UiPath Agentic Automation Associate Certification: 7 Pairs the UiAAA Exam Expects You to Tell Apart

The UiPath Agentic Automation Associate certification (exam number UiPath-AAAv1, usually shortened to UiAAA) is a 90-minute exam with a 70% passing score and a $150 fee. It has no prerequisite, and most of what it asks comes down to one skill: telling two similar-sounding ideas apart.

Those figures come from UiPath's own Agentic Automation Associate exam description, version 1.1, dated August 2025. The same document says the credential stays valid for three years and belongs to the Agentic Automation track of the UiPath Certified Professional program. It does not publish a question count or a weight per section, so any number you see for those is somebody's estimate.

Read the task list in that document and one verb keeps coming back: differentiate. Generative AI from agentic AI. Agents from robots. System prompts from user prompts. Deterministic evaluations from model-graded ones. So this guide does not walk through the sections in order. It sorts the exam into seven pairs and shows how to tell the two halves of each one apart.

What does the UiPath Agentic Automation Associate certification cover?

It covers ten exam sections and three products: UiPath Studio Web, UiPath Maestro and Autopilot for Everyone. UiPath aims it at technical and non-technical roles alike, from business analysts and project managers to automation developers and solution architects.

The "minimally qualified candidate" in the exam description has completed the recommended UiPath Academy training (the Agentic Automation Developer Associate learning plan) and can work independently. For hands-on experience the document lists three things: prompting, basic agent building, and working with large language models.

Here is how the ten sections map to the pairs in this guide.

Four of the seven pairs carry the word "differentiate" in UiPath's task list. The other three (prompt styles, grounding sources and Maestro calls) are my reading of tasks that name two options side by side.

Is it generative AI or agentic AI?

Generative AI produces content when asked. Agentic AI works toward a goal. The first exam section asks you to define ML, LLMs and generative AI, and then to "define agents and differentiate between generative AI and agentic AI."

A plain test helps: does the system only return something, or does it decide what to do next? A model that drafts a reply to a customer email is generating. A system that reads the email, looks up the order, checks whether the refund is within policy and then either sends the reply or hands the case to a person is acting as an agent. UiPath's documentation describes an agent as a software system driven by LLMs that can reason, act and adapt toward a goal, choosing tools and adjusting as it goes.

The trap in scenario questions is the word "AI" used loosely. If a scenario has no goal, no tool and no decision in it, it describes generation, however impressive the output sounds.

Should this step go to an agent or a robot?

Send predictable, rule-based steps to a robot and ambiguous, judgment-heavy steps to an agent. UiPath's side-by-side comparison of agents and robots calls agents probabilistic and adaptive, and robots deterministic and rule-based.

• Robots bring structured execution and reliability. They suit repetitive tasks and start from schedules, triggers or attended runs.
• Agents bring planning, decision-making and natural language. They suit unstructured tasks and start from user requests or system events.

The same exam section asks you to name the challenges of traditional automation "done only with robots and humans" and the benefits of adding agents: intelligent decision-making, adaptability, goal-oriented execution and multi-task coordination. Learn those four phrases as written.

One more idea belongs here: agentic orchestration. UiPath does not present agents as a replacement for robots. It defines agentic automation as agents, robots and humans orchestrated together, so an answer option claiming that agents make robots obsolete is wrong on UiPath's own terms.

A drill you can do on paper: take a process you know, such as supplier invoices, and label every step R, A or H. Typing invoice fields into the ERP is R. Working out what a vague supplier email is asking for is A. Approving a payment above a set limit is H.

Does this sentence belong in the system prompt or the user prompt?

The system prompt holds the agent's role, goal, rules and guidance on when to use tools. The user prompt holds the inputs for this particular run. UiPath's documentation on agent prompts puts it this way: the system prompt describes the role, goal and constraints in natural language, and the user prompt structures how inputs and arguments are passed to the agent.

A small example shows the split better than a definition does.

SYSTEM PROMPT
You are a leave-request agent for the HR team.
1. Read the request and find the employee ID and the dates.
2. Use the "Check Leave Balance" tool before you answer.
3. If the request is longer than 10 working days, raise the
   "Manager Approval" escalation. Do not approve it yourself.
4. Reply in three sentences or fewer.

