Why Your Web Scrapers Keep Breaking (And How to Build Self-Healing TypeScript Agents Using LLMs and Playwright)

The article argues that selector-based automation is brittle because CSS selectors and XPath often break when websites change.

It proposes a self-healing architecture built with TypeScript, Playwright, multimodal LLM vision, and Model Context Protocol contracts.

When the DOM shifts, the LLM can use page context to re-identify the right elements and keep scraping or form filling running.

WebGPU acceleration is intended to make recovery faster and more production-ready, reducing downtime from frequent UI changes.