How Multi-Agent Systems Mirror Team Collaboration: AI Agents Explained

Note: This article is for Product Managers, Product Designers, and Engineers who are building digital products, especially in the field of AI, and whose goal is to build a Multi-AI Agent System to optimize value for their products and businesses, while solving complex user problems through AI Conversation.

You might see that building an AI Conversation seems simple when looking at the interface, which includes a two-way interaction between the AI and the user, and an input placeholder for the user to ask questions. But behind that is a huge challenge for the Product Development team in general. It takes many steps to discover user expectations, improve, and iterate continuously to deliver value more clearly each day.

A few years ago (when the AI wave had not yet arrived), when we approached many chatbots, most of the scenarios were pre-written according to a fixed structure, and users were limited to those scenarios. Of course, those AI Chatbots were created to solve simple problems such as:

1. Answering frequently asked questions related to the business
2. Consulting on cosmetics based on the catalog and skin problems
3. Consulting on fashion based on height, weight, and product type
4. ….

So how do the currently popular chatbots like ChatGPT, Google Gemini, Claude, Microsoft Copilot, etc., manage to remember the user’s context and be so flexible that they can answer all questions and needs of the user? That will be the topic that I will explain in the easiest way in this time.

1. What is a Multi Agent System

Before understanding Multi-Agent, we can understand that this concept includes many different agents combined, right? So what is an AI Agent? Simply put, an AI Agent is a system designed to solve a specific problem, with a specific role, and a specific task. So a Multi-Agent is the sum of different AI Agents solving different problems combined to solve a larger problem. That’s a simple explanation for now, we will delve deeper into the correlation between Multi-AI Agent and Team Collaboration to understand and approach building Multi-Agent more easily and without being overwhelmed.

2. How Multi-Agent Systems Mirror Team Collaboration

Let’s start with something related to our work: your team collaboration at your company. Let’s use this example to compare it with how multi-agent systems work to identify the commonalities.

Now assume there’s:

• (1) Multi-Agent System helping you plan for the event and a. Goal: Plan the entire event
• (2) A team in the company is collaborating to build a health tracking feature in an app. Goal: Build a health tracking feature.

1. Agent 1: - Role-playing: Event Goals Planner. - Backstory: Designed based on event organization experts with the ability to gather and process information quickly. It “likes” helping users achieve clarity about the goals of the event.
2. Member 1: - Role: Product Manager - Backstory: With a passion for technology and health, the PM always wants to build products that improve lives. Having struggled with tracking their own health, they aim to create an easy-to-use feature for everyone.

1. Agent 2: - Role: Venue and Logistics Expert - Backstory: Agent 2 is built from data models about venues and logistics, with the ability to process information about space, equipment, personnel, and organizational conditions.
2. Member 2: - Role: Product Designer - Backstory: Has strong expertise in user interface and user experience design, especially in mobile applications. They understand the importance of designing interfaces that are not only beautiful but also easy to use and convenient.

1. Agent 3: - Role playing: Budget Manager. - Backstory: With the ability to analyze financial data, it can optimize spending and offer cost-saving options without reducing the quality of the event.
2. Member 3: - Role: Backend Engineer - Backstory: The Backend Engineer has expertise in building database systems and APIs. With the ability to process real-time data, they specialize in optimizing data storage and retrieval solutions.

1. Agent 4: - Role playing: Task Scheduler. - Backstory: Helps users organize tasks logically, from preparation to implementation, ensuring that no steps are missed.
2. Member 4: - Role: Frontend Engineer. - Backstory: Specializes in developing user interfaces and optimizing the user experience. They are concerned with how health data can be displayed smoothly and understandably.
3. Member 5: - Role: QA Tester - Backstory: QA Testers specialize in software testing with the ability to detect errors, ensuring data accuracy and product stability in all situationsg.

So, what are the common points between AI Agents when they work together (called a Multi-Agent) and team collaboration? They have common elements such as:

1. Clear Goals

The agents have their own tasks, communicating via API.

Team members have clear goals tied to their roles.

2. Deep Specialization and Distinct Roles

Automatic and fast, each agent performs its programmed tasks.

Each member has their own strengths, receives work, and delivers with the best outputs.

3. Interaction and Communication

Agents interact with each other to produce the final result. For example, Agent A can communicate with Agent B to receive information about the location, or with Agent C to calculate costs.

Team members often communicate and coordinate with each other, for example: Designers and Frontend Engineers work together to ensure the user interface is well integrated with the backend system.

4. Flexibility

Agents need to be able to adapt to changes in user requirements, such as changing locations, changing budgets, or changing event goals.

Team members also need to be flexible and adapt to changes in requirements or changes in product strategy.

5. Coordination and Synchronization

Agents must coordinate and synchronize work between the parties to produce a complete event plan, from choosing a location to budgeting.

Team members also need to align with each other on product direction and initial goals to deliver features that can grow the business and solve user problems.

-> There are many other similarities, but in general, the correlation between the two is the clearest example to help everyone understand how Agents operate to optimize products better.

Based on these characteristics, it will be very easy to find the answer to the next question:How to optimize a Multi-Agent System? and I hope that next time, I can share best practices for improving Multi-Agent systems as a Product Designer and Product Manager

Found this insightful?

If you have any contributions, feel free to share your thoughts with me. I’d love to hear them.

If you enjoyed this post, hit the like button and follow me on Medium & Linkedin for more updates — I’d love to connect and chat! 💡