How AI Can Auto‑Generate Test Cases from PRDs and Cut Design Time by Up to 70%

Project Background

Traditional test case writing relies heavily on manual effort, resulting in low efficiency, insufficient coverage, inconsistent styles, and a steep learning curve for new testers.

Labor‑intensive: designing high‑quality cases consumes significant time and effort.

Insufficient coverage: boundary, exception, and combinatorial scenarios are often missed.

Inconsistent quality: different QA engineers produce varied styles and completeness.

Knowledge bottleneck: high onboarding cost for newcomers.

Solution

Leveraging the rapid development of AIGC (Artificial Intelligence Generated Content), we built an end‑to‑end solution that combines large language models, classic testing theory, and human‑machine collaboration to convert PRDs into test cases automatically.

Technical Architecture

The system is divided into four layers:

User layer : UI for triggering generation, displaying cases, and adoption.

Business processing layer : Built with Flask, handles asynchronous tasks, file uploads, and generation status management.

AI core layer : Uses LangChain to interact with large language models, providing data streaming and callback mechanisms.

External service integration layer : Connects to test‑case management platforms and messaging services for seamless workflow integration.

Prompt Engineering

Prompts act as the thinking process of a senior test engineer. The workflow includes:

Structured PRD parsing : Extract key requirements, functional scope, and change points.

Requirement extraction : Identify test‑relevant information such as modules, user flows, and edge conditions.

Test scenario generation : Combine LLM capabilities with testing techniques (boundary analysis, equivalence partitioning) and knowledge‑enhancement to produce comprehensive scenarios.

Structured test case output : Convert scenarios into BDD‑style test cases with module name, case name, and detailed execution steps.

Project Outcomes

In a two‑week pilot with one team, the AI system generated over 1,500 test cases, achieving an adoption rate of 89.3% and a coverage rate of 80.6%. Design time decreased by up to 70% for medium‑size requirements.

Benefits

Efficiency boost: up to 50% reduction for large requirements, 70% for smaller ones.

Quality improvement: higher coverage of boundary and exception scenarios.

Scalability: low integration cost and rapid rule updates.

Human productivity: testers focus on exploratory testing, reducing onboarding time.

Future Plans

Knowledge‑enhancement: inject business knowledge bases and historical cases to build a smart knowledge graph.

Multi‑agent collaboration: adopt an agent‑based architecture for full‑process automation of generation, verification, and optimization.

Continuous iteration: refine models based on user feedback and real‑world performance.