agentic-workshop-novus: Spec-Driven Development Practice with AI Programming Assistants

This article introduces the agentic-workshop-novus project, a hands-on workshop focused on Spec-Driven Development (SDD) that aims to teach developers how to collaborate efficiently with AI programming assistants like GitHub Copilot and Cursor. The workshop covers core content such as the art of writing specifications, AI collaboration patterns, and hands-on project exercises, helping developers master the SDD methodology and its application in AI collaboration.

With the popularity of AI programming assistants like GitHub Copilot, Cursor, and Claude Code, software development processes are shifting from traditional linear models to dynamic collaborative models. As an emerging methodology, Spec-Driven Development (SDD) emphasizes writing clear and detailed specification documents first, providing clear context and constraints for human developers and AI assistants, and serving as a key link for human-AI collaboration. agentic-workshop-novus is a hands-on workshop built around this concept.

The workshop consists of three core modules:

1. The Art of Writing Specifications: Teaches methods to write clear, complete, verifiable, and hierarchical specifications, transforming ambiguous requirements into precise documents through cases.
2. AI Collaboration Patterns: Covers techniques such as prompt engineering, iterative refinement, AI-assisted code review, and test-driven AI.
3. Hands-on Project Exercises: Guides participants to complete web application development projects (RESTful API, database operations, front-end components, integration testing, etc.) to experience the full SDD process.

agentic-workshop-novus uses a modern technology stack:

• Backend: FastAPI/Express.js (emphasizing type safety and API specifications)
• Frontend: React/Vue.js + TypeScript
• Database: PostgreSQL/SQLite + ORM
• AI Tools: GitHub Copilot, Cursor, Claude Code
• Testing: Jest/Pytest
• Documentation: Markdown, Mermaid, etc.

Advantages of SDD compared to traditional development:

1. Improve AI Code Quality: Good specifications increase the first-pass rate of AI-generated code by 30-50%.
2. Reduce Communication Costs: As living documents, specifications help team members quickly understand the system.
3. Facilitate Maintenance and Evolution: Specifications make architectural decisions explicit, supporting updating specifications first before generating code.
4. Automated Verification: Specifications can be converted into test cases to ensure implementation aligns with requirements.

Best practices summarized by the workshop:

• Divide and Conquer: Decompose complex tasks into small specification units.
• Example-Driven: Provide specific input and output examples to help AI understand.
• Type-First: Prioritize defining data types and interface contracts.
• Continuous Verification: Run tests frequently and provide feedback for corrections.
• Version Control: Include specifications in version control to track evolution.

Applicable scenarios for SDD:

• API/microservice development (clear interface specifications)
• Data processing pipelines (clear input and output)
• Front-end component development (interface and visual specifications)
• Test case generation (converting specifications to executable code)

Limitations:

• Exploratory programming (unclear requirements)
• Highly creative tasks (algorithm design, architectural innovation)
• Legacy system maintenance (lack of specification documents).

With the improvement of AI assistant capabilities, SDD is expected to become a mainstream methodology. Future developments may include:

• Intelligent specification understanding and generation tools
• Bidirectional synchronization mechanism between specifications and code
• Specification-based automated test generation
• AI-assisted specification review optimization

Mastering SDD skills will be a key capability for developers to collaborate efficiently with AI, and agentic-workshop-novus provides learning resources and practical guidance for this purpose.