Getting Started with Agentic Playbooks - Workflow Automation CoE

Lisa Holenstein
ServiceNow Employee

2 hours ago

Workflow Automation CoE > Playbooks > Getting Started with Agentic Playbooks

 

Goal

In this Workflow Automation Center of Excellence article, you will learn how to get started with Agentic Playbooks in Workflow Studio to accelerate playbook execution and increase agent productivity by embedding AI workflows and AI agents into your deterministic workflows at runtime.

 

Technology

Product/Feature

Description

Workflow Studio

Central low-code workflow automation builder consolidating Flow Designer, Process Automation Designer, Decision Builder, and Action Designer

Playbooks

End-to-end process automation with guided user experience layered on top of flows and subflows

Playbook Experience

UI framework that improves end-user productivity, ensures response consistency, and reduces errors

Now Assist AI agents

Now Assist AI agents are entities that mimic human-like intelligence by using large language models (LLMs). AI agents can perform tasks that range from simple automated responses to complex problem solving.

Agentic Playbooks

AI agents embedded within playbook activities to automate data gathering, processing, and form filling

 

Watch the Creator Toolbox video about Agentic Playbooks

 

 

Introduction

 

What Are Agentic Playbooks?

Agentic Playbooks are an enhancement to ServiceNow’s playbook capabilities that embed AI agents directly into playbook activities. This feature closes the loop between deterministic workflows (which ensure reliable, predictable outcomes) and AI agents (which accelerate task execution and reduce cognitive load on human agents).

Key Concept: Agentic Playbooks operate within your existing enterprise workflow rules, policies, and logic. AI agents assist human workers by automating tedious data gathering and processing tasks while the overall process flow remains governed by your established business rules.

 

Business Value

Agentic Playbooks address common productivity bottlenecks in process automation:

Challenge

Agentic Playbook Solution

Human agents spend significant time gathering data from multiple sources

AI agents automatically fetch data from AI Search, Knowledge Graph, and web search

Manual form filling is time-consuming and error-prone

AI agents intelligently populate form fields with processed data

Reviewing lengthy knowledge articles and mapping them to cases takes considerable effort

AI agents summarize content and extract relevant information in seconds

Complex calculations and text processing slow down case resolution

AI agents perform mathematical calculations and text summarization automatically

Agents need to compare current cases with historical data

AI agents analyze patterns across past cases and provide recommendations

 

When to Use Agentic Playbooks vs. Traditional Playbooks

Understanding when to apply Agentic Playbooks versus traditional deterministic workflows is crucial for successful implementation.

 

Use Agentic Playbooks When:

Scenario

Rationale

Data collection from multiple sources

AI agents excel at gathering information from knowledge bases, previous cases, and external sources

Text summarization and generation

Tasks like summarizing case histories, KB articles, or incident details benefit from generative AI

Pattern recognition across historical data

Comparing current issues with past cases to identify solutions

Research-intensive activities

When agents need to review extensive documentation or search for specific information

Slot-filling complex forms

AI can intelligently populate form fields based on context and gathered data

 

Example Use Cases

  • Mapping a 50-page knowledge base article to a laptop provisioning request
  • Summarizing a case with all available information and comparing it to past cases
  • Triaging potential problems and solutions based on historical patterns
  • Gathering prerequisite information before approving hardware requests

 

Use Traditional (Deterministic) Playbooks When:

 Scenario

Rationale

Heavily regulated processes

Industries with strict compliance requirements (financial services, healthcare) need guaranteed outcomes

Audit trail requirements

When processes must produce identical outputs under identical conditions for legal/regulatory review

Simple conditional logic

Straightforward if/then scenarios that don’t require data interpretation

Critical business decisions

High-stakes outcomes where non-deterministic results pose unacceptable risk

Simple bundling or routing

Basic task assignment and routing based on clear criteria

 

Important distinction: AI agents are non-deterministic, meaning they may produce different (though contextually appropriate) outputs for the same inputs. If your process requires exactly the same result every time under the same conditions, use traditional flows, subflows, decision tables, and flow actions instead.

