AI Agents in Journeys

AI Agents allow you to automate tasks and enrich your workflows easily. Once you create an AI Agent, you can seamlessly integrate it into your workflows as individual actions.

Here are a few example use cases that you can consider using for your company 💡

🔎Application reviewing

Have the agent review application data such as CV's, Cover letters and other data in order to sift out the most qualified applications.

📄Document reviewing

This can be a great time saver when you need to review and validate documents such as:

Personal identification e.g. passports & drivers licenses
Criminal records
Education records
Tax documents

↗️ Categorizing and Routing Employee Complaints

Automatically interpret employee complaints or concerns submitted via an internal portal, categorizing them clearly (e.g., HR, Payroll, Management, IT), summarizing the key points, and routing them directly to the appropriate department for rapid resolution.

🫡Evaluating and Approving Leave Requests

Quickly review and structure leave requests submitted by employees, verify compliance with company leave policies, and automatically determine eligibility or highlight exceptional cases for manual review.

🏡Internal Policy Reference Guide

Quickly and accurately answer employee questions regarding internal policies, compliance guidelines, PTO rules, and company procedures.

🧥Career Development Coach

Provide structured, personalized responses for employees seeking career guidance, internal job postings information, professional development resources, and training recommendations.

👩‍💻IT Support and Asset Management

Assist employees in troubleshooting common technical issues, logging IT tickets, requesting equipment, or verifying asset inventory when onboarding or offboarding.

Step-by-Step Guide to Creating an AI Agent

In our example we are going to create an agent that will specialize in reviewing files and documents in order to determine if they are valid or invalid.

🚨Important information before getting started!

When you create an Agent you are really creating a template that is then usable in your workflows
If you update the base template, for example change the description or add a new file to the knowledge base, it is not updated on current agent actions in your workflows, only new ones that you create
You can individually edit these configurations on a single action in your journey workflow. For example if you want to use the same Agent with a different knowledge base, you will configure that on the agent action itself and it will not impact the original agent template or other agent actions

So let's get started!

First visit the AI Agents section from your main navigation sidebar and click on the "Create Agent" in the top right corner

Then start configuring your agent template

1. Set a Profile picture

Give your agent a recognizable profile picture, you can either select from a pre-defined list of images or upload your own.

2. Set a Name for Your Agent

Give your agent a recognizable name that helps you easily find it later, you decide the name!

Example 1: Mindy - More personable but maybe not very recognizable
Example 2: AI Document Reviewer - Not very personable but very recognizable

In our case we are going for Mindy!

3. Add a Description

Write a short, descriptive sentence explaining what your agent does. This description will appear in your action selection side panel.

Example: Review personal documents from new employees

4. Select an AI Model

Choose the AI model your agent will use to generate responses. Different models offer varying capabilities and performance.

Example: gpt-4o

Note: Currently the agent functionality only supports OpenAI, we will add more LLM options and models in the near future.

The following model types are currently supported in Journeys:

gpt-4o
gpt-4o-mini
o3-mini
o1

To read more about these models please visit the official OpenAI website

5. Set System Instructions

System instructions tell the AI exactly how it should behave or respond, similar to giving specific guidelines or rules. Think of this like giving clear directions to someone about how they should act in a specific role or situation.

Difference between System Instructions and Prompts:

When configuring agents for the first time, it is common to be confused about the purpose of the System instruction vs a prompt.

System Instructions: Set the overall rules or "personality" for the AI, guiding its general behavior.
Prompt: A specific request or question you give the AI each time you use it, like asking someone a question during a conversation.
System Instruction Example: You are a friendly support assistant that answers customer questions about billing clearly and politely.
Prompt Example: "How can I update my billing information?"

System instructions can be very simple or complex, all up to you.

We highly recommend using AI to generate a system instruction. So in our example I asked ChatGPT to generate a system instruction:

6. Temperature

"Temperature" controls the randomness of your AI's output. A lower number (close to 0) means more consistent responses, while a higher number (close to 1) creates more varied answers.

Default: 0.50 (Recommended for balanced responses)
Example adjustment: Set to 0.30 for more consistent, precise answers.

7. Knowledge Files

Knowledge files are documents that you upload for your AI Agent to reference. These files help the AI provide more accurate answers by giving it extra information it can use. Think of knowledge files as giving your AI agent its own personal library to refer to when answering questions.

Allowed file types:

PDF (.pdf)
Word Documents (.doc, .docx)
Plain Text Files (.txt)
Excel Spreadsheets (.xls, .xlsx)

Examples of useful knowledge files:
- Passport_document_rules.pdf
- Company_Policies.docx
- Employee_Handbook.pdf

8. Structured Output

Structured output means that the AI’s responses follow a specific, consistent format, like filling out a simple form. It makes responses easier to process automatically by other tools or software.

Why structured output?

Makes AI answers predictable and easier to handle.
In order to create Journey workflows that uses data from these agent actions, you need to know, beforehand, what the possible results can be

See more about structured outputs from OpenAI's documentation.

If you've chosen structured output, specify exactly what data your agent should generate. Clearly define each piece of information the agent needs to output, similar to filling out fields in a form.

Example 1: If your agent is designed to review if a document is valid the this would be one way to do it

You can now use the "Is the file valid?" when mapping from this action when creating your workflow.

Example 2: If your agent is designed to answer a question then this is how you could configure your structured output
Answer

Reason

Using AI Agents in Journey workflows

Once your AI Agent template is created, you'll see it available in your action selection panel whenever you're building a workflow. Just select your agent and add it wherever you need it.

After adding your agent, you need to tell it specifically what you want it to do in that part of the workflow.

Think of this step just like chatting with an AI (like ChatGPT): the agent already knows its general task, and you're now simply asking it to do something specific for you.

1. Add the Prompt

The prompt is your specific question or request, just like when you ask ChatGPT a question.

Example Prompt: "Can you check this passport document for errors?"

2. Add Additional Ffiles (optional)

Sometimes your workflow collects various documents along the way. Here, you can choose which documents you want your AI agent to review at a specific point.

Example: At one step, you may have the AI agent review only passports. At another step later on, you might have it check criminal records.

This lets you control exactly how and when your AI agent uses different files.

3. Advanced settings

If needed, you can adjust settings like system instructions, the AI model, or temperature for one specific action only. This won't affect your original agent template.

Testing

Always test your AI agent action to make sure it's working correctly. Open the agent action configuration side panel and click "Test" in the bottom-right corner to try it out before using it.