USER PROMPT
Employee email: {{employee_email}}
Request text: {{request_text}}

Everything that stays the same on every run sits in the system prompt. Everything that changes arrives through arguments in the user prompt, written in double curly braces.

UiPath's page adds a detail that makes a good exam fact: the agent sees only the arguments that are explicitly referenced in the user prompt. Define an input argument, forget to reference it, and the agent never receives the value.

The labels matter as well. Writing Employee email: {{employee_email}} tells the agent which value is which. A bare list of placeholders passes raw values and leaves the agent to guess.

Zero-shot or few-shot: which prompt does the task need?

Use a zero-shot prompt for a simple task and a few-shot prompt when the output has to follow a pattern. The prompt engineering section asks you to "craft zero-shot prompts for simple tasks" and to "design few-shot structured prompts."

Zero-shot means the instruction comes with no examples: "Classify this ticket as billing, access or hardware." Few-shot means you add a handful of worked examples so the model can copy the shape: three sample tickets with their correct labels, then the new ticket.

If you want to see how other builders write theirs, the #uipath tag here on DEV collects community posts about the platform.

The exam description also lists what a well-formed prompt includes: context, instructions, output format and constraints, written with clarity, specificity and conciseness. That is four parts and three qualities, and both lists are short enough to memorise.

Keep one caution from UiPath's writing advice in mind. A few illustrative examples help, but edge cases belong in evaluation sets, not in the prompt itself.

Storage bucket or Integration Service: where does the context come from?

Context Grounding reads from two kinds of source: an Orchestrator storage bucket, or a document system reached through an Integration Service connector. The exam asks you to "define context grounding sources (storage buckets, integration service) and configuration options."

Context Grounding is UiPath's retrieval layer. It indexes business documents so that a prompt can be grounded in them before the model answers. When you build an index you choose where the data comes from, and UiPath's steps for creating a Context Grounding index show exactly two choices under Data Settings: Storage Bucket or Connector.

• Storage bucket: files you upload into a bucket in a shared Orchestrator folder.
• Connector: files that stay where they are, in Confluence Cloud, Dropbox, Google Drive or Microsoft OneDrive and SharePoint, read through an Integration Service connection.

The quick tell in a scenario is where the documents live. "The policies are in SharePoint and change every week" points to a connector. "The team has forty PDFs to hand over" points to a bucket.

For the configuration options, know the search settings an agent applies to an index: the semantic search strategy, the relevance score threshold (raise it and only the best-matching chunks come back) and the maximum number of results.

Deterministic or model-graded: how should this output be scored?

Use a deterministic evaluation when the answer must match exactly, and a model-graded one when meaning matters more than wording. Evaluations are how you show an agent is reliable before it goes anywhere near production.

In the Evaluations page of the Agents user guide, deterministic evaluators (exact match, JSON similarity, CSV column exact match) use rule-based matching, never call a language model and always return the same result for the same input. The model-graded kind appears in the product as "LLM as a judge": semantic similarity, context precision and faithfulness evaluators that use a model to score the output and tolerate reasonable variation in language.

Notice the naming gap. The exam description says "model-graded" and the product screen says "LLM as a judge." They are the same idea. A third category, trajectory, judges the agent's whole run, including intermediate steps and tool use.

Try sorting a few outputs yourself before reading the right-hand columns.

Each evaluation pairs an input with an assertion about the output, and results are scored on a scale from 0 to 100.

Should the Maestro task start an agent or a workflow?

In UiPath Maestro, a Service task can start an agent or start an RPA workflow, and you choose by asking who should do the step. The exam names the option directly: configure Service tasks to invoke agents with "Start and wait for agent," and configure Service tasks to invoke automation workflows from Orchestrator.

Maestro models a business process in BPMN and maps each task to a robot, an agent or a human. In the Service task reference for Maestro, the two actions that matter here are "Start and wait for agent" and "Start and wait for RPA workflow." Both pause the process until the called item finishes, and each one runs as a separate Orchestrator job.

