01 | Introduction

In the previous article, we discussed in detail what an Agent is. Here, we will review it again.

An Agent is a way to encapsulate a large model to simplify user interaction. It understands the user’s intent based on their input, selects the appropriate Tool through the action field, and uses the action_input as the parameter for the Tool to handle the user’s request.

Jelly Pudding Rabbit, WeChat Official Account: Captain Lu’s GPT Introduction (11) Understand LangChain Agent in Two Minutes

Since the Agent is so important, when is it appropriate to use one?

We will analyze this based on the official website and the specific capabilities of the Agent.

02 | Introduction to Agent Application Scenarios

As usual, we will look for answers from the official website.

Some applications will require not just a predetermined chain of calls to LLMs/other tools, but potentially an unknown chain that depends on the user’s input. In these types of chains, there is an “agent” which has access to a suite of tools.

Depending on the user input, the agent can then decide which, if any, of these tools to call.

The literal translation is: Some applications not only need to execute a predetermined chain of calls to invoke LLMs or a set of Tools but may also need to call an unknown chain based on the user’s input. In this unknown calling chain, an Agent is used to maintain access to a series of tools. The Agent identifies the user’s intent based on their input to determine which Tool or series of Tools to use to execute the task.

Combining the application suggestions from the official website and the definition in the introduction, we can analyze as follows:

The Agent can identify the user’s intent to decide which Tool or Tools to use to execute tasks, even executing some chain calls that we have not pre-set.

The first statement is relatively easy to understand because the Agent itself can identify user intent, but the second statement gives us some usage scenarios, specifically when we are unclear about the user’s intent, the Agent decides which tools to use to meet the user’s needs.

03 | Agent Application Scenarios

With this in mind, we can generally understand the scenarios where the Agent is suitable for use:

The first type: Consumer applications.

Consumer applications provide users with an intelligent interaction experience. People may be accustomed to a tedious onboarding process when using new software, such as the typical onboarding flow shown in the image below (image source: internet, no specific direction), where users need to keep clicking “I understand” to start using it, but in reality, we don’t know what we have understood.

Another issue arises when we need to find some less commonly used features; we have to search through “more applications” or directly use the search box. Sometimes we don’t even know the application name, and just searching for the function may not match the application. Of course, Alipay does not have this problem.

However, if we use GPT capabilities, it seems we can easily solve this problem. Users just need to voice input “I want to take a taxi to the Forbidden City,” and the corresponding software can open directly and position the destination at the main gate of the Forbidden City without needing step-by-step guidance or an exhaustive search, greatly enhancing user experience.

The second application scenario: Intelligent workflow

Many students or young professionals should have a common pain point: that the OA system in companies or schools is really too cumbersome and difficult to use. A small request may take one or two hours to resolve, sometimes requiring various inquiries with others, and the key is that we may not even find the answer. This is when we genuinely need an intelligent workflow.

For these formatted scenarios, we don’t need to complicate matters. Users just need to fill in requests and requirements based on relevant fixed prompts, and then the Agent can directly invoke the pre-set Tool calling chain to execute the business process, providing complete controllability, reliability, and usability.

04 | Conclusion

This article mainly starts from the definition of the Agent and the recommended application scenarios from the official website, introducing what I consider to be two recommended usage scenarios: consumer applications and complex workflows. If readers have other suitable scenarios, please feel free to leave a message. Thank you.

​​​