Great point about the limitations of AI-powered scrapers on dynamic content. In my experience building 78 production scrapers on Apify, the most reliable approach is combining headless browser rendering with structured selectors rather than relying on LLM extraction alone. For sites with heavy JavaScript, I use Crawlee with Playwright — it handles SPAs, infinite scroll, and auth walls much better than pure HTTP+AI parsing. The key insight: AI is best for understanding scraped data (classification, entity extraction), not for the scraping itself. Curious if you have benchmarked latency differences between LLM-based vs selector-based approaches?