Skip to main content

7. Use task-master-ai as our AI agent and code builder

Date: 2025-04-17

Status​

Accepted

Context​

We are in an experimental phase of AI-assisted development, evaluating tools to help with code generation and project management. Our long-term goal is to develop our own agentic model, but we need an off-the-shelf solution to shape our understanding and requirements.

We evaluated several AI coding tools:

Cursor IDE​

  • Pros: Excellent tab completion, integrated AI chat, familiar VS Code interface
  • Cons: Limited task management, no structured project planning, primarily reactive rather than proactive
  • Experience: Good for individual coding tasks but lacks project-level orchestration

Claude Code (Anthropic)​

  • Pros: High-quality code generation, good reasoning capabilities
  • Cons: No integrated development environment, requires manual task management
  • Experience: Powerful but requires significant manual coordination

Aider (https://aider.chat/)​

  • Pros: Git integration, can work with existing codebases, Python-based
  • Cons: Python dependency conflicts with our JavaScript-focused stack, limited task planning
  • Experience: Good for code modifications but struggled with our monorepo structure

Task-Master-AI (https://github.com/eyaltoledano/claude-task-master)​

  • Pros: JavaScript-based (aligns with our stack), structured task management, MCP integration with Cursor, permissive license
  • Cons: Relatively new tool, smaller community, experimental status
  • Experience: Best fit for our workflow combining task planning with code execution

Decision​

We will use Task-Master-AI (https://github.com/eyaltoledano/claude-task-master) as our primary AI agent and code builder.

Key factors in this decision:

  • Technology alignment: JavaScript-based tool matches our frontend-focused technology stack
  • Integration capabilities: Excellent MCP (Model Context Protocol) integration with Cursor IDE
  • Task management: Structured approach to breaking down projects into manageable tasks and subtasks
  • Licensing: Permissive license allows for modification and customization as needed
  • Workflow fit: Combines high-level project planning with detailed code execution
  • Development tracking: Built-in progress tracking and task completion management

Consequences​

What becomes easier:​

  • Task breakdown and management through structured JSON format
  • Integration with Cursor IDE through MCP (Model Context Protocol)
  • JavaScript-based tooling that aligns with our frontend stack
  • Permission-based licensing allows for modification and customization
  • Better tracking of development progress and task completion

What becomes more difficult:​

  • Dependency on a relatively new, less mature tool compared to established alternatives
  • Potential need to contribute back fixes or features as the tool evolves
  • Learning curve for team members unfamiliar with task-master-ai workflow

Risks to mitigate:​

  • Tool abandonment risk - monitor project activity and have migration plan ready
  • Over-reliance on AI task generation - maintain human oversight of task quality
  • Integration complexity - ensure proper backup workflows if MCP integration fails