Building a Self-Service AI Agent With the Power of ASK_AI in Automic Automation V24.4

In a previous blog post, we discussed the concept of building an AI agent using Automic Automation. This post will go into the details of doing this with ASK_AI in Automic.

With the release of V24.4, Automic Automation delivers powerful new AI features like ASK_AI. These features enable you to add more intelligence to your automation use cases. In parallel, the idea of AI agents is increasingly moving into the spotlight.

So what happens if we combine those two things?

Agentic AI?

Agentic AI is a type of artificial intelligence that operates autonomously, making decisions and taking actions to achieve specific goals with minimal human oversight. At its core, agentic AI functions through three key processes:

1. Perception. AI collects and interprets data from its environment through sensors, APIs, or other input sources. This allows it to understand its surroundings, detect changes, and recognize patterns.
2. Cognition. Once data is gathered, AI processes it, applies reasoning, and makes decisions. This stage involves planning, learning from past experiences, predicting outcomes, and adapting to new information.
3. Action. Based on its cognitive processing, AI takes meaningful steps to execute tasks. This could involve controlling a robot’s movements, automating workflows, or generating responses in a conversational AI system.

By integrating these three capabilities, agentic AI can operate dynamically in real-world or digital environments, enabling advanced applications in automation, robotics, and decision-making systems. Unlike traditional AI, which typically responds to direct inputs, agentic AI can independently assess situations and take proactive steps toward achieving its defined objectives.

How does that translate to Automic Automation?

Perception:

For over 20 years, Automic has been a leader in event-driven process automation, which means that automation can be dynamically initiated in response to specific conditions.

Action:

With its powerful workflow engine and execution capabilities across platforms, including cloud, on-premises, and hybrid environments, Automic Automation acts as its core functionality.

Cognition:

The cognition part of an AI agent has been missing until now, as workflows had to be directly linked to their triggers, and the input data had to be defined in a certain format. Every decision had to be made at definition time, rather than execution time.

Why is ASK_AI in V24.4 a game-changer?

With the introduction of ASK_AI and its advanced capabilities, AI can be leveraged to gain a deeper understanding of events. ASK_AI adds more intelligence to the decision-making process, ultimately enabling the execution of a workflow.

The following scenario describes a real-life use case of triggering workflows based on triggers, such as emails or tickets created in Rally. The system already contains a number of workflows that can resolve common issues that a user might encounter, such as unlocking a user account, changing passwords, and so on. These workflows should include a brief description as part of the documentation sheet, and all parameters must be included in prompt objects with tooltips that describe the parameter.

Perception

The perception part of the example is essentially to gather all the necessary information about the event, as well as the available options.

The first part of the information we need is the list of available workflows and their details. In this example, all workflows that should be offered to the user are marked with the keyword “@SelfService” as part of the object title. With the “search” endpoint of the REST API, we first read all the object names that have this keyword.

Once we have the object names, we use the “objects” endpoint to retrieve the documentation sheet for the description, as well as all the prompt parameters, including input parameter names and tooltips. Because structured input works better for large language models (LLMs), we create a JSON structure with that information and store all of the workflows in a static VARA object.

Example:

{
  "ITADMIN.WORKFLOW.UNLOCK.USER": {
"description": "This workflow is used to unlock a user. The first task checks, if the user is locked and if so, the second task unlocks the user.",
        "input_variables": {
        "EMAIL#": {
          "description": "Email address of the user to unlock"
        }
     }
  }
}

The second part of the perception stage requires an event to trigger the agent. In this case, we use the Email Integration to check for incoming emails.  If a new email is found, its content is read and forwarded to the analyzer task.

Cognition

The cognition part of the AI agent is where it all comes together and where the magic happens.

First, the analyzer task has to be grounded and made aware of all the options. For this step, we use the information gathered about all workflows and incorporate it into the model's context.

Now that the LLM is aware of all the options, we can simply submit an inquiry to ASK_AI to determine which workflow could solve the issue highlighted in the email.

Email:

ASK_AI:

If an object name is returned, ASK_AI can also handle all the necessary details required to execute the workflow, as it is aware of the required input parameters. Since we already know the format, we need to start the workflow via the Automic REST API. We use ASK_AI to not only get the information out of the email but also instruct the LLM to create the JSON we need as input for the EXECUTION endpoint of the REST API.

With that, we identified which workflow to execute and how to execute it.

Action

Based on the analyzer's output, the workflow's execution is quite straightforward. We use a REST job that utilizes the full &JSON# returned by ASK_AI and calls the execution endpoint.

To enhance the user experience, we also utilize ASK_AI to generate a response for the customer. Since we receive the task’s Run-ID as a return value from the execution, we can also include it in the response.

Response:

Conclusion

This example demonstrates that Automic Automation v24.4, with its AI-powered features, enables the integration of proven technologies, such as event-based scheduling or workflow execution, with emerging technologies like AI-based decision-making. The solution enables you to bring more intelligence to every stage of your automation, whether it involves input parsing, decision-making, execution, or response generation.

To learn more, read the blog post on Automic 24.4’s powerful GenAI features or watch the Automic GenAI overview video.