An Agentic Processes represents a sequence of tasks that must be executed to achieve a specific result. These tasks may involve interactions with users, AI Agents and external tools.
Agentic Processes are organized as a logical flow that can be initiated by an end user action or by the availability of a previously generated artifact.
This is a step-by-step guide to create, test, and publish an Agentic Process that connects two AI Agents: the first one applies the Globant writing style to a user-provided text and converts it into Markdown format; the second one takes that Markdown content and translates it into the target language selected by the user.
First, log in to the Globant Enterprise AI Backoffice. In the Project Dynamic combo box, select the project you want to work with; in this case, Default(Docum Team) is used.
Next, on the left side of the screen, you will find the Backoffice menu. In this menu, click on The Lab.
By clicking on The Lab in the Globant Enterprise AI Backoffice, a new window opens in the browser with Globant Enterprise AI Lab.
Once inside Globant Enterprise AI Lab, you will see the Side Navigation Menu on the left edge of the screen. If the menu is collapsed, click on the arrow icon to expand it.
In this menu, select the Agentic Processes option to access the Agentic Processes Dashboard.
If no Agentic Processes have been created yet, a message saying "No Agentic Processes Created" will appear. Below that message, click on the New Processes button to start creating a new process.
After clicking on the New Process button, a pop-up window titled Process Details will appear on the screen. Here you must complete the following fields:
-
Process Name: Enter a unique and descriptive name for the process.
-
Process Description: Optionally, include a brief explanation of what the process is intended to do.
Once the information is complete, click on Save to continue.
After saving, a new tab with the name of your process opens automatically in the Agentic Processes dashboard. This tab shows a grid where you can design the flow of your process. By default, it includes a Start node at the center of the screen. This node represents how the interaction begins.
Click on Start to open the Properties panel on the right side of the screen.
In the Signal name field, you can define a label for the trigger, and in the Type dropdown, you must select one of the following options:
- User: The process starts based on a user-initiated action.
- Artifact: The process is triggered when a specific artifact is created.
If you select Artifact, an additional Inputs section appears, where you must define a variable representing the artifact that should trigger the process.
An artifact is the (intermediate or final) result produced by a task within the process. It can be, for example, a generated document, a response produced by an Agent, or any other type of structured content that represents a deliverable within the process flow.
Click on Add+ and select the appropriate artifact from the dropdown menu. This instructs the process to begin only after that specific artifact is available.
In this case, the Signal name is set to UserInput, and the Type selected is Artifact. In the Inputs section, Text is added.
This configuration is used because the Subject field that end users can fill when interacting with the Agentic Process is very limited in length. Therefore, if the user wants to apply the Globant writing style to a longer piece of content, that content must be uploaded as a file artifact.
To add a task node, move your cursor over the Start block. When you do this, you can see four circular anchor points around it (top, right, bottom, left). Hovering near any of these points displays the + icon.
Click on the + icon that appears when hovering over the anchor point of your choice. When you do this, a context menu appears with three options: Task, Event, and Finish.
Click on Task to add a new task node to the process. This task represents the first action to be executed once the process is triggered.
Once the task is added, the node appears and automatically opens the Properties panel on the right side of the screen, which allows you to configure the details of the task.
In addition to creating new tasks by clicking on the + icon next to a node, you can also add a task by dragging it directly from the Toolbox.
To do this, go to the Toolbox (located in the bottom left corner) and click on Task. Then, drag the new task node to the desired position on the canvas. Once placed, connect the task to the rest of the flow by clicking on the + icon from the node you want it to connect to. A dotted line will appear, indicating the connection. Once confirmed, the flow will integrate this task into the process.
A task in an Agentic Process represents a unit of action that is executed as part of the overall flow. It may involve different activities, such as invoking an AI Agent, requesting information from a user, executing an external tool, processing data, or generating an artifact as an output.
Each task can have inputs (for example, artifacts generated by previous tasks or user data) and outputs (new artifacts or actions that trigger subsequent tasks).
To configure the task, fill in the available fields in the Properties panel:
- Agent
- Task name
- Task description
- Task context
- Task instructions
- Examples
- Inputs
- Outputs
At the top of the Properties panel, within the Agent block, click on the Select link. This action opens a pop-up window listing all the Agents previously defined in AI Lab for the current project.
In this window, you can use the search field to filter Agents by name or description. When you select one of them, its detailed information is displayed on the right side, including:
- Agent name and description
- AI model used
- Reasoning strategy configured
- Tools
Once you've confirmed that this is the Agent you want to use, click on the Select button. The Agent is now assigned to the task, and its name will appear in the Properties panel instead of “Select”.
This task uses the Documentation Assistant agent to apply Globant’s writing rules and convert the text into Markdown format.
In the Task section, you must configure the following fields. They allow you to define precisely what the task should do and are combined with the Agent's capabilities to execute the corresponding action.
-
Task name: Specify a clear name that identifies the task.
In this case, since the goal is to review the article content, an appropriate name could be: ConvertArticleToMarkdown
-
Task description: Provide a brief description of what the task does. For example:
Convert a documentation article to Markdown format, applying correct structure, headings, and style conventions.
-
Task context: Include relevant information that gives context to the task, such as details of the overall process or considerations that the Agent must take into account.
For instance: This task is part of a documentation workflow in which generated articles must be transformed into Markdown to ensure consistency across publications. The Agent should format the content according to standard Markdown rules, maintaining clarity and proper use of headings, lists, and emphasis.
