How AI is Revolutionizing Software Testing: 2025 Roadmap and Real-World Successes

Since the end of 2023, the Quality team at Qunhe Technology has repeatedly emphasized the value AI can bring to testing, establishing a dedicated AIGC virtual group to build infrastructure and explore specific domains.

By the end of 2024, the release of DeepSeek demonstrated powerful AI inference capabilities, prompting industry‑wide exploration of AI‑driven quality testing and efficiency. The team concluded that staying on this trend requires equipping themselves with various AI tools, integrating them into workflows, and improving overall effectiveness.

For 2025 the team will focus on four key areas:

Exploring better AI models to enhance test efficiency.

Building a more user‑friendly AI testing platform for easy integration.

Continuously improving testing efficiency across processes.

Promoting AI awareness and usage among testers.

01 Exploring Better AI Models for Test Efficiency

Since the start of 2025, the testing platform has integrated several leading large models, receiving positive feedback.

2025‑01: DeepSeek v3&r1 – knowledge‑base Q&A –

2025‑02: text‑embedding‑3‑small bge‑m3 – vector model & knowledge base – accuracy improved by 20% –

2025‑02: Claude 3.5 & Claude 3.7 Sonnet – automated test code generation – stable local generation, significantly higher accuracy –

02 Building an Easy‑to‑Use AI Testing Platform

More than 20 applications have already connected to the Quality department’s AI testing platform FastQA, which continues to be optimized and upgraded.

2025‑02: Model integration optimization –

2025‑02: Platform version upgrade – added plugin support, workflow orchestration, and more stable versions.

2025‑01: Enterprise WeChat (Qixin) Q&A bot integration – supports @robot conversation for knowledge‑base queries.

03 Boosting Efficiency in Various Test Domains

3.1 AI‑Assisted Ticket Efficiency

Maintain an AI ticket knowledge base and expose OpenAPI for online Q&A and automatic ticket creation.

Best practice in the technical support team.

80% of internal inquiries are auto‑replied, reducing repetitive work.

Automatic ticket creation improves submission quality and saves clarification time.

Applications include group‑chat @AI virtual account for FAQ, auto‑create tickets, and in‑app “Customer Service Mini‑Bot”.

3.2 UI Automation Code Generation

Use VS Code plugins and AI knowledge base to generate automation scripts from test case descriptions, and Greasemonkey plugins to produce xpath expressions compliant with Kuke Home UI standards.

Accuracy between 50‑80% across three pilot projects, reducing xpath authoring cost by ~30% and improving quality.

3.3 AI‑Curated Document Sharing

AI crawls and analyses recent high‑quality documents, pushing selected ones for learning.

Deployed across the quality department, receiving strong internal approval.

3.4 Second‑Level AI Knowledge‑Base Construction

Parse cf tags to quickly convert cf documents into FastQA knowledge‑base entries.

Two weeks of operation added ~2k pages.

Provides end‑to‑end service: AI knowledge‑base creation, cf tagging rules, AI‑driven Q&A bots, OpenAPI integration.

Process review scheduled for Q2 to become a company‑wide ITM standard.

3.5 AI‑Generated Test Cases

AI agents parse documents and images, split functional points, and generate bilingual test cases.

Significantly improves case‑writing efficiency.

3.6 AI Page Inspection

Detect multilingual and layout issues in internationalized pages using AI knowledge‑base and preset correct/incorrect translations.

Typical page error detection rate exceeds 80%; the process is now automated and running continuously.

04 Integrating AI into the Test Management Platform

4.1 AI‑Assisted Test Analysis Report Writing

Leverage LLMs (text + vision) to generate draft analysis reports from requirement, cf, and design documents, providing testers with inspiration.

4.2 AI‑Generated Test Cases

LLMs generate initial functional cases from requirements; after review, cases are imported with one click, expanding coverage and detail.

4.3 Impact‑Case Automatic Identification (Research)

Use RAG/KAG models to retrieve relevant cases from the library, quickly presenting them to testers and saving search time.

4.4 AI‑Driven Test Case Review

Define strict review rules for LLMs; the model scores cases and offers targeted improvement suggestions, ensuring high‑quality test cases.

05 Exploring More Domains

Additional initiatives include AI‑assisted classification of SevenFish chat conversations, AI integration into the Apollo automation platform, and other projects aimed at improving automation issue handling and knowledge‑base enrichment.

06 Continuous AI Promotion

The department follows a three‑step plan: AI training and certification, familiarization with AI tools, and applying AI to business‑specific testing scenarios.

07 Conclusion

AI continues to deliver massive value to software engineering and testing. By learning, adopting, and deeply integrating AI with existing systems and processes, testers can unlock its full potential.