feeding one prompt's output into the next prompt is prompt chaining.
- D. The model still gives a final answer after its reasoning.
Question 16
Which statement best describes UiPath Maestro’s capability for deploying AI agents within a BPMN-modeled process?
- A. Maestro deploys agents from UiPath and external providers-such as LangChain, CrewAI, or Agentforce-through one consistent framework that includes human-in-the-loop orchestration.
- B. Maestro is a workflow engine similar to UiPath Studio, but it only allows you to invoke Agentic and Integration tasks.
- C. Maestro embeds external agents as inline code scripts inside the BPMN file and relies on each provider’s runtime instead of Maestro’s orchestration engine.
- D. Maestro deploys only UiPath-built agents in robot-driven processes; any third-party agents must be integrated through external platforms without human checkpoints
Show answer and explanation ▾
Correct answer: A
Maestro orchestrates agents built in UiPath and agents from external providers such as LangChain, CrewAI or Agentforce within one BPMN process, using a consistent framework that also includes human-in-the-loop steps.
Why the other options are wrong:
- B. Maestro is an orchestration layer for robots, agents and people, not a Studio-like workflow engine limited to two task types.
- C. External agents are invoked as tasks under Maestro's orchestration, not embedded as inline scripts in the BPMN file.
- D. Maestro is not limited to UiPath-built agents and supports human checkpoints.
Question 17
When exploring agentic automation discovery, which dimension ensures the solution aligns with the responsibilities and challenges of the individuals involved?
- A. Mapping systems, applications, and tools without understanding how they interact with human roles.
- B. Assessing structured and unstructured knowledge contexts required for the tasks but excluding the personas performing these operations.
- C. Defining the role or persona by considering the people performing the tasks and their needs, challenges, and responsibilities.
- D. Focusing solely on task dependencies while neglecting the daily pain points of individuals executing these tasks.
Show answer and explanation ▾
Correct answer: C
The role or persona dimension looks at the people who perform the tasks, including their needs, challenges and responsibilities, so the solution fits the individuals who will work with it.
Why the other options are wrong:
- A. Mapping systems and tools is the technology landscape dimension and here ignores the human roles.
- B. This describes the knowledge context dimension and explicitly leaves out the personas.
- D. Looking only at task dependencies neglects the pain points of the people doing the work.
Question 18
Which of the following steps is essential when assessing the agentic automation potential of business processes?
- A. Directly propose automation initiatives based solely on stakeholder suggestions without analyzing the “as-is” state.
- B. Prioritize automating every identified inefficiently in the “as-is” state, assuming automation is feasible for all processes.
- C. Focus exclusively on technology landscape evaluation and skip assessing the knowledge context or role dependencies.
- D. Analyze the “as-is” state by mapping workflows, roles, dependencies, and pain points to pinpoint areas for automation.
Show answer and explanation ▾
Correct answer: D
Assessing agentic potential starts with the as-is state. Mapping the current workflows, roles, dependencies and pain points shows exactly where an agent could add value.
Why the other options are wrong:
- A. Stakeholder suggestions alone are not enough without analyzing the as-is state.
- B. Not every inefficiency is feasible or worthwhile to automate.
- C. The knowledge context and the roles must be assessed along with the technology landscape.
Question 19
Which of the following represents a significant challenge in traditional automation systems that rely solely on robots and humans?
- A. Over-reliance on AI-powered natural language processing for all types of communication within the system.
- B. Inability to execute repetitive, rule-based tasks with consistency and precision.
- C. Excessive flexibility in automating complex, exception-heavy workflows without human intervention.
- D. Limited ability to handle dynamic, unstructured tasks that require contextual understanding and adaptive decision-making.
Show answer and explanation ▾
Correct answer: D
Robots follow fixed rules, so automation built only on robots and humans struggles with dynamic, unstructured work that needs contextual understanding and adaptive decisions. Those cases fall back to people.
Why the other options are wrong:
- A. Traditional automation does not rely heavily on AI natural language processing.
- B. Consistent execution of repetitive, rule-based tasks is the main strength of robots.
- C. Traditional automation is rigid in exception-heavy workflows, not excessively flexible.
Question 20
You are testing an agent that generates natural language responses to user queries, such as giving explanations or answering open-ended questions. The outputs need to be assessed for quality, semantic correctness, and reasoning.
Which evaluator type is best suited for this scenario?
- A. JSON similarity, as it assesses the structural similarity of outputs to ensure they align accurately.
- B. LLM-as-a-Judge, as it provides flexible evaluation of semantic correctness and reasoning beyond exact matching.
- C. Deterministic evaluators, as they guarantee all outputs strictly match predefined rules for correctness.
- D. Exact match, as it ensures the agent output matches the expected response word-for-word.
Show answer and explanation ▾
Correct answer: B
Open-ended natural language answers can be correct in many wordings. An LLM-as-a-Judge evaluator uses a model to grade semantic correctness, quality and reasoning, which exact comparisons cannot do.
Why the other options are wrong:
- A. JSON similarity compares structured outputs, not free-text explanations.
- C. Deterministic evaluators check fixed rules and cannot judge reasoning quality.
- D. Exact match fails any correct answer that is worded differently from the expected text.
Question 21
What are the characteristics of an agentic story within the ‘Do later’ quadrant in the impact and feasibility matrix?
- A. Low feasibility and Low Impact
- B. High feasibility and Low Impact
- C. Low feasibility and High Impact
- D. High feasibility and High Impact
Show answer and explanation ▾
Correct answer: C
Stories with high impact but low feasibility are worth doing, but not yet. They are placed in the Do later quadrant until the blockers to implementation are removed.
Why the other options are wrong:
- A. Low feasibility with low impact is not worth scheduling for later.
- B. High feasibility with low impact is easy to build but brings little benefit.
- D. High feasibility with high impact is the quadrant to act on first.
Question 22
What is the significance of the “as-is” process map in identifying agentic automation opportunities?
- A. It directly outlines the roles that agents will assume in the optimized process, ensuring alignment with automation requirements.
- B. It establishes the goals of the new process, serving as a foundation to later create the “to-be” process map.
