Session 1 - Agentic Automation Building the Enterprise Agent of Tomorrow

Discover UiPath Agent Builder with a focus on licensing essentials and practical guidance for creating enterprise-ready agents, including seamless integration with SAP and Salesforce. This is a virtual event being held by the UiPath California chapter.

Topics Covered:

• Introduction: Identifying enterprise-level challenges that can be addressed using UiPath Agent Builder

• Build, Test & Deploy: Step-by-step walkthrough to design, develop, test, and deploy enterprise-grade agents using UiPath Agent Builder

• Interactive Q&A: Get your questions answered by experts in a live discussion

• Agentic Orchestration with UiPath Maestro

1. Building the Enterprise Agent of
Tomorrow
Session –1 of 4 | Virtual Session by UiPath Community

2. 2
Agenda
Architecting Enterprise-Grade Automation Agents
Guardrails: Your Agent’s Built-In Safety Protocol
Context Grounding – Best Practices | Escalations : Why
escalations are important foragents
Testing & Evaluation Strategies
Build Your First Agent (Live Demo)
Q & A

3. 3
Meet the team!
Lead Consultant @ The Silicon Partners Inc.
UiPath MVP
Amit Tiwari
SVP Technoloyg, @The Silicon Partners, Inc.
UiPath MVP
Pankaj Banka
Co-Founder & COO @The Silicon Partners, Inc.
UiPath MVP
Diana Gray
Community Marketing Manager AMER
@UiPath
Mohd Faiz Khan

4. 4

5. 5
Evolution of Agentic AI Automation
Journey from Rule-Based Automations to Agentic AI
This evolution marks a significant shift in AI capabilities, where AI systems are designed to be proactive, engaging
in complex reasoning and decision-making processes.
This transformation is not just technological but philosophical, reshaping our approach to system design, product
thinking, and human-machine collaboration.
Process Automation Supervised Learning
Rule-based workflows to
manage end-to-end
processes, limited to
specific tasks.
Fixed machine learning
models trained on
historical data i.e.,
invoices and ID cards.
Generative AI
Large language models
began to respond to
prompts and contextual
information, interaction.
Agentic AI
The pinnacle of this evolution combines
Generative AI with automation to create
“Autonomous Agents " equipped with
memory.
These agents are not reactive but
proactive, driven by intentions and
capable of multi-step reasoning.

6. A unique combination of capabilities
Robots
Rules-based,
act predictably,
deterministic decisions
Agents
Goals-based,
act independently,
dynamic decisions
Best for tasks that require
high reliability & efficiency
Best for tasks that require
high adaptability

7. Processing Vendor Invoices
Robotic Workflow
Action
2-way match of vendor
invoice against the PO
Model
Invoice Model (DocPath)
to process invoice
Action
Update Records In ERP
Invoice Created Completed
Human
Dispute Resolution
Robot
Financial Accounting
Robot
Invoice Review
Human
Dispute Investigation
Manual
Investigate mismatch to
determine if a dispute is
warranted
Manual
Communicate with
Supplier via email
Error-prone, takes hours Distracts from business-
critical tasks, takes days

8. Processing Vendor Invoices
Robotic Workflow
Action
2-way match of vendor
invoice against the PO
Model
Invoice Model (DocPath)
to process invoice
Action
Update Records In ERP
Invoice Created Completed
Human
Dispute Resolution
Robot
Financial Accounting
Robot
Invoice Review
Human
Dispute Investigation
Manual
Investigate mismatch to
determine if a dispute is
warranted
Manual
Communicate with
Supplier via email
Error-prone, takes hours Distracts from business-
critical tasks, takes days

9. 9
Key components of a Robust Automation Agent
Guardrails
•Built-in safety checks for agent tools.
•Trigger actions: log, block, filter, or escalate.
•Ensures compliant execution paths.
•Example: Block tool if PII is detected or
escalate if approval > $10,000.
Context Grounding
•Feeds agents with business-specific knowledge.
•Storage Buckets: For stable mappings, SOPs, templates.
•Connectors: For dynamic sources like SharePoint or
Confluence.
•Why it matters: Enables accurate reasoning.
Effective Prompt Writing
•Clear, structured prompts guide agents
effectively.
•Use few-shot, chain-of-thought, or JSON-
formatted prompts.
•Define format, tone, and expectations.
•Pro Tip: Always start with the outcome in
mind.
Escalations
•Human-in-the-loop handoff for sensitive or
high-risk decisions.
•Triggered by guardrails or agent uncertainty.
•Example: Escalate HR issues or legal risks
to managers.
•Adds reliability and auditability.
Testing & Evaluation
•Evaluate agent behavior using real or simulated data.
•Use LLM-based, exact match, or JSON similarity
evaluators.
•30+ meaningful evaluations = healthy agent.
•Pro Tip: Focus on edge cases and realistic flows.

10. 10
Guardrails: Your Agent’s Built-In Safety Protocol
Why Guardrails Matter
•Prevent unpredictable tool behavior at runtime
•Block actions that violate policies or logic
•Trigger smart escalations only when truly needed
•Enforce enterprise-grade compliance and trust
How Guardrails Work
IF certain conditions are met,
THEN take a predefined action — before or after the tool runs.
Built Per Tool → You can add multiple guardrails per agent
tool.
Rules + Actions → Define when and how your agent reacts.
Action What It Does
Log Create logs with severity (Info, Warn, Error)
Filter Hide sensitive fields from input/output
Block Stop tool execution completely (with reason)
Escalate Trigger human review through escalation app
Guardrails are predefined rules that control agent behavior, ensuring safe, compliant actions by triggering logs, filters,
blocks, or escalations.
Pre-Execution Guardrails
Implement checks to validate inputs and environmental conditions before agent execution.
Post-Execution Guardrails
Verify outputs and ensure actions comply with business rules and security policies.