 

Prerequisites and Requirements

Before implementing Agentic Playbooks, ensure your instance meets the following requirements:

 

Version Requirements

Component

Minimum Version

Family Release

Zurich

Workflow Studio

28.0.5 (Zurich true up)

Process Automation Designer

28.2.3

Playbook Experience

28.2.2

Now Assist for Platform

9.1.0

Licensing

Any Pro Plus License (please contact your account team for details)

 

Supported Playbook Activities

Not all playbook activities currently support agentic capabilities. The following activities are enabled for AI agent integration (as of December 2025):

  • New Record Form
  • New Task Form
  • Questionnaire
  • Email Form
  • Record Form
  • Autocomplete Record Form

 

Required Roles

Authoring Roles

To create and configure Agentic Playbooks, users need standard playbook authoring rights:

  • admin
  • playbook.admin
  • pd_author
  • playbook.write

Note: No special roles are required for authoring Agentic Playbooks beyond standard playbook authoring permissions.

 

Runtime Roles

End users who will interact with Agentic Playbooks need one of these roles:

Role

Purpose

playbook.agent_user

Grants access to standalone playbook agents

playbook.agentic_workflow_user

Grants access to Playbook Activity Assist Workflow and Now Assist Panel integration

 

Note: If the end user or agent working the playbook does not have either of these roles, they can still access and complete the playbook and its activities, they’ll just need to complete the steps manually.

Example Use Case: Contract Renewal Playbook

AI Agent activity configuration

Enabling AI agents in your playbook activities is remarkably straightforward. When you open any supported activity in Workflow Studio, you’ll now see a fourth tab labeled AI Agent alongside the familiar Details, UI Layout, and Automation tabs. Simply toggle the AI Agent switch to enable the feature, then configure two essential elements: the impersonating user and the agent instructions.

The impersonating user determines which permissions the AI agent operates under. You can choose the person who triggered the playbook or a specific user that is available from the playbook records and data. This ensures proper access control and maintains audit trails while the AI agent gathers data and performs its tasks.

The real power comes from the agent instructions, where you write natural language guidance telling the AI what to accomplish. Use a step-by-step format, specify the exact fields and data sources the agent should access, and define the expected output format. For example, if you’re building a contract renewal workflow, your instructions might direct the agent to search for discount policy KB articles, fetch contract duration and value from specific fields, calculate the total contract value, and recommend an appropriate discount based on policy rules.

The agent instructions use conversational language rather than code, making this accessible to process owners who understand the business logic without requiring deep technical expertise. There’s no need to learn a new programming language or understand complex AI orchestration; if you can write a clear email explaining what needs to be done, you can write effective agent instructions.

 

Enhanced feature: Starting in December 2024, you can configure activities to auto-complete when AI agents provide high-confidence results.

In the Activity Settings, configure Complete activity independently:

Choice

Description

On

The AI agents update the record, complete the activity, and automatically move the playbook to the next activity.

Off

The activity waits until a human reviews and approves the AI generated outputs.

 

Note: Test the playbook extensively to make sure that the AI agents can complete the activities independently. Select View progress while testing the playbook to see the agent activities on the Now Assist panel.

For activities that cannot be completed independently by the AI agents, the system saves the data. The activity is completed if the data matches any Wait for condition in your playbook.

 

Important Note: This feature is currently in a controlled rollout, so you might not see the option immediately after updating the store apps to the December version. Please submit a Support case to have the option enabled for the respective activity definitions on your instances.

 

 

Runtime experience - AI Agent work in progress

When a playbook reaches an agentic activity, the user experience transforms into an interactive collaboration between human and AI. Form fields become temporarily locked with a clear indicator that Now Assist is working behind the scenes for this activity. Users can click View Progress to open the Now Assist side panel that provides real-time visibility into exactly what the AI agent is doing.

The panel displays a running commentary as the agent completes each task: “Accessed contract record,” “Retrieved contract duration: 5 years,” “Searching knowledge base,” “Found KB0012345: Discount Policy,” and “Recommending: 17% discount.” This transparency is crucial for building user trust, workers can see which data sources the agent is querying, what information it’s extracting, and how it’s processing that data.

