AutoDev Planner is a task planning system that enhances AI-assisted coding by making progress visible and interactive. Inspired by AI coding tools like GitHub Copilot Workspace, Cursor, and JetBrains Junie, it aims to improve transparency and adaptability in AI-driven development workflows.

Key Features:

  1. Visible Task Planning – Users can track task progress through the Planner ToolWindow and pin important tasks.
  2. Dynamic Task Adjustment – AI dynamically refines plans based on context, though responsiveness depends on the model used.
  3. Manual Execution – Users can manually execute incomplete tasks to fine-tune the development process.
  4. Task Review – Manual review of task plans using AI, allowing users to optimize their workflow while controlling token usage.

Core Technical Aspects:

  • Inference Model-Based Planning – Tasks are structured using reasoning models like DeepSeek R1, which requires custom prompts for better execution.
  • Interactive Task Management – Users can mark tasks as complete, open related files, edit plans, and review AI-generated strategies.

By making AI-driven coding tasks more transparent and controllable, AutoDev Planner enhances the development experience, allowing users to engage with and refine AI-generated code plans.

Design inside Planner

In Is Design Dead?, Fowler concludes that design is far from dead, but its role has changed. Instead of being a rigid, upfront process, design in an agile world is continuous, incremental, and driven by refactoring and testing. Agile methodologies, particularly Extreme Programming (XP), emphasize evolutionary design, allowing architecture to adapt naturally as the system grows.

As we know, AI models function as black boxes, capable of generating vast amounts of code. While this ability is powerful, it can also become harmful at scale if the generated code lacks proper design principles. Poorly structured AI-generated code can lead to technical debt, maintainability issues, and architectural inconsistencies, making it difficult for teams to evolve their software effectively.

The Need for a Well-Designed AI Agent

To mitigate these risks, we need to design an AI agent that enforces and supports structured software development. Such an agent should focus on:

  • Planned and Evolutionary Design – Combining strategic planning with the flexibility to evolve the design over time. Instead of enforcing rigid upfront designs, the AI agent should guide the developer in iterative refinement.
  • Growing an Architecture – Ensuring that as the system scales, its architecture remains cohesive and adaptable, avoiding monolithic, tangled structures.
  • Simple Design – Encouraging minimalism and clarity in the codebase, following principles like YAGNI (You Ain’t Gonna Need It) and KISS (Keep It Simple, Stupid).

For example, when asking an AI to design a complex frontend page, one good practice is ViewModel splitting. Instead of generating a monolithic UI structure, the AI should separate concerns properly, ensuring maintainability and reusability. By guiding AI-generated code with solid architectural principles, we can create scalable and sustainable software solutions.

Plan Example

Example of a Generated Plan:

  1. Identify Core Workflow Classes
    • [✓] Search for class definitions containing “Workflow” and “Sketch”
    • [✓] Analyze the execute method in SketchRunner
  2. Analyze AI Flow Execution Stages
    • [✓] Identify the context collection phase
    • [✓] Analyze the tool invocation decision module
    • [✓] Trace the code generation pipeline
  3. Validate Workflow Integrity
    • [✓] Check the exception handling mechanism
    • [✓] Confirm version control integration points