Governance, Deployment & Methodologies for Agentic Automation [2/3]

Agentic Automation Blueprint:
From Strategy to Scale

Governance, Deployment &
Methodologies for Agentic
Automation
The UiPath word mark, logos, and robots are registered trademarks owned by UiPath, Inc. and its affiliates. UiPath (R) is a registered trademark in the
United States and several countries across the globe. See TMEP 906. ©2025 UiPath. All rights reserved.

Speakers
Yuvatheja ER
Technical Account Manager
Shreya Ramprakash
Associate Technical Account Manager

01
02
03
04
Brief Introduction to agentic automation
Building & Managing the Agentic Pipeline
Designing for Change & Future-Proofing
Value Realization & Stakeholder
Communication
Human-Centric Design & Adoption
05
Recap of Session 1: Building a Value-DrivenAgentic Pipeline

Agenda
01
02
03
04
UiPath Governance Framework
Governance - AI Trust Layer
Governance - Automation Ops
Governance Promises and Hero
Scenarios
Unified Deployment
Deployment Methodologies
Quiz and Q&A
05
06
07

Agentic Governance ensures every action an
agent takes is reasoned, restricted, and recorded
—by policy, not hope.
It ensures agents—like people and robots—can only take
approved actions, on approved data, in approved conditions,
based on centrally defined and enforceable policies.
IT Governance is the rulebook that decides
who touches what — and how hard.
It protects sensitive automations with enterprise-grade
identity, role controls, and audit trails — so access is
always intentional, limited, and traceable.
Infrastructure Governance is the backbone of
trusted automation.
It enforces data residency, encryption, network isolation,
and compliance with global and industry-specific standards
(like GDPR, HIPAA, FedRAMP, ISO 27001).

Agentic
Governance
IT
Governance
Infra
Governance
Data Residency
Security Hardening
Industry / country
specific compliance
Network Boundary
Encryption / CMK
Identity
Audit & Logs
Delegation of Admin
Data Governance
Agentic
Processes
RPA
Workflow
API
Workflow
Agents Apps
Granular / Folder level
RBAC controls
Agency
Guardrails
NL Policy
Design
Project
Guardian
Data Access
Policies
Safety
Monitoring
Guardrails

Governance -AI Trust Layer
9
Trust Boundary
UiPath Products
& Services
Autopilot™
Communications Mining
Document Understanding
Process Mining
Test Manager
Apps
Marketplace
Studio Web
ACR/Serverless
GenAI Activities
GenAI Connectors
UiPath AI Trust Layer
UiPath Platform Services
Cloud Identity Service Azure Apps Insights Automation Ops
Semantic Search* Azure Cache for Redis UiPath Orchestrator
UiPath-managed
AI Models
UiPath-Managed
third-party LLMs
UiPath-Managed
third-party AI models
HTTPS
(TLS 1.2+)
UiPath
Products
Zero
data
retention
or
training
LLM Gateway
API
• Normalized Chat
• Embeddings
• PII detection
Capabilities
• Authentication
• Rate Limiting
• Governance
• Telemetry
• Monitoring/Audit
LLM Gateway
LLM Proxy
*private preview
HTTPS (TLS 1.2+)
Customer managed 3rd Party LLMs
Azure OpenAI (GPT) Google Vertex AI (Gemini) and many more…
HTTPS (TLS 1.2+)
All In Transit TLS 1.2 +
All At Rest AES 256
We manage all services within Trust
Boundary and customer data is never
retained/reused by our partners
Data Encryption

Governance -Automation Ops
Agentic Security &
Compliance
 Data encryption (at rest and in
transit)
 Identity access management
 Audit trails for agent actions
 Controlled rollouts and failover
mechanisms
 Enterprise grade- guardrails
AI Trust Layer Policies
 Enforce Ethical boundaries
 Agent access control
 User access control
Role based Access
Control (RBAC)
 Role based governance
Policy Deployment
Flexibility
 Per tenant
 Per group
 Per user
Integrated Controls
 Studio & Assistant
Governance
 Source Control
 Pipelines(CICD)
 Version Control
Environment and Folder
Strategy
 Multi Tenant -Dedicated and
isolated environments
 Custom governance policies
 Folder level controls
Automation
Ops
T
O
O
L
&
P
O
L
I
C
Y
G
U
A
R
D
R
A
I
L
S
A
I
T
R
U
S
T
L
A
Y
E
R
G
O
V
E
R
N
A
N
C
E
C
O
M
P
L
I
A
N
C
E

