Scalable Playwright Framework with AI

Because the way we build and maintain test automation frameworks is evolving fast, AI is becoming a practical part of the everyday testing workflow.
We will begin with a brief overview of AI, LLMs, and MCP, focusing on how these concepts fit into modern test automation and agentic workflows.
From there, participants will clone a GitHub repository containing a lightweight Playwright and TypeScript setup. Together, we will progressively add AI-powered capabilities to the project.

We will start by using Playwright’s Codegen feature to record a simple test against an e-commerce application and understand the baseline structure. We will then explore the custom instructions already included in the project and see how project-specific guidance can influence the way an AI coding agent works with Playwright and the project’s conventions.

Next, we will explore GitHub Copilot’s agentic capabilities and see how custom agents can be tailored for specific testing workflows. We will start with a Planner agent, using the Playwright MCP server to allow the agent to interact with the application in a browser, explore its functionality, and generate a structured test plan. We will then use the generated scenarios to create executable Playwright tests and explore how AI can assist with building and maintaining a Page Object Model.

Once we have generated some tests, we will introduce a custom Code Reviewer agent designed specifically for our automation project. It will review the generated code against our project’s conventions and provide feedback on areas such as test structure, maintainability, and Playwright best practices. We will also briefly explore the Agent Package Manager (APM) and how it can be used to package, version, and share reusable AI assets such as agents, instructions and skills across different project teams.

An important part of the workshop will also be understanding that not every AI task needs to rely on a cloud-hosted model. I will demonstrate how a locally running LLM can be integrated into the workflow using LM Studio and a Qwen model. We will look at the practical considerations behind running an LLM locally, including token usage, context windows, temperature, model capabilities, and how to decide which tasks are a good fit for a local model. The goal is to show how a hybrid approach can help us be more efficient with AI resources while keeping the right tools for the right tasks.

Finally, once the tests are stable, we will run them through a GitHub Actions CI/CD pipeline, demonstrating how AI-assisted test automation can fit naturally into a continuous testing workflow.

By the end of the workshop, participants will have both a conceptual understanding of AI-assisted test automation and a ready-to-extend Playwright project demonstrating practical workflows with GitHub Copilot, MCP, custom agents, APM, and a local LLM.

More importantly, they will have a better understanding of where AI can genuinely improve the daily work of an automation engineer: from test exploration and generation to code review, reusable AI workflows, maintenance, and CI/CD.