Build a Selenium-to-Playwright AI Agent in Just 90 Minutes. Watch Babu Manickam, CEO of Testleaf, build an AI agent live using Vibe Coding — without manually writing a single line of code.
Hundreds of scripts. Locators to rewrite, waits to change, assertions to redo. Most teams put it off because nobody has two spare days.
In this session we skip the rewriting. We build an agent that does it, and we build it in front of you.
You describe what you want in plain language. The AI writes it. See how far that goes, and where you still need to step in.
Built with a no-code approach, step by step. You can follow along and build your own after the session.
Babu once built an AI agent that ran up a ₹5 lakh bill overnight. He'll walk through what went wrong and how to stop it happening to you.
An agent that loops, retries or calls a model too often can burn through money while you sleep. Agents also get access to your code, keys and systems. So half of this session is about running them safely.
Where tokens actually go, why agents overspend, and the limits and checks to set before you leave one running.
What to keep away from an agent, how to limit what it can touch, and the mistakes that are easy to make.
Rough plan. The build happens live, so timings can shift.
What it is, what works, what breaks.
What the agent did overnight and how it got missed.
Live, no-code. Start with an empty screen, end with a converted script running.
Limits, guardrails and a checklist you can reuse.
Bring your own framework questions.
Full source code and setup notes. Yours to keep, nothing to sign up for afterwards.
Pick a field, or a whole page of fields, and it turns what it finds into a Cucumber feature file or locators and page code. It isn't tied to Selenium. The same approach works with Playwright, Cypress or Puppeteer.
You'll also see how it's built, so you can adapt it to your own stack.
Testers with a Selenium suite who are eyeing Playwright
SDETs who want to build agents without writing a lot of code
QA leads who want to know what AI agents will really cost
The testers who adapt first will lead. Build real AI skills while others wait.
The gap is opening between people who prompt and people who build the agent.
Token budgets and guardrails are turning into part of the job.
Tool user, workflow builder, LLM tester and AI engineer. Each asks for more depth.
Prompts, context and logic, with your domain knowledge as the edge.
Rewriting by hand is the slow part. It doesn't have to stay that way.
What you give an agent matters as much as what you ask it to do.
CEO of Testleaf. 25+ years in software and testing across four continents. He builds AI testing tools himself, which is why he's made the expensive mistakes already.
Logos represent where our learners work; no direct endorsement implied
No. The agent is built with a no-code approach. If you know Selenium, you'll follow the conversion easily.
No. It isn't tied to one tool. The approach works with Selenium, Playwright, Cypress or Puppeteer, and you get the source code to adapt.
Everyone who registers gets the recording. Live attendees can ask Babu questions directly.
It keeps the room to people who plan to show up.