Paul Klein IV Couldn't Get an Internship. So He Built the Browser Infrastructure Keeping AI Agents Online.

Paul Klein IV has been telling the story of his 500 rejected internship applications in a video interview now making rounds across developer communities. Depending on who's watching, it lands differently — founders hear a permission slip, engineers hear an origin story for a product they're probably already using. Browserbase, the company Klein built from that stretch of rejection, was last valued at approximately $300 million.

The company sells a specific kind of infrastructure: cloud-hosted browser sessions, available via API, designed for developers and AI agents that need to interact with live websites rather than clean, well-documented endpoints. That sounds narrow until you spend any time with web automation at scale. Websites detect and block automated traffic. CAPTCHAs appear at inconvenient moments. Sessions time out. Proxies need rotating. Fingerprints need spoofing. Browserbase absorbs all of that — so that agents built on top of it can focus on the actual task rather than the plumbing underneath it.

Browserbase has also shipped Stagehand, an open-source framework layered on top of Playwright that lets large language models navigate web pages using natural language. Rather than requiring agents to be programmed for specific page layouts, Stagehand lets them read a page and figure out what to click — which means agent workflows can survive website redesigns without breaking. Infrastructure companies that also ship frameworks tend to gain a stickiness that pure API providers don't, and that seems to be the logic here.

Browserbase sits in a cluster of startups building what investors are calling foundational primitives for autonomous agents. E2B handles sandboxed code execution; Tavily handles search; Browserbase handles the browser. The argument common to all three is that agents need reliable, scalable access to capabilities that look simple from the outside but aren't — and that managing them well is a real business.

Whether it's a $300 million business depends on how many agent workflows end up needing persistent, live web access rather than cached data or prebuilt API connectors. Right now, the trajectory suggests quite a few. For Klein, it's a long way from email number 500.