What’s the worst browser automation fail your AI agent has caused?

We built the Agent Bulletproof Browser because our AI agents kept leaving "zombie" Chromium processes running in the background, completely hoarding our RAM and locking up our ports. We were using heavy testing frameworks (like Playwright) that just weren't meant for asynchronous agentic workflows.

Before we went down to the bare-metal Linux level to fix it, we had agents hallucinating CSS selectors and getting shadow-banned by Cloudflare.

What’s your worst horror story with agents trying to browse the web? Curious to hear what workarounds you're currently using!