Overview and example use cases

Some use cases for Structured Visual Agents include all cases where your desired Agentic Logic cannot be expressed by a simple “ReAct-style” tool-calling loop.

Predefined sequence of steps and branches

Imagine an issue analysis agent. An employee enters the description of an issue. The agent then follows a set of predefined steps:

• Gathering information through multiple knowledge banks and datasets to get identifiers for previous similar incidents

• Identifying possible root causes

• If an obvious root cause is identified, skip to synthesis step

• Else, investigate several possible root causes in parallel, by using several agents, each one searching in different possible data

• Synthesizing all previous steps to generate a final report for the user

Enforcing this logic would be difficult in a Simple Visual Agent.

Gathering “mandatory” information at the beginning of the conversation

For example, an agent responding to customer support enquiries may want to gather all information about the customer, their subscriptions, etc… at the beginning of a conversation.

While a Simple Visual Agent can have these tools at its disposal, it’s hard to guarantee that it would call of them and properly gather all information prior to the run.