8-Minute Guide: Build an AI Team with Coze 3.0 – Complete a Project from Idea to Launch as a One-Person Company

Introduction

This tutorial shows you how to build a fully functional AI agent team to complete an entire project from idea to launch using Coze 3.0. The example project is an anti-procrastination app called Time Blocks, designed to break large tasks into small, actionable 10–30 minute chunks.

What You'll Learn

• How to build an AI agent team in Coze 3.0
• How to assign roles like product manager, designer, developer, and video producer to different AI agents
• How to use skill packs to give agents professional capabilities
• How to run a full project workflow: requirement definition → design → development → video production
• How to manage and iterate on your project using AI collaboration

Prerequisites

• A Coze 3.0 account
• Basic understanding of AI agents and workflow automation
• No coding experience required (though familiarity with tools like Claude Code or Codex CLI is helpful)

Step 1: Create a New Project in Coze

1. Open Coze 3.0 and click the + button in the top-left corner
2. Select New Project
3. Enter the project name: Time Blocks
4. Click Create Project

Step 2: Build Your AI Agent Team

Create a Product Manager Agent

1. Click + → New Agent → Cloud Agent
2. Select Coze Agent as the base model
3. Name it Product Manager and click Create and Deploy
4. In agent settings, select a model (e.g., GPT-4o, Claude 3)
5. Go to the Skill Store and add the Product Operations Skill Pack — includes product strategy, user research, PRD generation, and user flow design

Create a Designer Agent

1. Click + → New Agent → Cloud Agent, name it Designer
2. Add the UI/UX Design Skill Pack — includes wireframing, visual design, design system creation, and handoff documentation

Create Developer Agents

You'll create two developer agents:

• Development Planner: Uses Claude Code to create a detailed development plan. Create as Local Agent with Claude Code.
• Development Executor: Uses Codex CLI to write the actual code. Create as Local Agent with Codex CLI.

Create a Video Producer Agent

1. Create as Cloud Agent, name it Video Producer
2. Add the Video Production Skill Pack — includes script writing, storyboarding, video generation using Seedance, and editing

Step 3: Add Agents to Your Project

1. Go to your Time Blocks project
2. Click Project Settings → Agent Management → Add
3. Select all agents: Product Manager, Designer, Development Planner, Development Executor, Video Producer
4. Click Save

Step 4: Run the Project Workflow

4.1 Define Requirements with the Product Manager

Tag the Product Manager agent and send:

@ProductManager
I want to build an anti-procrastination app called "Time Blocks". The app should help users break large, overwhelming tasks into small, 10-30 minute time blocks.

Please:
1. Define the product positioning
2. Identify target users and their pain points
3. Outline core features
4. Create a user flow
5. Write a PRD for the designer and developer agents

Use the Product Operations Skill Pack to help you.

4.2 Design the UI with the Designer

Tag the Designer agent and send:

@Designer
Based on the PRD from the Product Manager, design the UI for the Time Blocks app.

Requirements:
- Single-page layout
- Visual style: clean, minimal, with a focus on time blocks
- Include a time board where users can drag and drop time blocks
- Provide handoff documentation for the developer agents

Use the UI/UX Design Skill Pack to help you.

4.3 Create a Development Plan

Tag the Development Planner agent and send:

@DevelopmentPlanner
Based on the PRD and design documents, create a detailed development plan for the Time Blocks app.

Include:
- Technology stack
- File structure
- Development phases
- Implementation details for each feature
- Timeline estimates

Use Claude Code to help you.

4.4 Write Code with the Development Executor

Tag the Development Executor agent and send:

@DevelopmentExecutor
Based on the development plan, write the code for the Time Blocks app.

Requirements:
- Use HTML, CSS, and JavaScript
- Implement the time block drag-and-drop functionality
- Include a timer for each time block
- Make the app responsive

Use Codex CLI to help you.

4.5 Generate a Promotional Video

Tag the Video Producer agent and send:

@VideoProducer
Create a 60-second promotional video for the Time Blocks app.

Requirements:
- Script: Explain how the app helps users overcome procrastination
- Visuals: Show the app in action, with time blocks being dragged and dropped
- Music: Upbeat and motivating
- Style: Clean and modern

Use the Video Production Skill Pack and Seedance to generate the video.

Step 5: Test and Iterate

1. Test the app to make sure it works correctly
2. If you find issues, tag the relevant agent and ask them to fix it:

@DevelopmentExecutor
The time block drag-and-drop functionality is not working correctly. Please fix it.

Step 6: Deploy the App

1. Once the app is working correctly, deploy it to a hosting service like Vercel or Netlify
2. Share the promotional video on social media to promote the app

Conclusion

Congratulations! You've just built a complete AI agent team and used it to create a full project from idea to launch. This workflow can be applied to any type of project, from app development to content creation to marketing. Coze 3.0 makes it easy to build and manage AI agent teams, allowing you to focus on the big picture while the AI agents handle the details.

Note: Some features mentioned in this tutorial, like multiple local agent access and faster video generation, require the Coze 3.0 Pro plan. The free plan includes basic agent collaboration and standard video generation.

常见问题

What's the difference between Cloud Agents and Local Agents in Coze?

Cloud Agents run on Coze's servers — they use Coze-hosted models (GPT-4o, Claude 3, etc.) and the Skill Store. They're ideal for knowledge work: product management, design, content creation. Local Agents run on your machine using tools you've installed — Claude Code and Codex CLI in this tutorial. They have direct file system access and can execute code, making them essential for development work. The article uses Cloud Agents for PM, Designer, and Video Producer (knowledge roles), and Local Agents for Development Planner and Executor (execution roles). This hybrid approach gives you the best of both: cloud convenience for thinking tasks, local power for building tasks.

Can I use this workflow for projects other than app development?

Yes — the agent team pattern (PM → Designer → Developer → Video Producer) is project-type agnostic. For a content marketing project: PM defines the content strategy, Designer creates brand assets, Developer builds landing pages or email templates, Video Producer creates promotional content. For a product launch: PM writes the go-to-market plan, Designer creates pitch decks, Developer builds the product page, Video Producer makes a demo video. The key insight: you're not limited to software projects. Any creative or business project that benefits from multiple specialized roles can use this workflow. Just swap the Skill Packs to match your domain.

How much does Coze 3.0 cost for this workflow?

The free plan supports basic agent collaboration and standard video generation — enough to complete the tutorial project. The Pro plan (pricing varies by region) unlocks multiple local agent access, faster video generation, priority model access, and advanced Skill Packs. For occasional projects, the free tier is sufficient. If you're running a one-person company with daily agent usage, the Pro plan's faster generation and multi-agent concurrency become worth the cost. The models themselves (GPT-4o, Claude) are billed through your Coze account or API keys — expect $20-50/month in model costs for regular usage in addition to the Coze plan.

How is Coze 3.0 different from just using Claude Code or Codex for everything?

Claude Code and Codex are single-agent tools — one AI, one conversation. Coze 3.0 is a multi-agent orchestration platform. In Claude Code, you'd ask one assistant to be PM, then designer, then developer — switching roles manually and managing context across conversations. In Coze, you have five persistent agents, each with specialized Skill Packs and separate memory. They work on their parts simultaneously (Designer can work on UI while Developer plans architecture). The project chat keeps everything in one shared context. For small solo projects, a single agent is fine. For multi-role projects where you want parallel work and role specialization, Coze's multi-agent approach is more efficient. Think of it as the difference between hiring one generalist freelancer vs. a small specialized team.