Browserbase gets AI Agents on the Web 🌐

Plus: Founder Paul Klein on why he believes AI-powered browsing is only just beginning...

February 20, 2025

CV Deep Dive

Today, we’re talking with Paul Klein, founder and CEO of Browserbase.

Browserbase offers a powerful cloud-based headless browser infrastructure designed for both AI-driven and traditional automation workflows. If an application needs to navigate the web like a human would—filling forms, clicking buttons, interacting with pages—Browserbase provides the underlying infrastructure so you don’t have to maintain your own fleet of headless browsers.

Paul’s experiences working at Twilio + Mux shapes his belief that infrastructure companies are uniquely powerful, offering the “building blocks” that power countless use cases.

After raising their seed round in June 2024, Browserbase raised a $21M Series A in October 2024 co-led by Kleiner Perkins and CRV with participation from Okta Ventures. 

Key Takeaways

  • Web Browser for AI: Browserbase is building the infrastructure that lets AI agents (and any automation) interact with websites as though they were human users.

  • Stagehand, the AI Web Browsing Framework: An extension of Playwright that allows natural-language instructions for web actions, making generative browser automation seamless.

  • Not Just AI: While AI automation is a major driver, Browserbase also supports any headless browser workflows for testing, web scraping, or automation.

  • Hands-off Scalability: The team has scaled to tens of thousands of browser sessions, handling tough Kubernetes and cloud infrastructure challenges so customers don’t have to.

In this conversation, Paul explains how Browserbase came to be, how it fits into the rapidly expanding web automation landscape, and why he believes AI-powered browsing is only just beginning.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Paul 💬

Welcome Paul - give us a bit of background on yourself and what led you to found Browserbase?

I’ve been working for infrastructure companies basically my whole career. I actually started out as an intern at Twilio during their IPO, which was a wild experience because everyone around me seemed to be getting rich overnight, and we could drink for free at the bar. I ended up staying for three years and watched Twilio go from about $100 million in revenue to over a billion, which really convinced me that building developer tools and infrastructure is incredibly powerful—giving devs the “Lego blocks” to build anything.

After Twilio, I co-founded Stream Club to create a browser-based live-streaming platform during COVID. Existing livestream software felt clunky, and we wanted to make it dead simple for creators to stream right from their browsers. We ran it for a couple of years, then sold to Mux, another infrastructure company that provides video APIs. It made sense because we constantly ran into video infrastructure headaches, and Mux solved those at scale. All of that led me to realize I get the most joy from building these horizontal platforms. Twilio does it for communications, Stripe does it for payments, and I wanted to do that for browsers—so I started Browserbase.

Can you give me an elevator pitch for Browserbase for the uninitiated AI developer.

The simplest way to explain it is: humans use browsers to go to websites and fill out forms or click on buttons. In the world of AI, we believe a lot of those routine tasks will be handled by AI on our behalf. But the browsers we use every day aren’t designed for AI; they’re designed for people on laptops. AI runs on servers in the cloud, so it needs something different.

Browserbase is building the infrastructure layer that runs thousands of browsers in the cloud. On top of that, we have Stagehand, which is the framework for telling a browser what to do using natural language. If you say “Click the Buy button” in plain English, Stagehand translates that into code that can drive a headless browser. That’s a perfect match for LLM-based agents because they speak and interpret natural language, so we’re essentially giving AI its own specialized browser to interact with the web.

I see, so you’re not an alternative to automation frameworks like Selenium or Playwright—you’re a complementary tool?

Exactly. Selenium, Playwright, and Puppeteer are automation frameworks that let you script browser actions, but you still have to run the actual browser somewhere. If you’re using Selenium or Playwright on your own, you might spin up containers in the cloud, handle scaling, logging, debugging—it can get complicated. Browserbase is that managed environment, so you can bring your Selenium, Puppeteer, or Playwright scripts, and we’ll run them at scale. Or, if you prefer, you can use Stagehand, which is our own extension of Playwright that’s built for AI agents.

