30th December, 2025
Testing is the safety net of any software platform — it’s what gives teams confidence to deploy changes without breaking production. But at Bizom, we realized that traditional E2E testing approaches were no longer sufficient for our fast-paced, complex B2B distribution platform.
In the world of FMCG and retail distribution, our platform serves multiple user types. Traditional test automation scripts, while helpful, had significant limitations: they were brittle, required constant maintenance, and couldn’t adapt to UI changes without developer intervention.
We needed E2E testing that could:
To solve these challenges, we adopted a hybrid agent-based approach, combining the power of Playwright’s robust testing framework with AI agents that could understand, plan, and execute tests intelligently.
We built a solid foundation using Playwright— a modern, reliable browser automation framework. Playwright provides auto-waiting, network interception, multi-browser support, and a built-in test runner.
However, even with Playwright’s robustness, writing and maintaining tests remained a manual, time-consuming process.
We created a three-stage intelligent testing system where specialized AI agents collaborate to plan, generate, and heal tests automatically, powered by large language models (LLMs).
1. Planner Agent: Understanding Test Requirements
The Planner Agent understands what needs to be tested. It explores the live application, maps user flows, analyzes different user journeys, and designs comprehensive scenarios, generating structured test plans in markdown. This ensures semantic coverage—testing based on intent and user goals.
2. Generator Agent: Translating Plans to Executable Tests
The Generator Agent figures out how to test it. This agent:
It understands the page semantically, looking for elements by their accessible name or role, making tests far more resilient.
3. Healer Agent: Self-Healing and Root Cause Analysis
When tests fail, the Healer Agent steps in. It executes failing tests in debug mode, performs root cause analysis (checking for selector changes, timing issues, etc.), edits test files to address issues, and re-runs tests to validate the fix. The Healer automatically fixes 80% of test failures due to minor UI changes.
We ensure E2E tests operate in stateful environments through:
AI agents leverage these fixtures intelligently, understanding when to reuse state versus when to start fresh.
We implemented the Model Context Protocol (MCP)—an open standard for connecting AI models with development tools. MCP acts as a bridge, allowing AI agents to interact with the browser, inspect test results, understand project structure, and generate code following our architectural conventions.
The three agents work in a coordinated workflow:
Our adaptive, intelligent testing system resulted in significant gains:
We built a next-generation E2E testing system by combining Playwright’s robustness with three specialized AI agents. We moved away from treating tests as static artifacts and instead embraced adaptive, intelligent testing.
The journey from manual test scripting to AI-powered autonomous testing has fundamentally transformed how we approach quality assurance at Bizom. Engineers can focus on building great features, confident that AI agents are continuously ensuring quality.