- C. It defines the current way tasks are performed, helping to highlight inefficiencies, bottlenecks, and areas for improvement that can uncover automation potential.
- D. It serves as a finalized map of processes ready for automation, removing the need for further adjustments or workshops.
Show answer and explanation ▾
Correct answer: C
The as-is process map documents how tasks are performed today. Seeing the current flow makes inefficiencies, bottlenecks and improvement areas visible, and those are where automation potential is found.
Why the other options are wrong:
- A. Agent roles in the optimized process belong to the to-be design.
- B. The as-is map records the current state and does not set the goals of the new process.
- D. It is a starting point that is annotated and discussed in workshops, not a finalized automation map.
Question 23
While configuring an index that points to a sales-ledger.csv file, you need the agent to answer SQL-style questions such as “total revenue per region”.
Which query type setting in the Contexts panel will make Context Grounding treat the CSV as tabular data and enable cell-level look-ups?
- A. Semantic
- B. Decrease relevance score threshold
- C. Let the agent infer
- D. Structured
Show answer and explanation ▾
Correct answer: D
The Structured query type tells Context Grounding to treat the CSV as tabular data. That allows look-ups and aggregations over rows and columns, such as total revenue per region.
Why the other options are wrong:
- A. Semantic search retrieves text passages by meaning and does not compute over table cells.
- B. The relevance score threshold filters semantic results and is not a query type.
- C. Leaving the choice to the agent does not guarantee the file is handled as a table.
Question 24
Four draft system prompts are shown for an “invoice-approval” agent. Based on UiPath guidance for context, instructions, output format, and constraints, which draft is the most robust choice?
- A. You are an invoice approver. After processing, output exactly the following JSON template: { “id”: “ABC-123”, “status”: “approved”, “amount”: 9999.99 } 1 Extract {{Invoice_ID}} from the email text. 2 When an {{Invoice_ID}} is found, run the LookupInvoice tool to retrieve invoice amount and supplier name. 3 If the total ≥ $10 000, escalate the case to Finance in Action Center, sending {{Invoice_ID}}, amount, and supplier. 4 If the total < $10 000, approve the invoice. Populate the fields above with real data.
- B. You are an invoice-approval agent who deals only with supplier invoices and rejects any other request. 1 Extract Invoice_ID from the email text. 2 When an Invoice_ID is found, run the LookupInvoice tool to retrieve invoice amount and supplier name. 3. If the total ≥ $10 000, escalate the case to Finance in Action Center, sending Invoice_ID, amount, and supplier. 4 If the total tags: use or as appropriate. Follow a concise, professional tone and refuse tasks outside invoice approval.
- C. You are an invoice-approval agent who deals only with supplier invoices and rejects any other request. 1 Extract {{Invoice_ID}} from the email text. 2 When an {{Invoice_ID}} is found, run the LookupInvoice tool to retrieve invoice amount and supplier name. 3 If the total ≥ $10 000, escalate the case to Finance in Action Center, sending {{Invoice_ID}}, amount, and supplier. 4 If the total tags: use or as appropriate. Follow a concise, professional tone and refuse tasks outside invoice approval.
- D. You are an invoice approver. After processing, output exactly the following JSON template: { “id”: “ABC-123”, “status”: “approved”, “amount”: 9999.99 } 1 Extract Invoice_ID from the email text. 2 When an Invoice_ID is found, run the LookupInvoice tool to retrieve invoice amount and supplier name. 3 If the total ≥ $10 000, escalate the case to Finance in Action Center, sending Invoice_ID, amount, and supplier. 4 If the total < $10 000, approve the invoice. Populate the fields above with real data.
Show answer and explanation ▾
Correct answer: C
This draft states the agent's role and scope, gives numbered instructions, references the input argument in double curly braces so the runtime value is injected, and sets constraints on tone and out-of-scope requests.
Why the other options are wrong:
- A. It has no scope or refusal constraint, and its sample JSON with fixed values invites the agent to copy them.
- B. Invoice_ID is written as plain text, so the input argument is not referenced with double curly braces.
- D. It combines the fixed sample JSON with a plain-text Invoice_ID and sets no scope constraint.
Question 25
What configuration options are available for escalations to ensure agents use them effectively?
- A. You can assign a single escalation to one agent.
- B. Agents can use multiple escalations automatically without any need for pre-configuration or context-specific descriptions.
- C. Escalations can use an Action app with a unique name and description to clarify specific circumstances.
- D. Escalation components cannot use unique Action app names or descriptions.
Show answer and explanation ▾
Correct answer: C
Each escalation is tied to an Action app and given its own name and description. The description tells the agent the specific circumstances in which that escalation applies.
Why the other options are wrong:
- A. An agent is not limited to a single escalation.
- B. Escalations must be configured and described before the agent can use them properly.
- D. Escalations do use unique names and descriptions.
Question 26
What is the role of indexes in UiPath Agents?
- A. To schedule agent tasks
- B. To manage automation package versions
- C. To provide agents access to permissioned knowledge bases
- D. To monitor agent performance
Show answer and explanation ▾
Correct answer: C
Indexes are the Context Grounding stores that an agent queries. They give the agent access to permissioned knowledge bases, so its answers can draw on the organization's own documents.
Why the other options are wrong:
- A. Scheduling is handled by triggers, not indexes.
- B. Package versions are managed in Orchestrator, not through indexes.
- D. Agent performance is monitored through traces and evaluations, not indexes.
Question 27
When passing runtime data into an Agent, which approach ensures the input argument is actually available inside the user prompt at execution time?
- A. Declare the argument in the system prompt; any text surrounded by angle brackets (e.g., ) will be substituted automatically.
- B. Use single braces like {CUSTOMER_EMAIL}, because the platform automatically normalizes the identifier.
- C. Simply mention the variable name in plain prose-the Agent will infer the value from the workflow without special syntax.
- D. Create the argument in Data Manager and reference it verbatim inside double curly braces e.g. {{CUSTOMER_EMAIL}}, so the name matches exactly.
