AI Agents are powerful but more expensive and potentially less unreliable (they introduce uncertainty in your workflow).
When possible, we recommend using regular actions and conditions instead of AI Agents.
AI Agents are typically used to automate conversations since you can never be entirely sure what the user will say next.
Just like actions, AI Agents have context for the workflow steps that came prior to the AI Agent.
You can reference this context in the prompt of the AI Agent.
AI Agents are Comprised of Five Parts: Prompt: The Prompt is used to instruct the AI Agent on what task to perform and how to execute it. Think of the Prompt as a guide for the AI Agent to follow in order to perform the task within your workflow. Model: The language model used by the AI Agent. Ask for Confirmation: Lets you bring a human in the loop and confirm actions before Lindy executes them. Skills: Actions that the AI agent can perform at any point to execute the task. For example, if you want the AI Agent to first search the web for a recent news article, and then send you a Slack DM, you will give it two actions: Search the Internet and Send a Slack Message. Jump Condition: Jump Conditions tell the agent when to continue to the following steps in the workflow. An AI Agent will keep “looping” and not move to the next action until a Jump Condition is fulfilled. Examples of Jump Conditions include: some textOnce you’ve found time with the emailer Once you’ve agreed on a price with the user Once you’ve completed the web searches Example AI Agent: This Lindy enters an AI Agent step once it has updated the Google Sheet Row.
The AI Agent will then search for the contact that initiated the trigger to find at least five interesting facts about them and send the user a Slack Direct Message with the details and links.
Once the AI Agent has done that, it will follow the “you sent a Slack DM” jump condition and proceed to the next step, updating the contact in Hubspot.
Using Multiple Jump Conditions: You can use more than one jump condition to tell Lindy when to leave the AI Agent and move onto proceeding steps.
In the example below, two jump conditions are used in case the AI Agent fails at finding five interesting facts about the contact. In this case, Lindy will exit the AI Agent and proceed to the next step of searching the internet.