My AI Development Workflow: From Feasibility Analysis to High-Fi Prototyping & Black-Box QA

People often ask me what my AI-assisted software development workflow looks like. I recently implemented a feature entirely with AI in my desktop app BaoCut, and this real case study serves as a complete reference.

1. Feasibility Analysis

Diving into code without feasibility analysis has cost me heavily in the past. Feasibility consists of two layers:

• Product Layer: Is this feature genuinely valuable? Does it align with the core product mission?
• Technical Layer: Is the technical path feasible, robust, and cost-controlled?

I passed the GitHub issue request into Claude Code. It analyzed the current codebase and proposed 4 approaches. I chose Option A (App native node service) paired with a lightweight HTTP API.

2. Drafting Design Documents

The design document acts as a hybrid specification combining product requirements, architectural design, and UI specs. Its primary purpose is to ground the AI in project constraints and allow the human engineer to verify direction before touching code.

3. High-Fidelity Prototyping ("Prototype IS UI Design")

Using the baoyu-design skill, I generated an interactive high-fidelity prototype directly in the App's dedicated design environment.

I iterated several times: splitting host and client into tabs, adding dynamic status indicators, and eventually elevating the server status card directly to the main screen.

4. Implementation

With complete design docs and high-fi prototypes, I instructed Claude Code using /goal to execute each milestone sequentially and verify via self-captured screenshots.

5. Testing & Black-Box QA