Show answer and explanation ▾
Correct answer: D
An input argument is defined in Data Manager and then referenced in the prompt inside double curly braces, with exactly the same name. At runtime the placeholder is replaced with the argument's value.
Why the other options are wrong:
- A. Angle brackets are not the placeholder syntax for arguments.
- B. Single braces are not substituted and names are not normalized.
- C. Plain prose does not bind the argument, so the value is not injected.
Question 28
What should be prepared for workshops aimed at identifying agentic automation opportunities?
- A. A list of potential agents and technical prototypes ready for deployment.
- B. Comprehensive automation implementation plans for all roles
- C. A completed “to-be” process map tailored to automation goals.
- D. Adjusted or annotated “as-is” process maps to highlight inefficiencies and automation potential.
Show answer and explanation ▾
Correct answer: D
Workshops are grounded in the current process. Bringing as-is process maps that have been adjusted or annotated to mark inefficiencies and automation potential gives participants a shared basis for finding opportunities.
Why the other options are wrong:
- A. Agent lists and prototypes come after opportunities are identified.
- B. Implementation plans for every role are an outcome of later stages, not a workshop input.
- C. The to-be map is produced after the workshop, not before it.
Question 29
Which situation is an appropriate reason for an agent to raise an escalation during execution?
- A. The agent has just started its first step
- B. The Agent Memory time-to-live reaches three months
- C. A tool call returns data with an unexpected structure
- D. The agent successfully finishes every planned step
Show answer and explanation ▾
Correct answer: C
An escalation hands a decision to a person when the agent cannot proceed safely. A tool returning data in an unexpected structure is such a case, because the agent cannot trust or interpret the result.
Why the other options are wrong:
- A. Starting the first step is normal execution and needs no human.
- B. A memory time-to-live setting is not an execution problem that calls for a person.
- D. Finishing every step successfully means there is nothing to escalate.
Question 30
Which of the following best illustrates how to optimize prompt engineering when dealing with domain-specific tasks?
- A. Provide sufficient context within the prompt to convey nuances and background information relevant to the domain.
- B. Focus exclusively on iterative refining of the prompt without considering the complexity of the specific domain.
- C. Use multiple examples, even if they are outside the domain, to help the model generalize better across tasks.
- D. Rely solely on simple, clear prompts without providing additional details, regardless of task complexity.
Show answer and explanation ▾
Correct answer: A
For domain-specific tasks the model needs the background that an expert would have. Supplying enough context in the prompt conveys the nuances and terminology of the domain and leads to accurate output.
Why the other options are wrong:
- B. Iterating on a prompt without regard to the domain's complexity does not supply the missing background.
- C. Examples from outside the domain can steer the model away from the task.
- D. A simple prompt without detail is not enough for a complex domain task.
Question 31
What configuration option is available in the Properties panel for a Service Task to start and wait for an RPA workflow?
- A. Implementation > Type: Start and wait for RPA workflow
- B. Implementation > Type: Execute connector activity
- C. Implementation > Type: Create and wait for queue item
- D. Implementation > Type: Execute Business Rule
Show answer and explanation ▾
Correct answer: A
In the Properties panel of a Service Task, the Implementation section sets what the task runs. Choosing Start and wait for RPA workflow starts the automation and pauses the process until it completes.
Why the other options are wrong:
- B. Execute connector activity runs an Integration Service connector action, not an RPA workflow.
- C. Create and wait for queue item works with an Orchestrator queue item, not a workflow run.
- D. Execute Business Rule evaluates a rule and does not start an RPA workflow.
Question 32
While designing a competitive-intelligence agent you need it to look up real-time information on “today’s top mergers in the telecom sector” without providing a specific URL.
Which tool should you add in the Tools panel so the agent can carry out that web search autonomously?
- A. Web Reader
- B. Google integration
- C. Context Grounding index
- D. Web Search
Show answer and explanation ▾
Correct answer: D
The Web Search tool lets the agent query the web for current information from a search phrase alone, which fits a real-time lookup where no address is supplied.
Why the other options are wrong:
- A. Web Reader reads the content of a specific URL, which is not provided here.
- B. A Google integration connects to Google applications and is not the tool for open web search.
- C. A Context Grounding index searches the organization's own indexed documents, not live web results.
Question 33
What key advantage do guardrails provide when developers configure them for individual tool calls?
- A. Their primary purpose is to speed up the agent’s reasoning cycle by shortening the time spent inside a tool call.
- B. They allow developers to define deterministic conditions that trigger human intervention only when specific, unexpected inputs or outputs occur, thereby reducing-but not eliminating-HITL tasks.
- C. Once enabled, they universally resolve all forms of AI unpredictability across the platform, so no additional customization is needed at the tool level.
- D. They autonomously correct every unexpected input or output without ever requiring human involvement after deployment.
Show answer and explanation ▾
Correct answer: B
Guardrails let developers set deterministic conditions on a tool call. Human intervention is triggered only when specific unexpected inputs or outputs occur, which reduces human-in-the-loop tasks without removing them.
Why the other options are wrong:
- A. Guardrails control behavior and are not a way to speed up reasoning.
- C. Guardrails are configured for each tool and do not remove all unpredictability across the platform.
- D. Guardrails can route cases to a person and do not correct everything autonomously.
Question 34
What is the primary role of guardrails in tools?
- A. Guardrails are designed to apply only after tool execution, without influencing pre-execution conditions.
- B. Guardrails are used exclusively to automate all tool corrections without the possibility of triggering human intervention.
- C. Guardrails control unexpected behaviors within tool calls deterministically, allowing developers to configure conditions for human intervention and escalations.
- D. Guardrails only validate tool inputs during development and do not address unpredictable behaviors at runtime.
Show answer and explanation ▾
Correct answer: C
Guardrails give deterministic control over unexpected behavior in tool calls. Developers configure conditions on tool inputs or outputs that trigger actions such as human intervention through an escalation.
Why the other options are wrong:
- A. Guardrails can be evaluated before a tool runs as well as after.
- B. Guardrails can trigger human intervention and are not limited to automatic correction.
- D. Guardrails act at runtime, not only during development.