UiPath
What’s to stop an autonomous agent from
triggering actions we never approved?
How can we ensure there’s always a human
reviewer in the loop for sensitive decisions?
How can developers be sure they’re not
building agents that bypass safety rules?
How do we know agents are tested, scored,
and safe before they go live?
How can we prove to legal that every agent
follows the rules?
When auditors ask who approved what and
under which policy, how do we show that
instantly?
How can we stop sensitive data like PII from
being exposed when agents call models?
How do we ensure the right model is selected,
used in the right region, and governed for
accuracy and compliance?
UiPath’s Agentic Governance
Your questions might be…
Automate processes with deterministic robots,
autonomous agents, and human reviewers
Controlled agency
Agent reliability
Centralized policies
LLM Governance
Agent score and agent guardrails give
developers the tools to build responsibly
Centralized policies ensure agents
operate within scope of compliance
Protect data being sent to models
while ensuring accuracy of automations
How

UiPath
01
02
03
04
05
‘How can we be sure our
data isn’t being used to
train models?’
‘What about third-party
AI vendors — are they
holding onto our data?’
‘Will using AI still allow
us to meet our data
sovereignty needs?’
‘What prevents AI from
accidentally pulling data
from the wrong system?’
‘Will AI models see
sensitive data like PII
or PHI?’
Never use
customer data to
train AI models
3rd party AI vendors
never use/retain
customer data
You control where
models and data are
stored and processed
AI features can
never access
unauthorized data
AI will not see personal
data, your workflows
retain 100% accuracy
Agentic governance
Your concern How
Customer Opt-Out for AI Training
Support for opt-out mechanism ensuring no
customer data is used for model training.
Supplier Compliance Tracking
Require all third-party LLM providers to
certify that no customer data is retained
and no training is performed on customer
data.
LLM Configuration/Bring Your Own Model
Ensure all GenAI activity complies with local
data residency and sovereignty
requirements.
Restricted Data Access Controls
Enforce folder isolation, role-based access
and coming soon with centralized policy so
AI features only access allowed data.
Mask/Filter PII before it's sent to any model
AI Trust Layer automatically masks/filters
out personal identifiers in prompts,
outputs
UiPath Governance Promises

UiPath
01
02
03
04
05
Rogue Agent Loops
Out of Control
Conversational Agent
goes out of bounds/context
Employee includes sensitive
data in a prompt
Over-Privileged Agent
Misuses Authorized Access
Cost Overruns from
Misconfigured Agent
An AI agent misinterprets its
prompt and triggers unintended
actions repeatedly
Conversational Agent responds with
unsafe/harmful content that bypasses
the context or system prompts
Sensitive user data is
accidentally included in a GenAI
prompt
An AI agent given excessive
permissions applies its access beyond
intended use, triggering actions or
data flows meant for higher-trust
scenarios.
A faulty loop causes an agent to call
LLMs thousands of times
Description
Scenario Governance Response
Controlled Agency: Governance
constrains agents to pre-approved tools
and steps. Confidence drops trigger
human escalation before harm occurs.
Jailbreak Protection: policy-constrained
prompt execution and jailbreak detection
ensure agent output aligns with
enterprise policies.
PII In-Flight Masking: Personal data is
automatically redacted before prompts
reach external models. Logs show exactly
what was masked and sent.
Human-in-the-Loop: All GenAI outputs
require explicit review and approval. No
action is taken without human validation
and audit tracking.
Usage Monitoring & Controls: Admin
dashboards detect anomalies. Token
limits, model kill switches, and rate caps
prevent runaway costs.
UiPath Governance Hero Scenarios

UiPath
Project Guardian: continuously monitors agent behavior and policy coverage. It flags a drift in expected output and
triggers a governance alert: “Detected sensitive data flowing to external endpoints. Recommend policy intervention.”
06
UiPath Governance Hero Scenarios
Policy Drift Causes Risky Agent Behavior
Governance Responses
NL Policy Designer: the platform owner uses natural language to respond:
“Block all agents from sending salary data to external APIs.”
This is instantly converted into an organization-wide policy, enforced across all agents — with full audit trail.
As automations scale, agents evolve. A previously safe AI agent is now exposing
sensitive salary data to an external API — a subtle but dangerous policy drift.
Description

Agentic orchestration End-to-end process lifecycle
Powered by Maestro
Model & implement - Build and deploy
processes in Studio Web
• Standard BPMN 2.0 constructs
• UiPath extensions
• Advanced decisioning
• Autopilot™ for Maestro
Operate - Orchestrator and Temporal
Service
• Process instance management
• Operational analytics
• Pause, resume, edit variables, retry, cancel, go to
step, and migrate.
Monitor and optimize - Process Intelligence
• BPMN models
• End-to-end processes
• Agents
• Automations
• HITL
End-to-end orchestration across agents, robots and people.
Connectors / APIs - Business applications and data sources
Agents
Prompt, Tools, Context,
Evaluations, Escalations
Robots
UI Automation, API
Workflows, IDP
People
Action Center, Apps, Assistant