We also record each session so you can watch what happened if something breaks. We log console output, network traffic, all those details that help you see why a script failed. That level of observability is essential when you’re running thousands of browsers, because the web is messy and unpredictable.

Who are your customers today, and what kind of people are finding the most value in Browserbase?

It’s a mix of AI startups and more traditional automation use cases. We have a lot of new AI companies building agents that need to do multi-step tasks on websites. For example, you might have an “AI SDR” tool that logs into LinkedIn, searches for prospects, and sends connection requests. Or you might have a workflow automation app in healthcare or insurance that logs into complex portals. All of that needs robust headless browsing.

Then we also see customers who are not explicitly AI-driven but have legacy scripts for web testing or scraping. They used to maintain their own browser fleet but realized how much time that eats up. For them, Browserbase is an infrastructure-as-a-service for browsers. So we end up powering everything from e-commerce automation to data collection. The common thread is: you need to run a browser reliably and at scale without spending half your engineering cycles on it.

A great success story is Structify, which built a web research agent that navigates sites lacking an API. The AI decides where to go, how to fill out forms, what to parse on each page, and that all happens on Browserbase. Without our infrastructure, they wouldn’t be able to spin up enough sessions or handle the complexity of real-world web flows at scale.

Another is Benny, which helps people on food stamps get rebates for healthy groceries. The normal process involves going to a painful web form, categorizing items, uploading receipts. Benny automates that on behalf of the user. It’s not even using AI —it’s just a specialized automation that Browserbase makes super simple to scale. They can run thousands of concurrent sessions without building their own headless environment. It’s a clear example of saving people real time and effort.

This is a really exciting and emerging market – how do you think about the competitive landscape, and how does Browserbase stand out?

I was a customer of older headless browser services and felt they never had the polish of a Vercel or a Stripe for browser automation. I saw the installation metrics for things like Playwright or Puppeteer going exponential—millions of downloads and growing fast. That signaled a real, underserved need. So we set out to build a best-in-class developer experience for headless browsers, with great logging, replay, reliability, and support for multiple frameworks.

We differentiate ourselves, frankly, by being a really great product. We handle concurrency, we’re stable, we offer detailed logs, session recordings, and we also have Stagehand specifically for AI use cases. It’s a wide net: people can do old-school statically-coded scripts, or they can do cutting-edge LLM-driven interactions. In both cases, you get a robust environment that scales and is easier to debug.

Web Agentic AI is accelerating fast. How do things like Operator or Anthropic’s “computer use” factor into Browserbase’s plan?
In many ways, they boost demand for Browserbase. Operator and Anthropic’s approach let an LLM call external tools to do complex tasks. If those tasks involve websites—clicking through, filling out forms, navigating multi-page flows—they still need a browser. Operator is basically a web agent living inside ChatGPT, and we power similar web agents. We released something called Open Operator as a reference project, to show you exactly how you can build something like Operator, but embedded in your product. Similarly, you can wire up Anthropic’s “computer use” to Browserbase so an AI can step through real websites.

The net effect is that as these agent frameworks get better, they need an infrastructure layer to power actual browsing. We’re not competing with them; we’re complementary. If Operator or Anthropic wanted to host a massive browser cluster themselves, that’s a different conversation, but so far, we see them as catalysts driving more interest in headless automation.

As Web Agentic AI becomes more common and AI Agents increasingly need to interact with websites, how do you feel the Internet and web dev might change in the long term?

It’s going to be a while before AI agents outnumber humans, because billions of people aren’t online yet. But yes, over time, more mundane web tasks will be offloaded to AI. That means an increasing volume of automated “users,” though I doubt websites will build entirely separate versions for bots. If the AI is sufficiently intelligent, it can work with the existing site just as a human does.