Question 35
Which of the following best describes the key difference between agents and traditional; RPA robots in UiPath’s automation ecosystem?
- A. Agents are designed for repetitive tasks and structured workflows, while robots excel in unstructured, exception-heavy environments.
- B. Agents are a probabilistic approach to make decisions based on patterns and real-time data, while robots follow structured logic and fixed rules.
- C. Agents can only communicate through code, while robots use natural language processing for all interactions.
- D. Agents require human supervision to operate, while robots work independently in all scenarios.
Show answer and explanation ▾
Correct answer: B
Agents are probabilistic: they use an LLM to decide based on patterns and real-time data. Robots are deterministic and follow structured logic and fixed rules.
Why the other options are wrong:
- A. This reverses the two: robots suit repetitive, structured work and agents suit unstructured, exception-heavy work.
- C. Agents work with natural language, and robots do not use it for all interactions.
- D. Agents can act autonomously, and robots do not work independently in every scenario.
Question 36
When you want to connector field value to be inferred dynamically at run time, which input method should you select in the activity tool?
- A. Prompt
- B. Static value
- C. Argument
- D. Clear value
Show answer and explanation ▾
Correct answer: A
With the Prompt input method the agent infers the field's value at runtime from its instructions and the tool and field descriptions, so nothing is fixed at design time.
Why the other options are wrong:
- B. A static value is fixed at design time and is the same on every run.
- C. An argument passes in a value defined for the agent and is not inferred.
- D. Clear value empties the field and does not set how it is filled.
Question 37
You are designing an evaluation for an agent. The test should compare the agent’s free-text summary of a ticket with an expected human summary, rewarding answers that are semantically equivalent even if wording differs. Which evaluator type best fits this criterion?
- A. LLM-as-a-Judge: Semantic Similarity
- B. Rely on the overall Evaluation score alone; it automatically detects semantic differences without configuring a specific evaluator.
- C. Exact match
- D. JSON similarity
Show answer and explanation ▾
Correct answer: A
The LLM-as-a-Judge semantic similarity evaluator uses a model to compare the meaning of the agent's output with the expected output, so a summary that says the same thing in different words still scores well.
Why the other options are wrong:
- B. The overall score only aggregates the evaluators that have been configured and detects nothing without one.
- C. Exact match fails any summary that is worded differently.
- D. JSON similarity compares structured output, not free-text meaning.
Question 38
An agent is being developed to calculate travel itineraries for event attendees. Inputs include location and event details, but the outputs fail to include critical time constraints (e.g. arrival before the event begins).
What should be added to the prompt to address this issue?
- A. Context that solely highlights the event details without timing constraints.
- B. Instructions allowing flexibility in timing to simplify the agent’s process.
- C. Constrains specifying that suggested itineraries must align with event schedules and start times.
- D. Examples of itineraries that include random arrival times.
Show answer and explanation ▾
Correct answer: C
Constraints tell the agent what a valid answer must respect. Stating that every itinerary has to align with the event schedule and start time makes arrival before the event a requirement.
Why the other options are wrong:
- A. Context without timing information leaves the same gap.
- B. Allowing flexible timing makes late arrivals more likely.
- D. Examples with random arrival times teach the wrong pattern.
Question 39
What is a correct Agent Story Blueprint?
- A. As an [agent role] operating in [context/environment], I want to [objective] by [actions/behaviors], interacting with [users/systems/agents], so that [desired outcome/benefit], Success is measured by [success criteria]
- B. As an [agent role], operating with [tools and applications], I want to [objective] by [success metrics], interacting with [users/systems/agents], so that [desired outcome/benefit], Success is measured based on [success criteria].
- C. As a [desired outcome/benefit], operating in [success criteria], I want to [context/environment] by [agent role], interacting with [objective], so that [users/systems/agents], Success is measured by [actions/behaviors].
- D. As an [users/systems/agents], operating in [actions/behaviors], I want to [desired outcome/benefit] by [objective], interacting with [success criteria], so that [agent role], Success is measured by [context/environment].
Show answer and explanation ▾
Correct answer: A
The blueprint reads: as an agent role operating in a context or environment, I want to reach an objective by certain actions or behaviors, interacting with users, systems or agents, so that a desired outcome is achieved, with success measured by success criteria.
Why the other options are wrong:
- B. It replaces the context with tools and puts success metrics where the actions belong.
- C. It scrambles the elements, starting with the desired outcome where the agent role belongs.
- D. It places the users, systems or agents where the agent role belongs and misorders the rest.
Question 40
You are evaluating an agent that extracts the invoice date from a PDF. You define a deterministic evaluation using the following ground truth JSON:
{“invoice_date”: “2024-09-01”}
What logic will the evaluation use?
- A. Ask a human to review and rate the date manually.
- B. Count how many characters match between output and input.
- C. Compare the extracted date string exactly to the expected value in the ground truth.
- D. Use a language model to determine if the date looks plausible.
Show answer and explanation ▾
Correct answer: C
A deterministic evaluation applies a fixed rule. Here the extracted invoice_date is compared exactly with the ground truth value 2024-09-01, and it passes only when the two match.
Why the other options are wrong:
- A. Human review is not part of a deterministic evaluation.
- B. Counting matching characters against the input is not how the comparison works.
- D. Using a language model to judge the output is a model-graded evaluation.
Question 41
While configuring an Integration Service activity as a tool for your agent in Studio Web, how should you set up the activity so the agent can decide the value of a required field (e.g. Channel ld) at runtime based solely on instructions in the prompt?
- A. Change every field, including Channel Id, to Variable because an agent cannot infer any field values without explicit arguments.
- B. Leave the field’s input method on Prompt (the default) and keep or refine the tool description; this lets the agent infer the value during execution.
- C. Change every field, including Channel ld, to Argument because an agent cannot infer any field values without explicit arguments.
- D. Declare the field as an output argument in Data Manager so the agent can feed a value back into the tool.
Show answer and explanation ▾
Correct answer: B
Prompt is the default input method for an activity tool field. Leaving Channel Id on Prompt and keeping a clear tool description lets the agent work out the value at runtime from the prompt instructions.
