Testing at scale @ Spotify
Our team discusses testing techniques from Swiftable BA.
Qubika at Databricks Data + AI Summit
Join us June 9-12 to see our leading Databricks capabilities
Financial services
Expertise in core banking, BaaS integrations, payments, and GenAI-enhanced financial solutions.
Healthcare
People-centric healthcare design and solutions, from virtual care, integrations, to smart devices.
Insurance
Modern solutions including self-service, on-demand, and algorithm-driven personalization.
Hi-tech & semiconductors
Semiconductor design, firmware and IoT development, and AI-powered embedded systems.
Qubika is a Databricks Select Partner
Learn more about our journey, our 200+ certified Databricks experts, and how we’re delivering solutions such as autonomous AI agents.
Databricks Capabilities
Learn more about Qubika's strong partnership with Databricks as a Select Partner, delivering solutions across the finance, banking, healthcare, hi-tech, and entertainment industries.
Databricks Impact on Financial Institutions
Databricks empowers financial institutions to harness unified data and AI, such as for real-time fraud detection, dynamic risk modeling, and personalized customer experiences.
June 20, 2025
This article explores the impact of AI tools on QA automation, from open-source vs SaaS architectures to RAG and MCP integrations. Learn how modern tools are changing the way we test software.
The integration of AI tools and models is rapidly reshaping the development landscape, especially in Quality Assurance (QA).
The shift from testing as code to testing as instruction marks a significant paradigm shift, demanding new approaches to standardization, safety, and collaboration. This article examines the transformative potential of AI in QA, highlighting critical tools and practices, including open-source versus SaaS architectures, sensitive data handling, language model flexibility, CI/CD integration, and the support for Retrieval-Augmented Generation (RAG) and Model Context Protocol (MCP).
This exploration will pave the way for understanding how these advancements are not just speculative, but are actively redefining the future of QA automation and coding in general.
The market for AI tools and models is transforming the development industry, with critical implications for QA. This document outlines relevant tools and practices, focusing on:
RAG is a technique that augments language models by injecting relevant external knowledge into the generation process. It bridges the gap between static model training and real-time, up-to-date context.
Core Components:
Why it matters for QA: Enables LLMs to answer questions or generate tests based on live product specs, API docs, and historical test failures — not just what they were trained on.
MCP is an open standard for securely connecting external data sources with LLM-powered agents and tools. It defines how context flows between systems.
Basic Flow:
Query -> Knowledge base search / MCP Query -> Relevant context -> LLM generation
Key Benefits:
Use Case: A QA assistant using MCP can pull logs from your CI/CD platform, compare them to recent test failures, and suggest potential flaky root causes — all without exposing raw logs to third-party tools.
The combination of RAG and MCP allow for QA use-cases such as:
Why it’s my tool of choice:
Features:
Model Support:
Feedback:
https://github.com/cline/cline
Why I Use Roo Code (Cline Fork)
Roo Code is my go-to personal AI coding tool. It’s a maintained, open-source fork of Cline that stays true to the original vision — no lock-in, full control, and powerful features baked right into VSCode. You can read here Qubika’s review of Roo Code in an enterprise setting.
While Cline has moved its focus toward a more closed SaaS model (while retaining its open-source counterpart), Roo Code retains the focus on its original API key–based architecture, giving you flexibility to work with OpenRouter, OpenAI-compatible models, or even local models like Ollama.
Why I use Roo Code:
Quick Review:
It feels like pair programming with an agent that actually understands your repo. I use Roo Code to scaffold side projects, prototype CLI tools, automate Markdown edits, and occasionally do quick browser automations. It’s lightweight, extendable, and respects your workflow.
https://github.com/RooVetGit/Roo-Code
Features Roo Code offers that Cline doesn’t (yet):
Strengths:
Feedback:
Features:
Features:
Feedback:
https://github.com/features/copilot
Main Features:
Enterprise Features:
https://about.gitlab.com/gitlab-duo/
Features:
Pricing:
Feedback:
Tool | Best For (in my testing) |
---|---|
Roo Code | Full-control, custom QA workflows, all-around best for daily driving |
Cline | Lighter users, SaaS-friendly setup |
Aider | CLI refactoring, multi-file ops |
Cursor | Predictive coding & IDE UX |
Copilot | Beginner-friendly suggestions |
GitLab Duo | Secure, enterprise DevSecOps flows |
Blackbox.ai | Public code search, snippet lookup, fast fixes |
Dual Trust Model:
Security Tactics:
SaaS Pros:
SaaS Cons:
Open Source Pros:
Open Source Cons:
The intersection of reasoning-capable models, local tooling, and AI-enhanced editors like Roo Code is beginning to reshape QA pipelines and workflows. Here’s a breakdown of promising components and how they can be leveraged:
These next-gen LLMs aren’t just completing code — they’re reasoning across documents, test flows, and structured data.
Use Cases for QA:
By combining browser drivers like Puppeteer with AI agents (Cline, Roo Code, AutoGPT variants), you can generate and validate UI test flows dynamically.
Examples:
LLMs can scan component trees and add intelligent, consistent testID or data-testid attributes.
Practical flow with Roo Code:
Perfect for selector-based tools like Detox, React Native Testing Library, and Playwright.
Both tools allow users to define custom terminal commands accessible directly via the AI agent, enabling:
“Flappy Bird or other small games from one prompt” — Compression Potential
This popular experiment showed how powerful models (like GPT-4, o1, Deepseek) can write entire games from a single, detailed prompt.
Why this matters for QA:
This shows that compressed QA workflows are within reach:
Imagine writing:
“Test the new KYC onboarding screen for successful flow, invalid email, and expired document in Playwright.”
…and getting the full suite scaffolded + commit message suggestions to launch your PR/MR.
These tools and AI integrations are no longer speculative — they’re here. The challenge now is standardizing safe workflows around them, ensuring test coverage is traceable, and enabling teams to collaborate with agents just like devs or testers.
In short, the industry is moving from testing as code to testing as instruction — and these tools are paving the way.
By Avi Tretiak
QA Automation Engineer at Qubika
Receive regular updates about our latest work
Our team discusses testing techniques from Swiftable BA.
Take a look at the different types of testing used by our QA Studio.
Qubika’s QA Studio shares the ultimate guide to validation testing, sharing everything you need to know about testing.
Receive regular updates about our latest work
Get in touch with our experts to review your idea or product, and discuss options for the best approach
Get in touchArtificial Intelligence Services
Accelerate AI
Healthcare Solutions
Data
Agentic Factory
Financial Services Technology
Platform engineering
Data Foundation