Deploying yourAgents
You define the guardrails. UiPath enforces them.
Outcome-driven
Agents
Guided Build
Experience
Enterprise-grade
Guardrails
Purpose-built,
context-rich agents.
Anyone can create powerful
agents – FAST.
Trust, choice &
visibility built-in.
Value components:
 Low-code/no-code wizard
 Unified visual builder
 Prompt auto-tuning + Health score
 Instant test-run within workflows
 Bring your own LLM (Vendor agnostic)
 Appropriate supervision
 Dynamic evals at design & run time
 Full audit log and governance
 Context Grounding
 Automations as tools
 Other agents as tools
 Attended and unattended execution
 Multi Agent workflows
01 03
02

Orchestrate AI agents, robots, and
people to exceed business
outcomes.
• Integrate AI agents, automation, and
people into end-to-end processes
seamlessly
• Scale automation to complex, intelligent
processes
• Simplify processes—and the way you
manage them

How agentic orchestration works
across the end-to-end process
lifecycle
IXP AUTOMATIONS
HITL
AI AGENTS
END-TO-END
PROCESSES
Model Implement Operate Monitor Optimize
Model end-to-end
processes in Studio Web
using BPMN, case modeling
or dev-friendly non-BPMN
flows, with decisions/policies
as guardrails
Implement and deploy
agents (UiPath or 3rd
Party),
automations, human-in-the-
loop tasks, and APIs in a
single orchestration layer
Operate processes with
real-time visibility and
control, resolve exceptions,
and allow people to
intervene when needed
Monitor processes with in-
model heatmaps, role-
based KPIs, and business
user insights via process
apps
Optimize processes by
uncovering bottlenecks and
opportunities, driving
continuous improvement
and ROI from agentic AI.

Business rules
configuration
Implement
Embed decision intelligence to
ensure governance and consistency
across agentic processes.
Leverage Decision Model &
Notation (DMN) to add business
rule tasks and manage decisions
directly within processes.
Import, export, and version
decision models for easy
integration with external systems
and ability to track and manage
changes.

20
Orchestrate agents,
robots, and people
across end-to-end
processes
Implement
Seamlessly integrate
agentic tasks (UiPath or
third-party), RPA, human-in-
the-loop, and API events into
your processes to maximize
productivity.
Test, debug, and validate
before deploying processes
to production.

21
Integrating with third-
party agents
Implement
Connect and coordinate the best
agents for each task. Vendor
agnostic, fully open, and seamlessly
integrated. UiPath Maestro is the
connective tissue that de-silos your
agents and empowers your
processes.
Compatible with leading third-party
agents:
• Salesforce Agentforce
• Microsoft Copilot Studio
• Microsoft Foundry
• Google Vertex AI
• Crew AI
Azure AI Foundry

UiPath Autopilot
Unified Deployment
Vertical / Specialized Agents
Other agentic assistants e.g.; SAP Joule
SAP Oracle
UiPath
(PEAK)
SFDC Others/ Industry
Unified development
Design, build, test, and deploy
Low-code & pro-code
RPA Workflows
Attended, Unattended, UI
Agents
Coded Agent, BYO
API Workflows
Connectors, business
applications, data sources
Flexible deployment
Cloud, dedicated, on-prem, air gapped
Unified administration
Access management, OOTB roles, audit
Assistant
/
UI
End-to-end
automation
and
orchestration
Specialized
/
vendor-
specific
OOB
automations

23
Best practices for publishing and deploying agents
Essential gate What to check Where to do this
Prompts and examples finalized System/User prompt includes role,
constraints, 3–5 input-mapped examples
Agent Builder → System and User
Prompt
Tools described and bound All tools have name, description,
input/output schema
Agent Builder → Tools
Guardrail logging enabled (optional) Tool calls are logged for audit/debug
(enable in guardrail configuration)
Tools → Guardrail builder
Context sources connected At least one relevant knowledge base is
grounded
Context Grounding → Sources
≥30 interactive tests conducted Manual tests cover typical, edge, and
malformed inputs
Agent Builder → Test Run
Evaluation set(s) created ≥30 curated test cases, covering real-
world usage
Agent Builder → Evaluations tab
Evaluation performance validated Evaluation set(s) score ≥70% with no
regressions
Agent Builder → Evaluations tab

Governance, Deployment & Methodologies for Agentic Automation [2/3]

More Related Content

Similar to Governance, Deployment & Methodologies for Agentic Automation [2/3]

More from suhanisingh58689

Recently uploaded

Governance, Deployment & Methodologies for Agentic Automation [2/3]

Editor's Notes