Why the other options are wrong:
- A. Agents can infer field values, so switching every field to Variable is not needed.
- C. Agents can infer field values without explicit arguments.
- D. An output argument returns data from the agent and does not let it choose a tool input.
Question 42
For a workflow designed to automate data entry from invoices where the format is consistent and the rules are predefined, ensuring compliance and efficiency which automation technology is better suited for this task?
- A. Agents, because they leverage probabilistic decision-making and perform best when the environment has fixed, predictable rules like invoice processing.
- B. Robots (RPA), because they excel in deterministic, rule-based workflows, ensuring precision and reliability when managing repetitive and structured tasks like data entry.
- C. Robots (RPA), because they integrate natural language capabilities to interpret unstructured data and adapt to changing formats in realtime.
- D. Agents, because they are adaptive systems designed to handle dynamic workflows and processes involving judgement or contextual awareness, even for structured tasks.
Show answer and explanation ▾
Correct answer: B
Consistent formats and predefined rules describe a deterministic task. RPA robots follow fixed logic precisely and repeatably, which gives the compliance and efficiency that structured data entry needs.
Why the other options are wrong:
- A. Fixed, predictable rules are where robots perform best, not probabilistic agents.
- C. Interpreting unstructured data and adapting to changing formats is an agent strength, and it is not needed here.
- D. Agents suit work that needs judgment, which this structured task does not.
Question 43
When evaluating agentic use cases using the impact and feasibility matrix, what does “impact” specifically refer to?
- A. The combined effect of process efficiency and organizational culture improvements.
- B. The degree to which the use case improves the process or benefits the business.
- C. The time impact resulting from use case implementation.
- D. The likelihood that the use case will align with stakeholder diversity and integration goals.
Show answer and explanation ▾
Correct answer: B
On the impact and feasibility matrix, impact measures how much the use case improves the process or benefits the business. Feasibility measures how practical it is to build.
Why the other options are wrong:
- A. Organizational culture is not part of the impact measure.
- C. Time is only one possible benefit, not the definition of impact.
- D. Stakeholder diversity and integration goals do not define impact.
Question 44
A company wants to automate customer support inquiries where messages vary widely in tone, urgency, and content.
Which type of automation technology should they use, and why?
- A. Robots (RPA), because they excel at handling workflows with ambiguous environments that require decision-making based on contextual awareness.
- B. Agents, because they operate best in deterministic environments with fixed rules and structured logic.
- C. Robots (RPA), because they use a probabilistic approach tailored to dynamic and exception-heavy workflows.
- D. Agents, because they are probabilistic and adaptive, making them ideal for unstructured tasks requiring flexibility and contextual awareness.
Show answer and explanation ▾
Correct answer: D
Messages that vary in tone, urgency and content are unstructured input. Agents are probabilistic and adaptive, so they can interpret each message in context and respond flexibly.
Why the other options are wrong:
- A. Robots follow fixed rules and do not handle ambiguous, context-dependent work well.
- B. Deterministic environments with fixed rules describe where robots work best, not agents.
- C. Robots are deterministic, not probabilistic.
Question 45
What is the primary recommendation for using bpmn.uipath.com to access the Maestro modeling canvas?
- A. Use bpmn.uipath.com as a sandbox since the tool does not offer implementation capabilities
- B. Always use bpmn.uipath.com as the primary tool for modeling processes with full implementation features.
- C. Use bpmn.uipath. com only to import predefined process templates from Studio Web.
- D. Utilize bpmn.uipath.com exclusively for creating agentic processes with implementation capabilities.
Show answer and explanation ▾
Correct answer: A
The canvas at bpmn.uipath.com is for modeling only and has no implementation capabilities, so it is recommended as a sandbox for drawing and trying out BPMN diagrams.
Why the other options are wrong:
- B. It does not provide full implementation features.
- C. Its use is not limited to importing templates from Studio Web.
- D. It has no implementation capabilities for agentic processes.
Question 46
What is the purpose of publishing an agentic process in Agents designer canvas in Studio Web?
- A. To prepare the process for debugging purposes
- B. To deploy so it can run on an automation agent or robot
- C. To enable collaboration by sharing the process with other users
- D. To schedule the process for automated execution within the platform
Show answer and explanation ▾
Correct answer: B
Publishing packages the process and makes it available for deployment, so it can be run outside the design environment instead of only being tested there.
Why the other options are wrong:
- A. Debugging is done in the designer before publishing.
- C. Collaboration is handled by sharing the project, not by publishing.
- D. Scheduling is set up with triggers after the process is deployed.
Question 47
What advantage does intelligent decision-making in agentic automation provide over traditional automation methods?
- A. It eliminates the need for any human oversight in all business processes.
- B. It ensures that all decisions are made with 100% accuracy, eliminating all potential errors.
- C. It allows for handling of unstructured, exception-heavy workflows where conditions and outcomes vary.
- D. It speeds up decision-making by always choosing the first available option without analysis.
Show answer and explanation ▾
Correct answer: C
Agents reason over context instead of following a fixed script. That lets agentic automation handle unstructured, exception-heavy workflows where conditions and outcomes change from case to case.
Why the other options are wrong:
- A. Human oversight remains part of agentic automation.
- B. Agents are probabilistic and do not guarantee perfect accuracy.
- D. Agents analyze the situation and do not simply take the first available option.
Question 48
How does Maestro enable organizations to manage processes at scale?
- A. By leveraging an event-sourced engine to execute processes at scale, with the ability to fix, restart, modify, and manage versions.
- B. By solely relying on AI agents to execute large-scale processes without additional tools or frameworks.
- C. By relying on manual execution of processes using BPMN notation, as Maestro does not support scalability or automated lifecycle management.
- D. Once processes are executed at scale, Maestro fully handles debugging, updates, and scalability without user input.
Show answer and explanation ▾
Correct answer: A
Maestro runs processes on an event-sourced engine built for scale. Operators can fix, restart and modify running instances and manage process versions.
Why the other options are wrong:
- B. Maestro coordinates agents, robots and people and does not rely on agents alone.
- C. Maestro executes BPMN models automatically and supports scale and lifecycle management.