The side panel also includes an interactive chat interface, allowing users to ask follow-up questions or request clarification while the agent works. For instance, if an agent is calculating a contract discount, a user might ask, “What discount range does the policy specify for 5-year contracts?” and receive an immediate answer.

This real-time progress view typically lasts 10-30 seconds depending on the complexity of the task, with agents gathering data, performing calculations, and preparing recommendations far faster than manual research would allow. What traditionally took 15-20 minutes of searching through documentation now happens in seconds while the user watches.

 

 

Runtime experience - AI Agent work completed

Once the AI agent completes its work, the form fields unlock and display the agent-populated data, ready for human review. In our contract renewal example, users would see the Discount Percentage field now contains “17,” accompanied by detailed reasoning explaining why this discount was selected: “Tier 3: Enterprise. Total cost $42330.44. Monthly subscription. Discount range 15-20%. Applied 17%.”

Importantly, all fields remain fully editable. This is human-in-the-loop automation, not autonomous decision-making. Users can accept the AI’s recommendation as-is, modify any values based on their judgment, or completely override the suggestion if business circumstances require it. The side panel remains accessible, now showing the sources the agent used, allowing users to click through to the original KB articles or records for verification.

This moment represents the core value proposition of Agentic Playbooks: what might have taken a human agent 15-20 minutes of searching through documentation and performing calculations has been reduced to a 10-second AI process, with the human providing final validation and approval in seconds rather than minutes. The agent does the tedious research and data gathering, while the human focuses on judgment and decision-making.

 

 

Runtime experience - AI Agent work summary

After an activity is completed, the playbook maintains a comprehensive history of what the AI agent accomplished, creating valuable documentation for audit, training, and optimization purposes. Users can access an Activity History view that provides a concise summary of the agent’s work without requiring them to review the entire conversation log.

For example, the summary might state: “AI Agent searched KB0012345 (Contract Renewal Discount Policy), retrieved contract duration (5 years) and monthly payment ($10,000), calculated total value ($600,000), and recommended 17% discount based on policy guidelines for contracts of 5+ years and values above $500K.” This summary includes the data sources consulted, the key information extracted, the processing or calculations performed, and the final recommendations provided.

More detailed users can click View Full Conversation to see every step of the agent’s orchestration, including alternative options it considered and the reasoning behind choosing specific data sources. This documentation serves multiple purposes: it provides an audit trail showing what information informed business decisions, helps identify opportunities to refine agent instructions for better performance, and serves as training material for new team members learning the process.

Organizations can analyze these summaries across multiple playbook executions to track agent accuracy rates, identify patterns in user modifications, and continuously improve their agentic workflows over time. This creates a feedback loop where the system becomes more effective with each use, learning from real-world performance data.

 

Conclusion

Implementing Agentic Playbooks allows you to combine the reliability and governance of deterministic workflows with the productivity benefits of AI agents. By following this guide, you can:

  1. Accelerate process execution by automating tedious data gathering and form filling tasks while maintaining human oversight for critical decisions
  2. Improve agent productivity by reducing cognitive load and allowing human agents to focus on judgment and decision-making rather than information lookup
  3. Maintain compliance and control through transparent AI operations that work within your existing enterprise workflow guardrails and business rules

Remember: Start with use cases involving data collection, summarization, and pattern recognition. Always maintain human oversight for final decisions and continuously monitor and optimize based on performance metrics and user feedback.

 

Resources

ServiceNow Documentation

Center of Excellence Articles

Training & Learning

Additional Resources

Academy Sessions

 

 

Center of Excellence Navigation

  1. Workflow Automation - Center of Excellence
  2. Workflow Automation Migration Considerations
  3. What's new for Workflow Automation?
  4. Resource Hubs
    1. Flows
    2. Decision Tables
    3. Playbooks
  5. FAQ
  6. Checklist
  7. Training
  8. Workflow Academy
0 Helpfuls
  • 19 Views
Version history
Last update:
2 hours ago
Updated by:
ServiceNow Employee
Contributors