We’ll likely see better differentiation between “good” and “bad” bots. CAPTCHAs are crude—they can’t tell if you’re an agent doing legitimate tasks or someone trying to spam or scrape without permission. A “good” bot typically represents a real user, logging in with their credentials, performing tasks with consent, and contributing some value. A “bad” bot is the one scalping tickets, scraping data for shady purposes, or spamming. We’re working with authentication providers like Okta, Clerk, and Stytch to help authenticate beneficial bots so they can pass through without friction. In theory, that would reduce the arms race of AI vs. CAPTCHAs.

Could websites add more alt tags or specialized endpoints for AI? Possibly, but that might be incremental rather than a total revamp. Accessibility best practices already encourage labeling UI elements thoroughly, and a well-structured HTML page is easier for an agent to parse. I don’t expect a wholesale rewrite of the web—there’s too much legacy out there. Browserbase is built to handle the web as it is, so even if some sites make AI-friendly features, the rest won’t change overnight.

How will Browserbase evolve over the next 12 months?

We’re doubling down on Stagehand, which has been TypeScript-only but is getting a Python version soon. A lot of AI devs are Python-focused, so that will be a big milestone. We’re also rolling out a visual web interface for Stagehand, kind of a WYSIWYG for building automation scripts. That should help people prototype or debug browser flows quickly.

We plan to offer more customization for things like cost vs. performance trade-offs. Some customers want extremely cheap concurrency for massive scale; others need bulletproof reliability or specific rendering requirements. Because we’re usage-based, we only succeed if you succeed. So we’re building more knobs you can turn to dial in your exact needs.

Overall, we’ll keep adding features that make cloud browser infrastructure easier—more logging, more debugging, better concurrency. Our goal is to remove every obstacle so teams can launch new products faster without wrestling with the low-level details of running browsers.

You guys seem to be on quite a rocket ship—seed round last summer, Series A by fall. How have you managed that insanity?

Rocket ships are exciting and chaotic at the same time. From the outside, it looks amazing, but inside, it can feel like everything’s on fire. I’m grateful, though. We’ve gone from one to fifteen team members, and we plan to be around thirty soon. I also got married last year and started focusing on my fitness again, which is a lot to juggle.

A lot of it was mapped out in this memo I wrote called “An Internet Browser for AI.” Things have lined up nicely, partly because the AI space is moving at breakneck speed. That can be overwhelming, but it’s also an opportunity. We’ve raised capital to scale quickly to meet demand, and if our customers flourish, our usage-based model means we grow too. We win when our customers win.

What’s the culture like at Browserbase? Obviously you’re hiring—what kind of environment are you building?

We’re fully in person, five days a week, which is unusual post-COVID, but we love the energy. We run together on Mondays, share ideas in real time, and basically have a hacker-house vibe, but more organized. It’s a place where engineers geek out over dev tools and infrastructure challenges.

We emphasize personal growth, so I once asked at an all-hands: “Who’s done something they never thought they could do in the last six months?” Everyone raised their hand. That’s what we want. People step outside their comfort zone and learn rapidly, whether it’s mastering Kubernetes scaling or building open-source tooling. Many of us are former founders or might become founders in the future. We share that passion for building real products people love.

If you’re someone who’s excited about dev tools and wants to solve challenging infra problems, we’re looking for you. It’s a rocket ship, but we try to keep it fun and supportive. We want a place where the slope of your learning curve is as steep as possible.

Is there anything else you'd like people to know about Browserbase?

If you’re building AI (or any) automation that needs to interact with websites, fill out forms, or replicate user actions, you should probably check out Browserbase. Instead of reinventing the wheel with your own headless fleet, let us handle the infrastructure. That way, you can focus on building a great product rather than debugging Chromium at scale.

Feel free to email me directly at [email protected] if you’re working on something interesting. We love supporting innovators—especially those using AI to do things that aren’t possible with regular APIs or static scripts. There’s a huge opportunity in letting AI act on the web, and we’re here to power that.

AI is definitely not zero-sum, so let’s help each other succeed. If we can empower you to ship faster and tackle the hard parts of headless browsing, everybody wins.

Conclusion

Stay up to date on the latest with Browserbase, learn more about them here.

Read our past few Deep Dives below:

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.