From the course: Agentic AI Human-Agent Collaboration Design Patterns
Join today to access over 25,500 courses taught by industry experts.
Correction and refinement loop
From the course: Agentic AI Human-Agent Collaboration Design Patterns
Correction and refinement loop
Even with the most sophisticated training data and prompt engineering, it is generally inevitable that agents will eventually make mistakes. For example, they might misinterpret a vague instruction or misread a complex document. These mistakes can easily go undetected, leading to the agent confidently proceeding with incorrect information or carrying out incorrect actions. Let's have a look at a sample scenario where an autonomous agent is helping an accounts payable team by extracting data from incoming digital invoice documents. It basically performs the same action repeatedly, reading each invoice and entering the data it extracts into an internal system. It's been trained to do this using a template that vendors comply with. But imagine that there is a new vendor that inadvertently sends a non-standard invoice where the layout is different. The agent correctly identifies the total amount but it mistakenly identifies the shipping address as the billing address because of where it…
Download courses and learn on the go
Watch courses on your mobile device without an internet connection. Download courses using your iOS or Android LinkedIn Learning app.