- D. Users still take part in fixing, updating and managing processes.
Question 49
Which advantage of providing Retrieval-Augmented Generation (RAG) as a service to UiPath generative-AI experiences most directly mitigates the limitations of an LLM’s context window?
- A. RAG speeds up document ingestion by caching previous user queries instead of consulting any knowledge base.
- B. RAG permanently removes all context-window limits for any present of future language model, regardless of its architecture or size.
- C. RAG uploads the entire knowledge base into the LLM’s prompt, effectively bypassing the need for selective retrieval.
- D. RAG pulls only the most relevant snippets from indexed knowledge bases, thereby improving accuracy, reliability, and efficiency even when the LLM’s context window is small.
Show answer and explanation ▾
Correct answer: D
RAG retrieves only the most relevant pieces from indexed knowledge bases and adds them to the prompt. The model gets the facts it needs within a limited context window, which improves accuracy, reliability and efficiency.
Why the other options are wrong:
- A. RAG consults the knowledge base and is not a cache of earlier user queries.
- B. Context window limits still exist, and RAG works within them.
- C. Loading the whole knowledge base into the prompt is the opposite of selective retrieval.
Question 50
A team is building an AI agent that drafts personalized marketing emails. The quality of the drafts depends on tone, alignment with brand voice, and personalization.
What evaluation approach is best?
- A. Model-graded evaluation to capture nuanced style and relevance.
- B. Deterministic evaluation using a checklist of key phrases.
- C. Random sampling with A/B testing.
- D. Evaluation using a character count threshold to assess message quality.
Show answer and explanation ▾
Correct answer: A
Tone, brand voice and personalization are subjective qualities with no single correct wording. A model-graded evaluation uses an LLM to judge these nuances of style and relevance.
Why the other options are wrong:
- B. A phrase checklist cannot judge tone or brand voice.
- C. Random sampling with A/B testing is not an evaluator for the quality of each draft.
- D. Character count says nothing about tone or personalization.
Question 51
Which of the following is a key configuration step when using ‘Start and wait for agent’ in UiPath?
- A. Setting up a parallel gateway to handle multiple agents simultaneously.
- B. Using a subprocess to encapsulate the agent logic.
- C. Configuring the agent specifying input/output variables.
- D. Configuring a timer event to trigger the agent.
Show answer and explanation ▾
Correct answer: C
When a Service Task uses Start and wait for agent, the agent is selected and its input and output variables are mapped, so the process passes data to the agent and receives its result.
Why the other options are wrong:
- A. A parallel gateway is a process design choice, not a required setting of this task.
- B. A subprocess is not needed to invoke an agent.
- D. A timer event is not part of configuring the task.
Question 52
What happens when you set the input method of an activity to Agent (the default choice)?
- A. The agent prompts a human in Action Center to supply the value
- B. The field always applies one constant that you enter during design
- C. The agent derives the value at runtime from the field’s Description and its own reasoning
- D. The field uses a workflow argument that was defined earlier in the job
Show answer and explanation ▾
Correct answer: C
With the default input method the agent decides the field's value at runtime. It uses the field's description together with its own reasoning over the prompt and available data.
Why the other options are wrong:
- A. Asking a person in Action Center is an escalation, not this input method.
- B. A constant entered at design time is the static value method.
- D. Using a previously defined value is the argument method.
Question 53
While adding the Slack Send Message connector action as a tool in Studio Web, you want the agent-not the workflow designer- to decide the value of the Message field each time it runs.
Which configuration best enables this behaviour?
- A. Keep the field’s input method on Prompt and, if needed, fine-tune the tool’s description so the agent can infer the value at runtime.
- B. Change the input method to Variable and rely on an undeclared variable, assuming the agent will still infer the text.
- C. Change the input method to Argument and rely on an undeclared variable, assuming the agent will still inter the text.
- D. Convert every field in the activity to Argument because agents cannot infer any field values without explicit arguments.
Show answer and explanation ▾
Correct answer: A
Keeping the Message field on the Prompt input method leaves the value to the agent, which infers it at runtime. A well-written tool description helps the agent compose the right message.
Why the other options are wrong:
- B. An undeclared variable supplies no value and takes the decision away from the agent.
- C. An argument with nothing declared behind it gives the agent nothing to pass.
- D. Agents can infer field values, so converting every field to Argument is unnecessary.
Question 54
You are working on a chatbot designed to assist users with troubleshooting software issues. You notice that the chatbot is providing inconsistent answers to the same question due to unclear prompts.
What approach should you apply to improve the chatbot’s performance?
- A. Use step-by-step thinking exclusively, even for simple troubleshooting tasks, to improve overall accuracy.
- B. Add vague but creative instructions to make the AI responses flexible and adaptable to user queries.
- C. Incorporate a broad range of diverse examples to solve the inconsistencies without clarifying the initial prompt structure.
- D. Use clear and specific language, including key details that guide the AI toward the desired outcome for accurate troubleshooting responses.
Show answer and explanation ▾
Correct answer: D
Inconsistent answers to the same question point to an unclear prompt. Clear, specific language with the details that define the desired outcome guides the model to the same accurate troubleshooting response each time.
Why the other options are wrong:
- A. Step-by-step thinking for every simple task does not fix an unclear prompt.
- B. Vague instructions increase variation in the answers.
- C. Diverse examples without a clear prompt structure leave the cause of the inconsistency in place.
Question 55
A team is designing an agent to convert plain text meeting notes into a formatted agenda (e.g., structured bullet points). Despite providing a few example transformations in the prompt, the agent generates agendas in inconsistent formats.
What critical step was likely overlooked?
- A. Adding clear instructions detailing the output format.
- B. Providing only examples without additional context about the task.
- C. Adding randomized formatting examples to test the agent’s creativity.
- D. Including constraints to limit the length of the agenda for simplicity.
Show answer and explanation ▾
Correct answer: A
Examples alone leave the model to guess the pattern. Explicit instructions that describe the output format, such as the required headings and bullet structure, make the agenda consistent.
Why the other options are wrong:
- B. Relying on examples alone is what the team already did.
