6 Practical AI Examples in QA

1. Generating test cases from a user story (CRISPE + Few-Shot)

You are a senior QA engineer with 8 years of experience in fintech.
Your task: generate test cases based on a user story.
Know that: the project uses React + Java 17 + PostgreSQL.
Format: ID | Title | Preconditions | Steps | Expected Result | Priority.

Example:
"As a user I want to change my password"
→ TC-001 | Password change - happy path | Logged-in user |
  1) Profile, 2) Change password, 3) Old and new, 4) Save |
  Password changed | High

Now: "As a user I want to pay by card with 3D Secure"
Include: happy path, expired cards, insufficient funds, 3DS timeout.

Tool: ChatGPT or Claude Sonnet

2. Edge case analysis (Tree of Thought)

What edge cases exist in the card payment module (3DS2, Visa/MC, PLN/EUR/USD)?

3 perspectives:
A) User (UX/flow)
B) Integration (API, timeout, retry)
C) Data (currencies, limits, card formats)

For each: 5 edge cases + risk + how to test.
Which 5 across all are the most important?

Tool: Claude Opus

3. Bug description for Jira (ARC)

As a senior QA, write a bug description for Jira.
Context: Checkout, BLIK payment, clicking "Pay"
triggers no action. Chrome 122, Firefox 124, Windows 11.
Console: "TypeError: Cannot read property 'submit' of null".
Staging, v2.4.1.

Format: Title | Description | Steps | Expected | Actual |
Environment | Severity

Tool: ChatGPT (speed), Rovo (Jira context)

4. Business case for automation (SCQA)

Situation: 8 QA, 12 modules, 650 manual test cases,
release every 2 weeks.
Complication: Regression takes 5 days = 40 person-days/month.
2 critical bugs leaked to production.
Question: How to justify automation to management (non-technical)?
Answer: Business case, max 300 words, with ROI.
No technical jargon.

Tool: Claude Opus or ChatGPT

5. Test code from API documentation (RAG + TDS)

Based ONLY on the following documentation:
---
POST /api/v2/payments
Headers: Authorization: Bearer {token}
Body: { "amount": number, "currency": "PLN"|"EUR"|"USD",
        "card_token": string, "idempotency_key": string }
Responses: 201, 400, 401, 409, 422
---

Write REST Assured tests (Java 17 + JUnit 5).
Cover all response codes.
@ParameterizedTest for currencies.
Assertions on response time (<2000ms).
Do not invent fields not in the documentation.

Tool: Claude Code, Codex, Copilot

6. Sprint quality report (CO-STAR)

Context: Sprint 14. 22 stories, 18 accepted, 4 rejected.
7 bugs (2 critical, 3 major, 2 minor). Auto coverage: 62%.
Objective: Quality status for stakeholders.
Style: Professional, data-driven.
Tone: Factual, positive but honest.
Audience: PO + CTO.
Format: Email, max 200 words.
Sections: Summary, Risks, Recommendations, Next Steps.

Tool: ChatGPT, Claude Sonnet

---

In the next post: Multi-Agent — how to orchestrate multiple AI tools together.