11. 11
Context Grounding – Best Practices
Context groundingwithstorage buckets
Storage Buckets offer a centralized, secure, and version-controlled space within UiPath Orchestrator to store static or semi-static reference files,
enabling reliable and consistent grounding for agents.
Pro tip: Choose connectors for dynamic content and storage buckets for static files to ensure reliable grounding and minimize maintenance.
Context Grounding: Why it Matters?
•Grounding helps agents make smart, context-aware decisions.
•Both Connectors and Storage Buckets enable access to external knowledge.
•Data access method impacts performance, reliability, and scalability.
•They differ in functionality, flexibility, and best-use cases.
•Choosing the right method ensures efficiency and reduces maintenance effort.
Context grounding ensures agents use accurate, business-specific data. Without it, they rely on generic knowledge—like
asking a general doctor to perform surgery without specialized expertise or current information.
Connectors enable real-time access to dynamic business content—such as Live CRM records, Support ticket statuses (Zendesk, Freshdesk etc.
allowing agents to stay updated with the latest information.
Context groundingwithConnectors

12. 12
Context Grounding – How does it work and why it matters?
The key benefits of context grounding
• Accuracy: Instead of generic responses, your agents provide answers based on your actual policies and procedures.
• Relevance: Agents can access up-to-date, company-specific information rather than outdated general knowledge.
• Reliability: Reduces “hallucinations” (when AI makes up information) by grounding responses in verified data sources.
• Compliance: Ensures responses align with your organizational standards and regulatory requirements.
How does context grounding work?
Context grounding operates through a process called Retrieval Augmented Generation (RAG). Context grounding is a key part of RAG.
In simple terms, it is like giving your agent a research assistant. Instead of relying solely on its training, the agent first searches for
relevant information in your knowledge base, then uses that information to craft its response. This ensures answers are both intelligent
and accurate.

13. 13
Effective Prompt Writing: Guiding Your Agent
Crafting clear, concise prompts is crucial for effective agent performance.
<goal> Defines the agent's objective.
<tools> Specifies allowed tools/activities.
<response> Dictates the desired output format.
{{Input}} Placeholder for dynamic input data.
Prompt Type Best Used For Agent Use Case Example
Zero-Shot Simple classification Tagging request as IT/HR
Few-Shot Pattern-based tasks Summarization, reply formatting
Chain of Thought Multi-step reasoning Leave policy validation
ReAct Tool-based decision making Should agent call Jira API or escalate?
Self-Reflection Reducing hallucination, improving reliabilityHR policy clarification
Structured Output Consistent outputs for automation Return user, type, urgency in JSON
Instruction (System/User)Agent personality, tone, guardrails Compliance-focused finance agent
Common Prompting Techniques for Agents

14. 14
Escalations : Why escalations are important forAgents?
Escalations: The Agent’s Built-In SOS Button
Imagine your employee support agent receives one of these queries:
“Can you help terminate an employee due to misconduct?”
“I’m not being paid fairly. What does our HR policy say?”
“Can you override my manager’s approval on this request?”
Should your agent try to resolve these on its own?
No. Even the smartest agent shouldn’t play HR Director, Legal Advisor, or CFO.
Key Takeaway
Smart agents know what they don’t know
Escalations are not failures—they’re built-in safeguards for enterprise-grade trust and control.
Best practice:Before creating an escalation, ask: "What situations truly require human intervention?" Not every agent
decision needs a manual check. Reserve escalations for moments where user intent is unclear, data is missing, or business risk
is high. This keeps the human-in-the-loop experience efficient and meaningful.
Escalations are like the agent raising a hand and saying: “This looks important. Better loop in a human.”
They ensure that:
Sensitive matters get the right level of review
Complex edge cases aren’t mishandled
Your agent doesn’t overstep its role

15. 15
Testing & Evaluation Strategies
Testing & Evaluation Strategies ensure your agent performs reliably by simulating real-world scenarios, validating
outputs, scoring accuracy, and refining behavior—ultimately building trust before deployment at scale.
Evaluations : Why It Matters
Before trusting your agent in production, test if it can:
•Make correct decisions
•Handle edge cases
•Respond consistently
•Avoid unexpected behavior
What’s an Evaluation?
Think of it as a test case:
Input +
Expected Output +
A rule (Evaluator) =
A pass/fail score (0–100)
Evaluator Type Use Case
LLM-as-a-Judge Semantic accuracy in natural language
Exact Match Precise matching (e.g. "Yes" or "No")
JSON Similarity Complex structured output validation
Trajectory Evaluates reasoning path of agent
Types of Evaluators
Pro Tip: Use simulations early in development to catch logic flaws without hitting live systems—this saves costs, avoids API throttling,
and supports rapid iteration.

16. 16
Demo : Employee Support Agent – Breakdown
Problem Context
•Organizations handle diverse employee support requests: IT
issues, HR queries, facilities, and more.
Agent Objective
•Build an intelligent employee support agent to automate
request classification and resolution.
Agent Capabilities
•Understand employee queries using natural language.
•Identify and classify the request into correct problem
categories.
•Match the category to the relevant support group using a
reference CSV.
•Respond with appropriate action steps or next-level
instructions.
Data Handling
•CSV mapping file is securely stored in a Storage Bucket
(UiPath Orchestrator).
Outcome
•Faster triaging of support issues.
•Reduced manual intervention.
•Improved employee experience and operational efficiency.

17. Live Demo…..

18. Q & A

19. 19
Register for Next Session : -