Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DesignGPT: Multi-Agent Collaboration in Design

Shiying Ding, Xinyi Chen, Yan Fang, Wenrui Liu, Yiwu Qiu, Chunlei Chai

TL;DR

The paper addresses the friction in embedding generative AI into product-design workflows by bridging design thinking with machine reasoning. It introduces DesignGPT, a multi-agent collaboration framework that uses SOPs, role-based agents, and a chat-room interface to assist designers during the conceptual stage. Empirical results show DesignGPT improves novelty, completeness, and feasibility of design schemes compared to using standalone image and text tools, with higher consistency across evaluators. This work contributes a concrete method for AI-assisted, human-centered design and offers practical guidance for deploying multi-agent systems in design practice and future research.

Abstract

Generative AI faces many challenges when entering the product design workflow, such as interface usability and interaction patterns. Therefore, based on design thinking and design process, we developed the DesignGPT multi-agent collaboration framework, which uses artificial intelligence agents to simulate the roles of different positions in the design company and allows human designers to collaborate with them in natural language. Experimental results show that compared with separate AI tools, DesignGPT improves the performance of designers, highlighting the potential of applying multi-agent systems that integrate design domain knowledge to product scheme design.

DesignGPT: Multi-Agent Collaboration in Design

TL;DR

The paper addresses the friction in embedding generative AI into product-design workflows by bridging design thinking with machine reasoning. It introduces DesignGPT, a multi-agent collaboration framework that uses SOPs, role-based agents, and a chat-room interface to assist designers during the conceptual stage. Empirical results show DesignGPT improves novelty, completeness, and feasibility of design schemes compared to using standalone image and text tools, with higher consistency across evaluators. This work contributes a concrete method for AI-assisted, human-centered design and offers practical guidance for deploying multi-agent systems in design practice and future research.

Abstract

Generative AI faces many challenges when entering the product design workflow, such as interface usability and interaction patterns. Therefore, based on design thinking and design process, we developed the DesignGPT multi-agent collaboration framework, which uses artificial intelligence agents to simulate the roles of different positions in the design company and allows human designers to collaborate with them in natural language. Experimental results show that compared with separate AI tools, DesignGPT improves the performance of designers, highlighting the potential of applying multi-agent systems that integrate design domain knowledge to product scheme design.
Paper Structure (18 sections, 4 figures)

This paper contains 18 sections, 4 figures.

Table of Contents

  1. Introduction
  2. SYSTEM DESIGN
  3. Core components
  4. Requirements Import Form
  5. Role Definition and Selection
  6. Meeting room
  7. Generative AI settings
  8. Image Generation AI settings
  9. Text Generation AI settings
  10. Composition of roles
  11. USER STUDY
  12. Participants
  13. Procedure
  14. Evaluation indicators
  15. Experimental results
  16. ...and 3 more sections

Figures (4)

  • Figure 1: System schematic. The system defines the job roles and workflow in the design company. Initially, the designer user needs to input design requirements, such as designing three cups for young people in the office. The DesignGPT system initializes multiple employees in the design company, allowing participants to select roles to start meetings. Employees complete the work in the order of design SOP , and finally output the results. The designer user can check the output of the design process in real time, and intervene by typing midway to make Scheme 1 more innovative.
  • Figure 2: The role card encompasses information such as the role's name (e.g., Xiao A), job responsibilities (e.g., Product Manager), as well as skills and behavior logic, which include prompt engineering and API tools.
  • Figure 3: A designer is using DesignGPT system.
  • Figure 4: Mean with SD Plot of the designing results. The plot shows the average scores of the participants' design schemes evaluated by experts in the three dimensions of novelty, completeness and feasibility under the two strategies