-
Task instructions: Specify clear instructions that guide the Agent in the execution of the task. These instructions are part of the prompt that will be sent to the Agent.
For example: Take the provided article and convert it into a well-structured Markdown document. Use appropriate Markdown syntax for headings, bold, italics, bullet points, and code blocks where needed. Make sure the content remains clear, professional, and aligned with documentation standards.
If you click on the expand icon, a larger editing window will open to make typing easier.
Examples allow you to define input/output pairs that help the Agent understand what kind of result is expected. This is especially useful in tasks involving content transformation, such as converting text to Markdown.
To add an example, click on the + Add option under the Examples section. Two fields will appear:
Although this section is optional, it is recommended to include at least one example when the output must follow a specific structure or formatting style.
In this case, since the task consists of converting an article to Markdown, a useful example would be:
-
Example input:
This is the introduction. It explains how the system works and what the key components are. Features include: easy setup, automatic updates, and compatibility with all major platforms.
-
Example output:
# Introduction
This is the introduction. It explains how the system works and what the key components are.
## Features
- Easy setup
- Automatic updates
- Compatibility with all major platforms
Although it is not mandatory in this case, you can indicate an artifact that works as input if this task must work with a previously generated content.
Insert Text as the input artifact. This artifact represents the content provided by the user when triggering the Agentic Process. Since this is the initial material to be processed, it must be configured as the input of the task.
This field is mandatory to ensure that the task generates an output that can be used by the following tasks in the process.
In this case, since the task involves validating a documentation article, a suitable output could be an artifact of type MarkdownArticle—or similar—containing the corrected and validated content.
Once you have configured all the necessary fields, the task node will be displayed as follows:
To continue building the process, add a second task node following the same procedure described earlier.
This task uses the Expert Translator agent to translate the article— already converted to Markdown—into the target language specified by the user.
- Agent: Click on the Select link in the Properties panel and choose the ExpertTranslator agent from the list.
- Task name: TranslateMarkdownArticle
- Task description: Translate a Markdown article into the target language provided by the user.
- Task context: This task is executed after the article has been reformatted to Markdown. Its purpose is to make the content accessible in the language selected by the end user, while maintaining structure and readability.
- Task instructions: Translate the provided Markdown content into the target language {{language}}. Make sure to preserve the original formatting, structure, and headings used in the Markdown document. The output should remain clear, professional, and easy to read.
Note: The {{language}} syntax refers to a process variable. You must define this variable by clicking on the process name (in this case, Documentation Assistant) at the top of the screen, selecting "Edit process variables", and adding a new variable named language of String type.
- Input: Select MarkdownArticle, which is the output of the previous task.
- Output: Define an artifact named TranslatedMarkdownArticle.
Once you have configured the tasks, you can continue building the flow by adding a new task or event.
To add a new task, follow the same procedure explained in the previous step: go to one of the connection points of the current node (for example, the one at the bottom), click on the + icon and select Task.
If you want to add an event instead, click on the same + icon and select the Event option from the context menu. This action will insert an event type node that can be used to handle intermediate situations within the process.
After selecting Event, a new node is inserted and automatically connected to the task. The right panel (Properties) opens automatically, allowing you to define the event that will trigger the next action.
In this panel, complete the following fields:
- Signal name: Enter a descriptive name for the event signal (e.g., MarkdownReady).
- Type: In this field, a dropdown menu will appear with two options: User and Artifact. In this case, you must select User because the event will be triggered when a specific artifact is generated, acting as a continuation signal within the process.
- Inputs: When you select Artifact, the Inputs section is enabled automatically. Here, you must define which artifact should be available for the event to be triggered. This behavior is similar to how the Start node works when configured with the Artifact type.
The event node acts as an intermediate pause in the process and can be triggered either by the generation of a specific artifact or by a user action. In this case, the Type is set to User, which means the process will wait until the end user confirms that they want to proceed with the next step.
To complete the flow, add a Finish node after the event by clicking on the + icon and selecting Finish, just as you did for the previous elements.
Once the flow is complete, you can test how the process behaves from an end-user perspective.
Click on the Run Test button at the bottom right of the screen. This opens the testing panel, where you can simulate the interaction with the Agentic Process.
In the testing panel:
- Fill in the Subject field with a brief title or description.
- Upload a document in the Artifact (Text) field. This document will be used as the input for applying the Globant writing style and converting it to Markdown.
- Enter the Language you want the final output translated into (e.g., English, Portuguese, etc.).
Once all fields are completed, click on Run to execute the process.
After clicking on Run, the Agentic Process begins execution. It processes the input using the first task and pauses when it reaches the event node.
At this point, the process waits for user confirmation to proceed. As shown in the image, a MarkdownReady section appears with a Send button next to it.
To continue and complete the process, the end user must click on the Send button. This confirms that the Markdown version is ready and triggers the next step in the workflow, enabling the translation task to run and allowing the process to reach the Finish node.
Once the entire Agentic Process has been configured and tested, the final step is to publish it. To do this, click on the Publish button located in the bottom-right corner of the screen. This action makes the process available in the Playground, where it can be accessed and used like any other assistant.
To delete a node you must click on it. You will know that the node is selected when you see the circles around it. Then press the Delete key on your keyboard to remove it from the flow.
Since April 2025 release.