I’m currently learning the process of building an Agentic Automation, and I wanted to know if it’s possible to keep the system prompt and user prompt dynamic?
An example:
A user sends an email containing the system prompt
That system prompt must be updated in the Agent Builder
An explanation would suffice, don’t need a workflow.
Yes, it is absolutely possible to make the system prompt and user prompt dynamic in an Agentic Automation setup — including one built with UiPath’s Agent Builder, OpenAI APIs, or similar tools.
An agent typically interacts with a large language model (LLM) like GPT using prompts that include:
System prompt: Defines the agent’s personality, role, and behavior.
User prompt: The actual request or instruction from the user.
To make both dynamic, you need a way to:
Receive new prompts from the user (email, form, UI, etc.).
Store and manage the system prompt state.
Inject those prompts into your LLM interaction at runtime.
After you’ve published and deployed your agent, You can use the Run Agent Activity in your process. Using this you can Invoke your Agent in the process:
The Prompts extracted from the “Extract Prompts from Email” task will become the Out-Property of that Task.
Which will become the In-Property for the Agent.