🚀 My First Experience with MCP & AI Agents (and it blew my mind) 🤯 After seeing more and more posts about MCP—especially the bite-size videos from Debbie O'Brien — I had to investigate. The whole Agentic AI wave was all around me, and I had never paid much attention to it, being busy with other things. But once I decided to dive in… my mind was blown 🤯. 👨‍💻 The Mission I decided to start my adventure with MCP and the GPT Agent in VS Code (Insider Edition)by trying to rebuild the Playwright test suite for a portfolio page I had built a while ago. It’s nothing special—just vanilla Playwright tests using Page Object Model (POM)for reusability, clean structure, and easy maintenance. 🧠 Zero Code Written by Me Yes, literally—I didn’t type a single line of code or run any terminal commands.Istarted by adding the MCP Playwright Server. Super easy: -> Run Add Server -> Type @playwright/mcp -> Hit Enter -> Click Allow -> ✅ Done. Switched to Agent Mode in Copilot… and the rest is history. I simply told the Agent: -> Install Playwright & dependencies -> Initialize the project -> Use a prompt (originally by Debbie O'Brien, slightly adjusted for my setup) 🎯 After that, it generated all the tests for me. ❌ Some tests were failing… I just said: 🛠️ “Fix all the failing tests and run them again.” And like magic ✨—it kept fixing them until they passed! I then asked: 🧱 “Add POM support.” Copilot: “Sure, no problem.” ✅ 🍬 But what came next made me feel like a kid in a candy shop! It asked: “Would you like further enhancements?” And presented a list of suggestions! 🤯 🚀 Suggested Improvements: ✅ More assertions for images, content, dynamic elements 🔁 Navigation flow tests using page objects 🔗 External link validation (e.g., GitHub, LinkedIn) 📊 Parameterized tests using test.each() ♿ Accessibility checks with axe-core 🖼 Visual regression via screenshot diffs 🌐 API testing & mocking 📦 Shared base classes for reusable checks ❌ 404/error page testing ⏱ Performance tracking (page load times) It even offered to help implement any of them on the spot. ⏱ All of this took just 20 minutes. 😳 🤔 This experience got me thinking… Is this the future of SDET work? Are we heading toward a world where we: -> Generate prompts 🎤 -> Let AI handle the boilerplate ⚙️ -> Focus more on quality oversight, strategy, and edge cases? Part of me says: ✅ Yes—it removes time-consuming grunt work, puts the human in control, and lowers the barrier for non-coding QAs to join automation. But also… 💻 I like to code! Either way, the possibilities are HUGE. And this is just the beginning. 🌱 If you like to see the repo it is here.