When the Development's "Groove" Was Disrupted by Claude Code: The Tuning Art of a Triple-Blade Workflow Crafted with AI
Everyone has likely experienced encountering unexpected technical noise in development. I, too, recently faced local dependency conflicts and build errors during a session with AI (Claude Code) on a project. At that time, the development groove—that is, the rhythm and creative flow—was temporarily disrupted, and I experienced almost losing sight of the product's essential value design.
TL;DR
In development with AI, the author believes it is essential to not only solve local problems but also to re-present the product's business goals and overall architecture to the AI, and to adopt the perspective of a 'conductor' who fine- tunes the entire system.
Background and the Author's Three-Sword Style Workflow
Based in Osaka, I am a sole proprietor with three faces: musician (founder of Hormonize Records), web engineer, and business owner. In my daily development, I leverage modern technology stacks such as Supabase and GCP, centered around Next.js, TypeScript, and Python (FastAPI). And to maximize development efficiency, I have actively utilized the AI assistant, Claude Code.
In this three-sword style workflow, while implementation skills as an engineer are certainly important, perspectives such as 'harmony' and 'improvisation' as a musician, and 'overall optimization' and 'vision presentation' as a business owner are also indispensable. AI was supposed to be a powerful partner to accelerate this very workflow.
Stumbling Block: The Disrupted Development Groove
This is a story from a recent project where I was developing a new feature module in collaboration with Claude Code. When I proceeded with the implementation based on AI's suggestions and tested it in the local environment, unexpected dependency conflicts and build errors frequently occurred.
For example, problems such as different modules defined within package.json referencing incompatible library versions, or Typescript type definitions not matching specific framework versions. Claude Code accurately analyzed the error messages and immediately presented several makeshift solutions.
However, the suggested solutions sometimes triggered new problems or had the potential to undermine the integrity of the existing architecture. I became so engrossed in the localized debugging work, as if stuck in quicksand, that I felt my awareness drifting away from the essential business goals I should have been considering, such as 'what value this feature provides to the user' and 'how it contributes to the overall growth of the product.' It was precisely the moment when the development groove was disrupted, and the session stalled.
Solution: Redefining AI as an 'Orchestra Tuner'
In this situation, I decided to stop coding for a moment. Then, I re-approached Claude Code with the following steps.
- Temporary Suspension of Coding and Situation Assessment
Instead of reacting emotionally to the immediate error, I calmly organized what the problem was and where I was heading. - Re-presenting the Product Vision and Architecture to Claude Code
Beyond just reporting an error, I explained to Claude Code the business goals of the product under development and the current overall system architecture design (responsibilities of each module, data flow, key technology choices and their reasons, etc.) in a more detailed and comprehensive manner. - Requesting a Solution Considering Global Optimization
Instead of just fixing a specific file or line, I asked for 'a solution that considers the impact of this dependency conflict or build error on the overall system health, is highly consistent, and easy to maintain in the future.' It was like an orchestra conductor seeking harmony in the overall sound, not just the pitch of a specific instrument. - Review and Application of AI-Proposed Structural Improvements
Claude Code deeply understood my intentions. What was presented was not just a version upgrade proposal, but more fundamental and structural improvement proposals, such as redesigning dependencies in specific modules, abstracting into a common utility layer, or refactoring to more rigorously apply TypeScript's Strict mode.
For example, I conveyed my intentions with a prompt like the following.
「現在開発中のNext.jsアプリケーションでは、ユーザー管理をSupabase、API連携をFastAPIで行っています。今回の機能(例: リアルタイム通知)は、ユーザーエンゲージメント向上というビジネスゴールに直結します。現在、通知モジュールとSupabaseクライアント間の依存関係でビルドエラーが出ていますが、これは単にバージョンを合わせるだけでなく、今後スケールすることを考慮し、各モジュールの責務を明確にした上で、システム全体として最もクリーンで一貫性のある解決策を提案してください。」Based on this broad context, Claude Code logically explained not only specific code modification proposals but also why those modifications were necessary and how they would affect the overall architecture. For example, it proposed transitioning from monolithic dependencies to a more loosely coupled interface design and suggested concrete refactoring code to achieve that.
// 提案された修正の一例(ダミー)
// BEFORE: 直接的な依存関係で型定義が衝突
// import { SomeService } from 'shared-library-v1';
// AFTER: 抽象化されたインターフェースを介した依存関係
interface INotificationService {
sendNotification(userId: string, message: string): Promise<void>;
}
class SupabaseNotificationService implements INotificationService {
// ... Supabase specific implementation ...
}
// DIなどを活用し、具体的な実装を分離This approach allowed me to regain my development groove, improving the overall system quality without wasting time on localized problem-solving.
Summary and the Tuning of a New Three-Sword Style Workflow
What surprised me in this particular instance was the depth of Claude Code's contextual understanding and the speed of its response to problem-solving. It felt like the exhilaration of improvising with a skilled musician.
AI is not just a code generation tool. If I, as the 'overall conductor,' clearly present the product's vision and the entire architectural overview, AI understands the role of individual 'instruments (modules)' and becomes a reliable co-creator that optimizes their tuning and performance (implementation and debugging).
Through this experience, my three-sword style workflow has deepened further. I delegate routine tasks like detailed implementation and debugging to AI, while I, as a manager, strategize for product growth, and as a musician, focus on 'overall conduction' and 'vision presentation,' designing the system's overall 'groove' and 'resonance.' This was an experience that truly made me realize the establishment of a new workflow, balancing creativity and efficiency.
Going forward, I intend to further explore the 'groove' of development through collaboration with AI, pursuing the creation of unique products where technology, business, and art converge.
