Now, more than ever, improving software quality is consistently a top performance objective for all organizations.
“Quality” encompasses both the characteristics of the software product (such as its reliability, usability, performance, and security) and the “processes” used to develop and test it. Processes is therefore of no less importance. Given what AI has been allowing us to achieve.
Onboarded to a mission-critical software solution for a prominent client in the financial sector, Our Quality Assurance Engineering team faced an industry-wide bottleneck. A complete absence of functional requirements and system documentation.
Navigating an undocumented codebase forces teams into reactive, manual testing that slows down release velocity. To break this cycle, we engineered a custom, AI-augmented Quality Engineering framework that reverse-engineers application logic and automates end-to-end (E2E) test generation.
Phase 1 – Reverse Engineering the Functional Baseline
Before executing the automation model, we had to solve the documentation debt. Operating with AI as an architectural co-pilot, our team reverse-engineered the undocumented application from the ground up.
In a short-focused sprint, we have drafted three comprehensive functional requirement specifications, 0covering complex macro-functionalities, and 15 structured User Stories, establishing a human-verified baseline, for both, the business and technical stakeholders.
Phase 2 – The Multi-Agent Test Generation Engine
With the functional baseline now in hand, we engineered an autonomous multi-agent workflow. Coordinated by a central orchestrator, five specialized AI agents execute tasks in a strict, deterministic sequence:
- The ISTQB Analyst (Gherkin Phase): Parses User Stories to generate structured logical scenarios (Happy, Edge, and Error paths).
- The Explorer (MCP Phase): Navigates the live interface in real time to map DOM locators with surgical precision.
- The Data Engineer (Data-JSON Phase): Extracts and structures test data payloads into execution-ready JSON format.
- The E2E Developer (Playwright Phase): Translates specifications, locators, and data into modular, production-ready Playwright JavaScript test suites.
Phase 3 – Human-in-the-Loop Governance & Impact
To prevent the common risks of unmonitored AI output, the introduced framework embeds strict Human-in-the-Loop governance checks. Engineers validate generated Gherkin scenarios and conduct rigorous peer reviews on all Playwright code before merging to staging. Gherkin scenarios are plain-English, structured descriptions of software behavior used to link the language between business stakeholders and technical teams.
By combining autonomous agent velocity with human engineering oversight, we achieved remarkable results:
- 138 Individual Tests Generated: Across 32 robust E2E Playwright scripts.
- Exponential Velocity Boost: Reached a test volume in weeks that previously required six months of fragmented manual effort.
- Maturity & Scalability: Advanced our testing model from reactive manual checks (Level 1) to an optimized, AI-driven tooling ecosystem (Level 4) now being adopted across our internal testing practice.
By embedding multi-agent AI workflows directly into the SDLC, we didn’t just automate test execution, we eliminated technical debt, reclaimed engineering time, and established a modern benchmark for enterprise Quality Assurance.
The client, after expressing satisfaction with the results, enormous long-term value, and the team’s innovative proactivity, decided to adapt the methodologies internally to other solutions.
Learn more about application performance
By Lisa Scarpellini