- Agent = Agent
- Components of Agent
- My Practical AI Agent in Enterprises
- Pain Points of Enterprises
- The AI Agent App I Built
- Project Open Source & Installation Package Download
Hello everyone, I am Engineer Linghu. Today I would like to explain AI agents and the integration of AI agents with enterprises. In this article, I will share my practical projects involving agents in enterprises to help everyone better understand this concept and the relationship between agents and enterprises.
The reason for discussing this topic today is that AI agents are considered the most likely path to achieving AGI.
Agent = Agent
AI agents, simply put, refer to agents. They allow AI agents to understand, make decisions, and implement projects.
When we take the college entrance examination, the entire preparation process over three years is complex, and we can understand AI agents through this process.
During this preparation, we need to understand the following concepts:
-
Persona -
Goals -
Tools
To execute the entire event, we need to know:
-
Persona: Student -
Goal: Get into college -
Tools: Exercise books, school, teachers
Now we need to use AI agents to complete this event. How should we do it?
First, it is important to clarify that in the agent system, the large language model (LLM) corresponds to the human brain. We need to provide this brain with some prompts, which must include persona. We tell the large language model: you are a student and clarify what its goal is, which is to get into college! During the preparation, the agent system needs to use tools.
When we talk about tools, they can be understood as:
-
Internet access -
Numerical computation -
Code execution -
File access -
Database access -
…
Through the above analogy, we find that the agent system is very similar to human thinking, decision-making, and execution. We can break down the agent system into two parts: Planning and Action.
Components of Agent
As mentioned above, we can classify the agent system into:
-
Planning -
Action
This idea comes from human decision-making and execution.
After providing the agent with the appropriate prompts, the agent system will return:
-
Reasoning process -
Breaking down small goals -
The actions to be executed now
In other words, let AI decide what we should do next, how to do it, and in how many steps!
We take the reasoning results from the above agent system as new parameters and import them back into the agent system, allowing it to return new reasoning processes, small goals, and execution actions repeatedly. The agent system can then push back complex matters like a sage, providing step-by-step guidance to execute automatically!
The execution process above will also be saved in the corresponding memory module for continuous generalization. The true quality of this AI agent is determined by its reasoning ability, which is the reasoning capability of the large model.
In my practical process below, I will fully utilize the reasoning ability of the large model to realize AI agent cases. I will choose enterprises as the background for this scenario.
My Practical AI Agent in Enterprises
Alright, now we come to the much-anticipated practical segment.
We need to clarify our goal:
-
Building this AI agent is aimed at enabling those without programming skills to complete programming tasks through AI. [For example, our company does laundry, and we want delivery personnel and laundry colleagues to bypass the R&D department and achieve programming goals themselves, rather than needing programmers to pre-code programs for them.] -
To complete programming using natural language, lowering the barrier to the minimum. Product managers without development experience can directly realize their ideas.
Pain Points of Enterprises
In the early stages of a company, departmental decoupling is a challenge, and unclear division of responsibilities is a common pain point for startups. Since startups need efficiency to ensure business growth, the focus shifts to business, leaving little energy for the management system itself. Everyone has to meet their own targets, leading to communication issues: everyone goes to the R&D department for direct connections and inquiries, which affects the efficiency of the R&D department due to numerous trivial distractions.
For example, a very small company is well-suited for a flat management model because there are few people, so the boss can communicate directly with everyone more efficiently. However, once the number of employees increases, business will consume a lot of the manager’s energy. If departmental decoupling and process standardization are not well executed, efficiency will become a problem.
Is it possible for employees from other departments to solve their issues before coming to the R&D department?
Of course, we can build an AI agent that allows employees from other departments to converse with the large model, which helps them make decisions, provides guidance, and even helps them achieve their goals. If it cannot solve the problem, they can then connect with colleagues in the R&D department.
The AI Agent App I Built
During the holiday, I developed an AI Agent application focused on enterprises, primarily customized for our company’s scenarios, though it can also be applicable to other companies.
My application includes:
-
Server [Large Model Reasoning] -
Client APP
The functions I aim to achieve are as follows:
-
I hope to build an agent that can help me convert natural language into SQL and Java business code. -
The agent will automatically execute SQL or Java programs and return results. -
Ensure communication security using a mobile APP. -
Utilize the reasoning ability of the large model to implement the agent.
When colleagues from other departments, such as customer service, want to find information about a person in the company’s backend management system, they don’t need to go to the R&D department to ask colleagues. Even if they don’t understand technology or MySQL, they can directly complete data queries through the agent.
The above shows the app’s functionality.
In addition to the above functions, there are many unexpected advanced features, which are completed based on the reasoning ability of the large model.
So why do I want to make it an app? Because the app and server still use HTTP transmission, and it can only be used when the terminal connects to my hotspot or the company hotspot.
In fact, the biggest advantage of this agent is that colleagues from other departments do not need to approach R&D colleagues for every little issue to obtain data from the backend management or database. They can directly retrieve information through the agent. For some data calculations and business logic clarifications, the agent can help answer questions and even complete certain actions, such as deleting data from the backend system. At that point, the backend system is no longer solely controlled by programmers in the R&D department; once authorized by the boss, employees throughout the company can directly operate core company data through the agent.
The traditional development model has programmers controlling core company data, achieving product managers’ expected goals through business logic coding, in collaboration with testing and operations teams to deliver product prototypes. Marketing starts marketing, designing logos, and expanding the market. Throughout this process, the steps from programmers controlling core company data to coding business logic to realize products can be directly replaced by AI agents.
Project Open Source & Installation Package Download
Alright, this concludes the introduction of this article. You can click below to experience the product:
“JieKe Intelligent Agent Terminal App Project Source Code”
Project Source Code: https://gitee.com/jackieling/jieke-intelligent-agent