When Perplexity uses Kernel's browser infrastructure instead of building it themselves, watch what that means. Companies with world-class engineering teams are choosing to pay for browser automation rather than treat it as product differentiation. That choice reveals where complexity actually lives in agent systems and where the next infrastructure categories will emerge.

Kernel's $22 million Series A this month isn't remarkable because browser automation is novel. It's remarkable because it's becoming its own funded category. Browserbase raised $6.5 million in May 2024 for similar infrastructure. Multiple companies are betting that "make agents interact with websites at scale" is a problem worth solving once, centrally, rather than repeatedly by every team building agents.

This matters for anyone architecting agent systems. The build-versus-buy calculus has shifted. Five years ago, you'd build your own browser automation because the problem seemed tractable: spin up Puppeteer, write some scrapers, handle a few edge cases. Today, that "few edge cases" includes session persistence across thousands of concurrent workflows, authentication flows that expect human behavior, CAPTCHA solving at scale, browser fingerprinting that needs to look organic, and anti-bot measures that evolve weekly.

The pattern here mirrors earlier infrastructure waves. Observability started as "add some logging." Then it became Datadog. Authentication started as "hash some passwords." Then it became Auth0. A technical problem becomes specialized enough that building it yourself stops differentiating your product, and suddenly it's infrastructure someone else should handle.

Browser automation has crossed that threshold. The complexity isn't in the concept. It's in the accumulated knowledge of how to make programmatic access work reliably when websites actively resist it. That knowledge is worth paying for because the alternative is dedicating engineering cycles to debugging why your agent got blocked by Cloudflare instead of improving your core product.

What this predicts: more agent infrastructure categories will emerge. If browser automation is too complex to build yourself, what about workflow orchestration across multiple agents? What about handling failures gracefully when the web doesn't cooperate? What about auditing agent decisions at scale? Each of these feels like a feature today. Each will become infrastructure tomorrow.

The defensibility here comes from operational excellence rather than proprietary technology. The techniques for evading bot detection are well-known. What matters is maintaining browser fleets that work reliably, updating fingerprinting strategies as detection evolves, handling edge cases across thousands of different websites. That's the kind of defensibility that comes from doing something hard, repeatedly, at scale.

Teams building agents face a straightforward choice about what they're actually buying. You're buying focus. You're buying the ability to treat "interact with any website" as a solved problem so you can concentrate on the reasoning and workflow orchestration that actually differentiates your product. You're buying accumulated expertise in a specialized discipline that doesn't advance your core mission.

Browser automation at scale is infrastructure, not product. The companies that recognize this early will ship faster. The ones that insist on building it themselves will spend months debugging session management instead of improving their agents' capabilities.