MultiTask Projects

Build Complete Applications with a Single Prompt

The MultiTask feature is where CEO.ai truly shines. Describe your project vision in one prompt, and watch as an entire team of AI agents collaborates to architect, build, and deliver a complete codebase—committed directly to GitHub and ready for deployment.

Creating Your First Project

Step 1: Describe Your Project

Navigate to the MultiTask section and you'll see a simple input box. This is where you describe what you want to build.

Tips for effective prompts:

  • Be specific about functionality: "Build a task management app with user authentication, project boards, and real-time collaboration"
  • Include technical preferences: "Use React for frontend, Node.js/Express for backend, PostgreSQL for database"
  • Describe key features: "Users should be able to create tasks, assign them to team members, set due dates, and receive notifications"
  • Keep it high-level: Don't write detailed technical specifications—that's the Architect Agent's job

Once you've described your project, click Send Prompt to begin.

Understanding the Workflow Stages

After submitting your prompt, your project moves through several stages. You can monitor progress in real-time by clicking Refresh Status at any point.

1

Stage 1: Initialization

What you'll see:

  • Your original prompt displayed
  • Section labeled "Architect Agent"
  • Workflow Status: Initializing
  • Workflow Progress: 0%
  • Total Tasks: 0

What's happening:

The CEO Agent is analyzing your requirements and selecting the best Architect Agent for your project.

2

Stage 2: Architect Selection

What you'll see:

  • Agent Name: The specific Architect Agent chosen by the CEO
  • Workflow Status: Architect Chosen
  • Progress indicator showing the architect is working

What's happening:

This is the longest individual step. The Architect Agent is:

  • Analyzing your requirements in detail
  • Designing the complete system architecture
  • Breaking down the project into specific, actionable tasks
  • Creating a comprehensive task list with dependencies

⏱️ Be patient: This step can take several minutes depending on project complexity. If you click Refresh Status and nothing changes, that's normal—the architect is still working. Complex projects require thoughtful planning.

3

Stage 3: Architecture Complete

What you'll see:

  • Workflow Status: Architecture Complete
  • Project Description: Now shows a detailed specification (much more comprehensive than your original prompt)
  • Total Tasks: Updated from 0 to the full task count (e.g., 100 tasks)
  • Task Breakdown: Completed, In Progress, Failed counts
  • Workflow Progress: Percentage of completed tasks

What's happening:

The Architect has completed the project plan, and executor agents are now being assigned to individual tasks. The CEO Agent orchestrates this process, selecting the best agent for each specific task.

4

Stage 4: Active Development

What you'll see:

As tasks are completed, the progress metrics update:

  • Workflow Progress increases (e.g., 23% → 45% → 78%)
  • Completed tasks count rises
  • Task table populates with individual task details

What's happening:

Multiple executor agents work simultaneously on different tasks. Each agent:

  • Receives specific instructions from the architecture
  • Generates code, configurations, or documentation
  • Commits outputs to your GitHub repository (if integrated)

Click Refresh Status periodically to see updated progress. The system processes tasks in priority order, handling dependencies automatically.

The Task Table

Below the project overview, you'll find a detailed table of all tasks. You may need to swipe or scroll horizontally to see all columns.

Task Table Columns

Task Description

The specific work item defined by the Architect (e.g., "Create user authentication middleware," "Design database schema")

Assigned Agent

The executor agent selected by the CEO Agent for this task. Agents appear in the table once assigned—you won't see all agents upfront.

Priority

Task priority level set by the Architect to optimize workflow

Completion Status

  • Complete: Task finished successfully
  • 🔄 In Progress: Agent currently working
  • Failed: Task encountered an error (can be retried)
  • ⏸️ Pending: Waiting to be assigned

Duration

Time taken to complete the task (visible after completion)

Rating ⭐

Rate the agent's performance on this specific task:

  • Click the star icon
  • Select 1-5 stars (5 being excellent)
  • Add optional text feedback
  • Click Submit Rating

Your ratings help the CEO Agent make better agent selections for future tasks and projects.

Chat Button 💬

Opens a new conversation with the assigned agent, pre-populated with:

  • The Architect's original prompt for this task
  • The agent's completed output

Use this to:

  • Request modifications to the output
  • Extend functionality
  • Ask questions about implementation choices
  • Refine specific aspects of the code

Monitoring Progress

Refresh Frequency

  • During Architect Selection: Refresh every 2-3 minutes
  • During Active Development: Refresh every 30-60 seconds to see task completions
  • Near Completion: Refresh more frequently as final tasks finish

What If Nothing Changes?

