arunpundir
ServiceNow Employee

Agentic Desktop – Bringing AI to Legacy Systems

For traditional organizations that rely on — or must continue to manage — legacy systems such as thick-client ERPs, legacy general ledger applications, or Linux servers accessed via SSH, a significant challenge exists: these systems were built long before modern APIs and AI technologies, and they offer no native path to connect with today's intelligent automation landscape.

 

This is precisely the gap that ServiceNow Agentic Desktops was designed to close. Rather than waiting for legacy systems to grow a modern API, Agentic Desktops gives AI agents the ability to see, read, and interact with any screen — navigating menus, entering data, and completing multi-step workflows exactly as a human operator would.

 

The outcome is a seamless bridge between the legacy systems an organization depends on today and the intelligent, AI-driven enterprise it is building for tomorrow — no system replacement, no API dependency, and no disruption to the underlying infrastructure that keeps the business running.

 

How does Agentic Desktops connect with a Legacy System?

With Agentic Desktop, ServiceNow AI Agents do not need backend or Agentic integrations to connect with the target system. It connects through the screen – Frontend integration, same way as a human does. It sees the screen, look at whatever is displayed on the screen and understand the context.  The agent clicks, types, select dropdown value and submit forms like humans. Once the task is complete, agent returns the results to ServiceNow.

 

Agentic Desktop is a client application installed on the Windows operating system. It has two workspaces:

 

  1. Design workspace – To build multi-step desktop actions with intelligence and adaptability. Desktop actions are the actions which a human user does on a screen.
  2. Executive workspace – To run desktop actions in an isolated desktop session. This workspace is launched automatically while testing a desktop action or from Now Assist Panel.

Where does Agentic Desktop fits in ServiceNow Agentic Workflows?

With Agentic Desktop Design Workspace application, you can design & enable reusable Desktop actions to automate repetitive tasks mainly desktop applications. While configuring the agentic workflows, the desktop actions can be used as a Tool for AI agents to be used in agentic workflows.

arunpundir_0-1777907038043.png

 

Key Configuration considerations:

  • Desktop Action Type: There are two types of desktop actions – UI Block and Non-UI Blocks
    • UI Block: These actions are basically recording of human actions i.e. UI interactions, such as pressing buttons, typing values in fields, selecting values from dropdown list or more.
    • Non-UI Blocks:  These are prebuilt components for example reading data from Excel, extracting emails from outlook, Invoke Command Powershell, execute SQL query or more. These actions run in the background and there is no need to interact with UI. Following applications are supported for non-UI block actions:
      • Microsoft Excel
      • Microsoft Outlook
      • Microsoft Word
      • PDF
      • Powershell Connector
      • SQL
      • SSH
      • SystemAction

 

  • Execution Mode: Every desktop action can be executed either as Supervised or Autonomous
    • Supervised – Inputs from Live Agent are needed for the execution
    • Autonomous – Executes autonomously without any live agent input
  • Processing Messages for Users: Desktop actions can display two types of messages
    •  In-Progress Message: This displays an in-progress message to end-users while the tool is running.
    • Completion Message: Displays a completion message to end-users.

 

Agentic Desktop vs RPA

Till now the agentic desktop and desktop actions sounds very similar to RPA Bots. So where is the difference? Think of it this way: both a bicycle and a car can get you from A to B. But if there's a detour, heavy traffic, or you need to make a judgment call — one handles it far better than the other.

 

RPA bots work by memorizing exact screen coordinates and sequences. "Click button at position X. Type text in field Y." It follows a rigid, pre-written script every single time — no flexibility, no understanding. While AI Desktop Actions works by understanding what it's looking at. It uses AI vision to recognize "the submit button" regardless of where it appears on screen. It reads error messages and adapts. It can reason.

Feature / Situation

Desktop Actions

Traditional RPA

How it reads the screen

Understands visual context with AI

Memorizes pixel coordinates & selectors

If the UI changes slightly

Adapts - finds element by meaning

Breaks - can't find old coordinates

Handles unexpected pop-ups

Reads and responds intelligently

Crashes or stalls

Processes unstructured data

Yes - reads free text, varied formats

No - needs perfectly formatted input

Decision-making

AI reasoning - next best action

Only pre-programmed if/then rules

Ongoing maintenance

Low - resilient to UI changes

High - breaks often, needs constant fixes

Cost per run (compute)

Higher — uses AI/LLM inference

Lower — simple rule execution

Best for

Complex, decision-heavy, varied tasks

Simple, identical, high-volume tasks

 

Note: Do not run RPA Attended desktop mode and Agentic Desktop execution workspace at the same time.

 

Technical Requirements for Agentic Desktops

Hardware requirements

Resource

Minimum

Recommended

CPU

Intel Processor (Core i5)

Intel Processor (Core i7)

RAM

4-GB RAM

8-GB RAM

Disk space

Minimum 20-GB free disk space after installing the OS, patches, and base software

Minimum 50-GB free disk space after installing the OS, patches, and base software

 

Software requirements

  • Windows 11 operating system is used.
  • A .NET 9.0 runtime v9.0.10 and .NET 9 Desktop Runtime v9.0.10 is installed.
  • No extended monitors are connected.
  • Add the end users who interact with the Execution workspace of Agentic Desktop to the Remote Desktop Users group on the target machine and provide Remote Desktop access permissions for seamless automation execution.
  • If your organization uses Group Policy, add the end users to a Microsoft Active Directory group that is permitted to use Remote Desktop through Group Policy on each target machine where desktop actions run.
    • Local changes to the Remote Desktop Users group are temporary unless they align with Microsoft Active Directory entitlements.
    • If the user is not entitled, Group Policy refresh automatically removes them from the group.
  • Theme must match between the systems used for recording and execution.

Note:

Screen resolution and scaling must be the same between the systems used for recording and execution of desktop actions that are created prior to Agentic Desktop v1.0.1.

 

Disclaimer: "All views expressed in this post are my own and reflect my understanding of the ServiceNow Agentic Desktop."