- C. Randomized formatting examples make the output less consistent.
- D. A length limit does not define the format of the agenda.
Question 56
What methods can be used to manage the ingestion pipeline for Context Grounding in UiPath?
- A. Ingestion pipelines can only be managed via Studio Web, excluding other UiPath tools.
- B. Use manual upload of data into shared folders without indexing options in Orchestrator.
- C. Manage ingestion pipelines directly in Orchestrator or use the Update Context Grounding Index activity.
- D. Ingestion pipelines can only be managed through third-party tools integrated outside of UiPath.
Show answer and explanation ▾
Correct answer: C
Ingestion for Context Grounding can be managed from the index in Orchestrator, where data is synced, or from an automation with the Update Context Grounding Index activity.
Why the other options are wrong:
- A. Ingestion is not restricted to Studio Web.
- B. Uploading files without indexing does not make them searchable by Context Grounding.
- D. Ingestion is managed inside UiPath, not only through third-party tools.
Question 57
When configuring escalations for an agent, what is a key step to ensure the agent knows when to use the escalation during execution?
- A. Utilize required fields in the inputs section of the escalation to define conditions for triggering escalations dynamically.
- B. Configure escalation behavior entirely within the outcome behavior section, specifying how each resolution should be handled.
- C. Directly assign an escalation recipient to ensure proper routing, which eliminates the need for agent-specific prompts in the escalation logic.
- D. Add a prompt in the properties panel to help the agent determine the appropriate circumstances for using the escalation.
Show answer and explanation ▾
Correct answer: D
The agent decides when to escalate from the instructions it is given. Adding a prompt for the escalation in the properties panel describes the circumstances in which the agent should use it.
Why the other options are wrong:
- A. Required input fields define what data is sent, not when to escalate.
- B. Outcome behavior covers what happens after the person responds, not when the escalation is raised.
- C. Assigning a recipient routes the task and does not tell the agent when to escalate.
Question 58
How does the “human in the loop” characteristic enhance agent performance?
- A. It allows humans to intervene only when automation performs poorly, failing to incorporate consistent oversight mechanisms.
- B. It guarantees error correction after automation fails without providing a proactive validation step.
- C. It defines the specific event or trigger for the agent’s tasks, even though “human in the loop” focuses only on oversight.
- D. It ensures humans can validate responses for accuracy, tone, or compliance before the agent completes its workflow, addressing businesscritical concerns.
Show answer and explanation ▾
Correct answer: D
Human in the loop places a person at the points that matter. They can validate a response for accuracy, tone or compliance before the agent completes its work, which protects business-critical outcomes.
Why the other options are wrong:
- A. Human in the loop provides planned oversight, not only intervention after poor performance.
- B. It adds proactive validation and is not limited to correction after a failure.
- C. The event that starts the agent's work is the trigger characteristic, not human in the loop.
Question 59
After Agent Memory is enabled, what happens by default to each resolved escalation at runtime?
- A. It is stored as a question-response pair tied to the latest agent version
- B. Is it stored as a question-response pair and used as input for validation Action App
- C. It is deleted immediately
- D. It is achieved indefinitely in a Context Grounding index
Show answer and explanation ▾
Correct answer: A
With Agent Memory enabled, each resolved escalation is saved at runtime so the agent can reuse the human's resolution in similar cases. By default the stored pair is associated with the latest version of the agent.
Why the other options are wrong:
- B. The stored resolution is recalled by the agent and is not fed into a validation Action app.
- C. The resolution is kept for reuse, not deleted.
- D. It is kept in Agent Memory, not archived indefinitely in a Context Grounding index.
Question 60
What key feature of agentic automation allows it to handle dynamic, non-deterministic environments effectively?
- A. Its reliance on fixed rules and structured logic for all decision-making processes.
- B. Its ability to use a probabilistic approach for decision-making based on patterns and real-time data.
- C. Its capability to execute only pre-programmed, repetitive tasks without variation.
- D. Its exclusive use of human intervention for all complex decision-making scenarios.
Show answer and explanation ▾
Correct answer: B
Agentic automation makes decisions probabilistically, based on patterns and real-time data. That allows it to adapt when inputs and conditions are not known in advance.
Why the other options are wrong:
- A. Fixed rules and structured logic describe deterministic robots.
- C. Running only pre-programmed, repetitive tasks describes traditional RPA.
- D. Agents decide autonomously and bring in humans only where needed.
Question 61
After adding a policy-manual PDF as a context, the agent returns long, vaguely related text parts. To improve the results without cutting the number of chunks outright, which properly is designed for Semantic queries to filter out low-similarity results?
- A. Raise the model’s token limit; token budget affects LLM context size, not the semantic filtering of retrieved chunks.
- B. Switch the Query type to Structured; that setting is for CSV/tabular data and will not help with text-heavy PDFs.
- C. Lower Max results generated; this just limits chunk count and could drop useful rather than improve precision.
- D. Increase the Relevance score threshold so only chunks scoring above the higher similarity cutoff are grounded into the prompt.
Show answer and explanation ▾
Correct answer: D
For Semantic queries the relevance score threshold sets the minimum similarity a chunk must reach. Raising it filters out loosely related passages, so only closely matching chunks are grounded into the prompt.
Why the other options are wrong:
- A. The token limit controls how much text the model handles, not which chunks are retrieved.
- B. The Structured query type is meant for tabular data, not a text-heavy PDF.
- C. Lowering the maximum results only cuts the count of chunks, which the scenario rules out.
Question 62
What is the primary role of Escalations in the context of agent-controlled execution?
- A. They serve solely as a mechanism for capturing agent errors after execution for compliance purposes.
- B. They replace the need for Agent Memory by directly correcting outputs through manual overrides.
- C. They are used to completely bypass agents when humans are available for decision-making.
- D. They enable human intervention to validate agent outputs, inform Agent Memory, and calibrate future execution plans.
Show answer and explanation ▾
Correct answer: D
Escalations bring a person into the agent's run to validate its output or make a decision. The resolution can also inform Agent Memory, which calibrates how the agent plans similar cases in the future.
