What's broken today.
Research agents need fresh, real-time web data to ground their answers. Static training data is months behind. Hardcoded API keys to scraping providers don't scale across thousands of agent instances or rotate cleanly when keys expire.