So this pair is less about definitions and more about reading a diagram. If a box in the process says "extract the form fields and post them to the ERP," the Service task should start a workflow. If it says "work out what the customer is asking for," it should start an agent. If it says "manager signs off," that is a human step and a different task type.

What about the exam sections that are not pairs?

Four areas ask for plain recall, not comparison, and they are quick to learn.

• Agentic discovery. Know the five characteristics of an agent story, the format used in the Agent Definition Document: trigger, instructions, context, human in the loop, and the tools it integrates with. Opportunities are prioritised by feasibility and impact.
• Escalations. Human-in-the-loop is how an agent hands a decision to a person. In UiPath, escalations run on Action apps in Action Center, and each escalation carries a prompt that tells the agent when to raise it.
• Platform components. Agent tools are created in Studio Web, each with an input and output schema, and connections are created through Integration Service.
• Autopilot for Everyone. The exam only asks when to use it. It is UiPath's assistant for business users' daily tasks: generating content, answering questions from Context Grounding indexes, and recommending and running automations.

How do you book the exam, and what happens if you fail?

You schedule through Pearson VUE, and a failed first attempt means a two-week wait before the next one. The UiPath candidate page on Pearson VUE sets out the rest: from the third attempt on, the wait is at least one month between attempts, and every attempt is paid for separately.

Rescheduling or cancelling has to happen at least 48 hours before the appointment. Inside that window, or if you do not show up, the fee is lost. The first and last name on your registration must match your government-issued ID exactly.

It also helps to know where this exam sits. UiPath's certification program page lists a new Agentic Automation Builder Professional certification with a launch date of September 22, 2026, so the Associate is no longer the only agentic credential on offer.

How should you practise once the pairs are clear?

Build one small agent, then test yourself on scenarios. The hands-on part matters because every pair above shows up as a setting on a screen: a prompt box, a context source, an evaluator type, a Service task action.

A weekend plan that covers the hands-on list in the exam description:

1. In Studio Web, create an agent with a system prompt and a user prompt that references two arguments.
2. Add one tool and one Context Grounding index.
3. Add an escalation with a clear trigger sentence.
4. Write five evaluations: three deterministic and two model-graded.
5. Call the agent from a small Maestro process with a Service task.

For the reading side, CertFun's UiAAA syllabus with every exam topic on one page makes it easy to tick off tasks as you cover them.

A word of caution from UiPath itself: the exam description opens with an alert about websites selling fake exams. Practice material is for learning to recognise a concept in new wording. A memorised answer key does not survive a reworded scenario.

When you want to see how the concepts turn into question wording, work through these sample questions for the Agentic Automation Associate exam and write down which pair each question is really about. That labelling habit is worth more than the score.

Do the same with your wrong answers. Most misses on this exam are not gaps in knowledge. They are a correct fact attached to the wrong half of a pair.

Once your labels are right most of the time, a timed UiAAA practice test shows whether you can hold that accuracy for a full 90 minutes.

Agentic automation is new enough that the vocabulary is still settling, and that is exactly why this exam leans on distinctions. Get the seven pairs straight and most scenario questions turn into a reading exercise.

Frequently asked questions

1. What is the passing score for the UiPath Agentic Automation Associate exam?

The passing score is 70%, according to UiPath's exam description (version 1.1, August 2025). The exam lasts 90 minutes and the fee is $150.

2. How many questions are on the UiAAA exam?

UiPath's exam description does not state a question count. It gives the duration, the passing score and the fee only, so treat any specific number of questions as unofficial.

3. Is UiPath-AAAv1 the same exam as UiAAA?

Yes. UiPath-AAAv1 is the exam number, and UiAAA is the common short name for the credential, UiPath Certified Professional Agentic Automation Associate.

4. Do I need RPA development experience or another certification first?

No. The exam has no prerequisite exam or certification, and UiPath includes non-technical roles such as business analysts and project managers in the target audience. It does expect the recommended Academy training plus some hands-on prompting and basic agent building.

5. How long does the UiPath Agentic Automation Associate certification stay valid?

It is valid for three years, as stated in the exam description.