Agentic AI: Plain-English Requirements to Automated Tests

This project is a working proof-of-concept built with Flowise that demonstrates how an agentic AI system can convert natural language software requirements into structured, automated test steps.

The intent is not to create a production-ready pipeline, but to explore and showcase how modern large language models (LLMs) can bridge the gap between manual QA input and executable automation.

Core Features

• Prompt Template Node: Captures plain-English input such as “the user logs in with username and password.”
• LLM Node: Interprets the requirement using OpenAI’s model (or compatible LLMs) to reason about appropriate interaction steps.
• Structured Output Parser: Validates the output against a defined schema to ensure it matches a predictable JSON structure.
• Custom JS Function: (optional) For transforming, filtering, or routing results between nodes.
• Set Variable & Write File Nodes: Store results and optionally write out the generated test steps to file.

Example Input

"The user logs in with username and password and clicks submit."

Example Output

{
  "steps": [
    { "action": "type", "target": "#username", "value": "testuser" },
    { "action": "type", "target": "#password", "value": "password123" },
    { "action": "click", "target": "#submit" },
    { "action": "wait", "target": "#dashboard" }
  ]
}

Why This Matters

QA automation is often bottlenecked by the manual translation of requirements into test cases and then into automation code. This POC highlights a future where AI can support testers by rapidly prototyping test steps, even if human oversight remains essential for validation and deployment.

Limitations

While promising, this is an empirical experiment. Real-world usage requires hardened flows, richer context handling, input validation, and integration with production-grade testing tools like TestCafe, Playwright, or Cypress.

This POC was created using Flowise (v2.2.8) running locally on my computer — not the hosted Flowise cloud version. Also using the chatGPT API, You can download the canvas JSON file to load it into your own local Flowise setup: Download Flowise Canvas JSON Also you will need to setup on your flowise configuration the API key to your own OpenAI API