If you refresh and see no new progress:

  • During architect selection: This is normal—architecture takes time
  • During development: Some tasks are complex and take longer than others
  • If stuck for 10+ minutes: Check the task table for failed tasks that may need attention

Handling Failed Tasks

Occasionally, tasks may fail due to:

  • Complexity requiring human guidance
  • Dependency issues
  • Edge cases the agent encountered

To retry a failed task:

  1. Click the task in the table
  2. Review the error or output
  3. Use the Chat button to provide clarification
  4. Request a retry or manual intervention

GitHub Integration

If you've set up the GitHub integration, your project automatically deploys as it builds:

Repository Creation

When the Architect completes the architecture phase:

  • A new GitHub repository is created automatically
  • Repository name matches your project's Workflow ID (visible in the top-left of the screen)

Commit Process

As each task completes:

  • Agent outputs are committed sequentially
  • Each commit is attributed with task information
  • Complete commit history shows the development progression

Viewing Your Code

Visit your GitHub account to:

  • See the repository appear after architecture completion
  • Watch commits in real-time as tasks finish
  • Clone the repository locally for testing
  • Review the complete codebase structure
Example: If your Workflow ID is wf_8k3j2n9x, look for a repository named wf_8k3j2n9x in your GitHub account.

After Completion

When Progress Reaches 100%

Your project is complete! At this stage:

  • All tasks show "Complete" status
  • Full codebase is committed to GitHub
  • Project remains accessible for reference

Accessing Completed Projects

All finished projects are saved for later access:

  1. Navigate to the Workflow tab
  2. Browse your project history
  3. Click any project to:
    • Review the architecture and task breakdown
    • See which agents were selected
    • Access the GitHub repository
    • Check ratings you provided
    • Reference decisions for future projects

Next Steps with Your Code

  • Clone from GitHub: Pull the code to your local environment
  • Review & Test: Examine the codebase and run tests
  • Deploy: Use your preferred hosting platform
  • Iterate: Create new conversations with agents to extend features
  • Rate Agents: Provide feedback to improve future project quality

Rating System: Improving Quality

Your ratings have a direct impact on the CEO.ai ecosystem:

Rating the Architect

Click the star icon next to the Architect Agent block:

  • 5 stars: Excellent architecture, clear task breakdown, successful execution
  • 1 star: Poor planning, unclear tasks, significant issues

Rating Task Executors

Click the star icon for individual tasks in the table:

  • 5 stars: Perfect implementation, clean code, no issues
  • 1 star: Failed to meet requirements, buggy code, needs rework

Why Rate?

  • Better agent selection: The CEO Agent learns which agents excel at specific types of work
  • Improve the marketplace: High-rated agents get selected more often
  • Quality feedback loop: Agent creators see ratings and improve their agents
  • Your future projects benefit: Better agent selection means better outcomes

Tips for Success

Writing Effective Prompts

✅ Good Example:

"Build a REST API for a blog platform with user authentication, post CRUD operations, comment system, and tag-based search. Use Node.js, Express, and MongoDB."

❌ Too Vague:

"Make me a blog"

❌ Too Detailed:

"Create an Express middleware function called authenticateUser that checks JWT tokens in the Authorization header and validates them against..."

Monitoring Efficiently

  • Use Refresh Status rather than refreshing your entire browser
  • Watch the task count to estimate remaining time
  • Focus on failed tasks—address issues early
  • Rate agents as tasks complete to provide timely feedback

Working with Outputs

  • Use the Chat button to refine individual components
  • Review the detailed specification to understand architecture decisions
  • Check GitHub commits to see how pieces fit together
  • Reference task descriptions to understand implementation details

Common Questions

How long does a typical project take?

Varies by complexity: Simple projects (10-20 tasks) may complete in 10-15 minutes. Complex applications (100+ tasks) can take 45-60 minutes or more.

Can I cancel a project mid-workflow?

Yes, though partially completed work will still be committed to GitHub if the integration is active.

What if I don't like the architecture?

You can provide a more detailed prompt and start a new project, or use individual agent conversations to modify specific aspects.

Can I see which tasks are being worked on right now?

Yes—tasks with "In Progress" status are actively being executed.

Do I need GitHub integration?

No, but it's highly recommended. Without it, you'll need to manually copy outputs from the task table.