UiPath Agent Builder L200 Flashcards
(22 cards)
What are the current limitations of Agent Builder regarding productivity activities?
Productivity activities (o365 + Google Drive) are not available as direct IS activity tools in Agent Builder yet.
Workaround: Use workflows or reference microautomations as tools.
How can files be processed as argument inputs to agents?
Agents can’t process files, images, audio, or video directly. Users need to extract text from a document before invoking agents and pass a string as input.
Strings are limited to 2,000 characters. So you may need to pass an array of strings.
Users can use OCR in a workflow before invoking the agent.
What is a workaround for interacting with Generic Resources in Agent Builder?
Users can use workflows containing activities that interact with Generic Resources.
Example: Get Record.
What is the workaround for IS activity tools not having argument support?
Agents can infer fields unless it is a reference field where a static value is required. Otherwise, use a workflow.
What should be done if custom connectors and activities aren’t available as direct tools?
Use a workflow containing the custom activities needed.
What is a recommended practice for designing agents?
Scope agents narrowly to simplify integration, minimize risks, and enable better performance.
What does ‘Human-in-the-loop’ refer to in agent design?
Incorporating human review, validation, and escalation to improve agent reliability and learning.
What are the key components to establish for agent goals?
Define the agent’s objectives, success criteria, and performance metrics.
This ensures alignment with business goals.
What is the purpose of evaluating agents?
To test agents using comprehensive datasets that cover edge cases, adversarial examples, and domain-specific nuances.
What is the first step in creating evaluations for agents?
Define clear evaluation objectives, such as output quality, specific constraints, content accuracy, and format compliance.
What types of assertions can be selected for evaluations?
Assertions types include:
* Equals
* Regex
* Contains
* Factuality
* Custom LLM-graded
What should be included in comprehensive evaluation sets?
Evaluation sets should cover:
* Typical use cases
* Edge cases
* Boundary conditions
* Diversity of inputs
What information can be obtained from evaluation set runs?
Evaluation runs provide:
* Score
* Latency
* Trace
What does ‘deterministic evaluations’ mean?
Predictable and repeatable output under the same given input conditions.
What are the three types of deterministic evaluations in UiPath Agent Builder?
The three types are:
* Equals
* Contains
* Regular expressions
What are the aspects to assess in agent evaluations?
Aspects include:
* Output quality
* Specific constraints
* Content accuracy
* Format compliance
What is the significance of ‘Trajectory based evaluation’?
It checks if the agent did what is expected in the way it is expected to happen.
What is the main goal of continuously reviewing traces and logs?
To monitor agent performance and identify problems.
What is the objective of ‘Iterate to success’ in agent development?
Continuously refine agents, prompts, and workflows through testing and iteration.
What should be done to structure the output of an agent’s evaluation?
Evaluate the most wanted parts of an agent’s output, ensuring they have their own output properties.
What is the recommended number of evaluations in each set?
Aim to have 3 to 5 evaluations in each set to focus on edge cases and coverage.
True or False: Evaluations are a critical step to creating reliable enterprise agents.
True