Why the other options are wrong:
- A. Escalations act during execution and are not an after-the-fact error log.
- B. Escalations feed Agent Memory and do not replace it.
- C. Escalations support the agent at specific points and do not bypass it.
Question 63
In UiPath Integration Service, when is a “Connection” officially created between a user and an external application?
- A. It is set up as soon as a developer drags the corresponding connector into a UiPath Studio project for the first time.
- B. It is established the moment the user successfully authenticates with the external application’s API provider.
- C. It begins only after a robot executes its initial activity that calls the external application’s API.
- D. It is automatically created for all members of an organization as soon as Integration Service is enabled, regardless of individual authentication.
Show answer and explanation ▾
Correct answer: B
In Integration Service a connection links one user to one external application. It is established when that user successfully authenticates with the application's API provider.
Why the other options are wrong:
- A. Adding a connector to a project does not authenticate anyone.
- C. The connection must exist before a robot can run an activity that calls the application.
- D. Connections depend on each user's authentication and are not created for the whole organization automatically.
Question 64
What are the prerequisites for using Context Grounding with UIPath GenAI Activities?
- A. Use any version of Studio Desktop without the GenAI activities connector.
- B. Ensure a connection to the UiPath GenAI activities connector and use Studio Web or Studio Desktop version 2024.4 or newer.
- C. Ensure manual data ingestion without relying on UiPath connectors or Studio versions.
- D. Create an Orchestrator index but avoid using GenAI activities in workflows.
Show answer and explanation ▾
Correct answer: B
Context Grounding is called from workflows through the UiPath GenAI Activities, so you need a working connection to the GenAI Activities connector in Integration Service and a supported design tool: Studio Web or Studio Desktop 2024.4 or newer.
Why the other options are wrong:
- A. The GenAI Activities connector is required, and not every Studio Desktop version supports these activities.
- C. Data is ingested into an index through UiPath services, so the connector and a supported Studio version are still needed.
- D. An index alone does nothing for a workflow, since the GenAI activities are what query it.
Question 65
What is the purpose of UiPath coded agents?
- A. To replace UiPath-built agents in all scenarios.
- B. To enable highly tailored automations with complex logic using custom code.
- C. To operate without integration with UiPath Orchestrator.
- D. To handle only repetitive and rule-based tasks.
Show answer and explanation ▾
Correct answer: B
Coded agents are written in code by developers, using their own frameworks and logic, which makes them the choice for highly tailored automations with complex logic that a low-code design cannot express. They still run and are governed inside the UiPath platform.
Why the other options are wrong:
- A. Coded agents sit alongside agents built in the UiPath designer and do not replace them in every scenario.
- C. Coded agents are deployed to and managed through UiPath Orchestrator.
- D. Repetitive, rule-based tasks are the domain of classic RPA robots, not the purpose of coded agents.
Question 66
When configuring a system prompt for an agent, what is the primary purpose of including rules, behaviors, and constraints?
- A. To allow the agent to interpret the user prompt in multiple flexible ways depending on the data accessed.
- B. To provide a general overview of the types of documents or data formats the agent might encounter.
- C. To define the agent’s role clearly and ensure consistent, goal-aligned operation within specified boundaries.
- D. To automatically generate context and escalation pathways for task execution.
Show answer and explanation ▾
Correct answer: C
The system prompt sets who the agent is and how it must act. Rules, behaviors and constraints define its role clearly and hold it inside set boundaries, so it works toward the same goal in the same way on every run.
Why the other options are wrong:
- A. Rules and constraints narrow how the agent acts, they do not invite several loose readings of the user prompt.
- B. Describing document or data formats is background information, not the purpose of rules, behaviors and constraints.
- D. Context and escalations are configured separately on the agent and are not generated by the rules in the system prompt.
Question 67
What is the next design step after you add a UiPath process as a tool so the Agent understands when to use it?
- A. Change the retry policy in Orchestrator settings
- B. Define output mappings in the Agent Memory panel
- C. Add a function description in the system prompt
- D. Map the Agent argument to the tool arguments in Data Manager
Show answer and explanation ▾
Correct answer: C
An agent picks a tool from what it is told about that tool. After adding a UiPath process as a tool, you describe what the process does and when to call it in the system prompt, so the agent can match the tool to the right situation.
Why the other options are wrong:
- A. The retry policy controls how failed jobs are rerun and tells the agent nothing about when to use the tool.
- B. Output mappings in a memory panel do not explain to the agent when the tool applies.
- D. Mapping arguments passes data to the tool, but it does not tell the agent when to choose it.
Question 68
What is the primary significance of understanding the “knowledge context” in agentic discovery when developing automation solutions?
- A. It ensures all automation solutions work independently of data type and source complexity, making them universally applicable to any scenario.
- B. It focuses exclusively on identifying structured information sources like databases but omits consideration of unstructured data, limiting automation scope.
- C. It evaluates how automations directly reduce errors in task execution without considering the type of information or complexity involved.
- D. It helps determine the type and complexity of information required to perform tasks effectively, whether structured or unstructured, enabling more accurate and tailored automation solutions.
Show answer and explanation ▾
Correct answer: D
Knowledge context describes the information a task depends on: what kind it is, how complex it is, and whether it is structured or unstructured. Knowing this lets you design an automation that fits the real task and is more accurate.
Why the other options are wrong:
- A. Knowledge context shows that solutions depend on the data type and source, so it does not make them universal.
- B. Knowledge context covers unstructured information as well as structured sources such as databases.
- C. Knowledge context is about the type and complexity of information a task needs, not a measure of error reduction.
That was 68 of 176.
The full UiPath AAAv1 pack has all 176 questions, each with the answer, the explanation and why the other options are wrong, plus a questions-only copy for timed runs. $39.00 USD, paid once, with free monthly updates and a pass-or-your-money-back guarantee.
Get the full packNot ready yet? Get the first 70 free as one PDF by email. Get the Half Pack →
About the AAAv1 exam
These 68 UiPath AAAv1 questions are free here. The free PDF has 70, free by email.
They are drawn from the same bank as the full AAAv1 pack, which has 176 questions in total.