Agents will use the internet - Navigate 2026 Keynote

0:14 Please welcome founder and CEO of Browserbase, Paul Klene. >> >> Welcome to Navigate, our first conference gathering some of the best teams in AI discovering what's possible on the frontier of intelligence. In this room, we have our customers, our partners, our investors, and our team.

0:45 And I'm so thankful to have you all here today. This might be the biggest day of my life outside of my wedding. So, we're all in this together, and it's going to be a lot of fun. Over the past 24 months, we've seen so many interesting things happen from our customers building with browserbase. And today, I want to share a state of the web because at browserbase, we have one core belief. In the next 12 months, every task humans do will be assisted by AI. And we're seeing that already on the Twitter timeline or in the hackathons. People are pushing the boundaries more and more and unlocking new cases that we didn't think were possible. But we are just getting started because you see in the next 12 months every industry is going to benefit from AI being able to access the entire web. But there's still so much work to be done and we have to kind of wonder like why is this not ubiquitous yet? Why is AI not completely revolutionized all of our jobs? It might be able to book me a reservation but why is it not doing my taxes? Well, we think it comes down to a couple key things.

1:52 You see, intelligence isn't faster than the world can put it to work. And we're seeing models solve every single type of task in isolation, but that's not being translated to real world impact that I'm experiencing in my day-to-day lives. And even at browserbase, we've actually saturated some of our tasks on Frontier model in sandboxes. But then when we see our agent go out into the real world web, it gets a little more complicated. Basically, the models are ready, but the web isn't. And this is a capabilities overhang that we've been talking about all the time in this age. It doesn't apply explicitly to the web, but that's the area that we're focused on the most here at Browserbase. And there's a reason why. You see, humans do over 4.45 trillion hours of browsing every single year. And that's across a range of countries and tasks and socioeconomic status. It is one of the most ubiquitous things that we do in our day-to-day lives, looking at screens. I want to do less of that and maybe AI can help us.

2:52 But you see, the web was built by humans and for humans, not for AI. And translating these human interfaces into AI ready interfaces, is a pretty big challenge. But thankfully, we've seen this problem before because when you look outside on the roads of San Francisco, you see AI driving alongside humans right now using cars just like the ones that we do. Whimo didn't have to rebuild the roads for AI. They just taught it how to drive like a human by giving it a car that looked like ours and the rules of the road that we follow as humans, too. We believe the same thing applies to AI using the web. We won't have to rebuild the web for AI. Instead, we can teach it to work with the web that we already have. Now, there's a lot of work to actually make that happen, and we'll we'll talk about that through the rest of this conversation, but I kind of want to end with one more thing here, which is that we believe at browserbase, we're not just an infrastructure company.

3:49 We're a diffusion company. And the fastest way to get AI into the tasks that we experience every single day is by giving it the same capabilities that human have to actually do these tasks. If AI can browse the web, it can do so many more tasks right now. That's what we're doing at browserbase. And we've made a lot of progress so far. You see, over the past few years, we've run 500 million sessions on our platform across four different geographies and touching so much of the web. When you add all of that together, that's 2,000 human years of browsing. 2,000 years just deleted from the world that a human would have otherwise had to do. That's so inspiring to me because we're only a team of 75 strong. To have that impact of automating away lifetimes worth of work just gets me out of bed every morning.

4:39 But we've only actually touched 7% of the active web and I keep thinking about what about the other 93%. What are the other tasks and things that we can go do online that we haven't seen yet at browserbase? And it kind of begs the question, what capabilities does AI need to unlock the rest of that web? That other 93% that I keep thinking about every single day. Well, let's take a flashback. Let's think about the web that we had so far. The ugly web, the nostalgic web, Neopets in MySpace, those are not built for AI, right? How do we actually work with this web? This web that might not ever change. There might not be developers maintaining this web, but there's an incredible source of information of the existing internet that we already have. We shouldn't just throw this web away. We need to give AI a way to access it. And we've assembled the right prim to make that happen in the browserbased platform. We've taken every single tool that your agent needs to access both the new and old web, the legacy and the modern, the ReactJS or the AngularJS.

5:41 All of that can be accessed with browser bases infrastructure platform. But we started with one tool, one tool that I care a lot about, which is the browser. You see, I've spent most of my career getting browsers to run on servers. And the browser that you run on your laptop is pretty similar to the run that we run in our cloud. But the browser that runs your laptop kind of sucks. I mean, it uses all of my memory, all of my CPU. To actually scale that up into cloud infrastructure is extremely challenging.

6:07 You need to build all these features like file upload and download, screenshot support, recordings and replays and then run it at web scale. Well, our browserbased infrastructure platform is one of the largest atscale infrastructure platforms out there. And we've now cut our hands so many times that you can see the scars so you don't have to experience those pains when you're building your own browser infrastructure products. But once your agent can browse, it might need to be able to find out where to go. That's why early this year we launched our search API. Our search API takes a single query and returns a list of URLs to help your agent know where to start. This is an essential complement to the browser because the browser helps you interact while the search API helps you know where to go. But then what if you're searching and you don't necessarily need to interact with the page? You just want some data off of that. Well, that's where our fetch API comes in. Our fetch API allows you to give a URL and extract information in a structured format. And the cool part about it is that it actually uses the browser infrastructure we have underneath. We found our fetch API to be more reliable, performant, and accurate than other providers out there because it benefits from all the investments we've made in our browser core. The browser is the primitive. If you have a great browser, you can interact with the web the same way people do, but you need a way to control that browser. That's where we looked upwards into the open source world and found out there wasn't a good solution.

7:32 So, Stage Hand is our open-source browser automation framework. I've spent a lot of my career working with things like Puppeteer, Playright, and Selenium. Open-source frameworks to enable programmatic control of a browser that often fall short in AI workloads because they weren't built for AI. They were built for humans writing tests. So, with stage, we went from the ground up. We thought about how do we build in open source so the world can benefit from a new control plane for browsers. And we've seen some incredible results.

8:02 Stage hands now downloaded 1.4 million times per week. It is actually 80% more token efficient than Playright and two times as fast. And once again, it's all open source. You can use it with any browser. You can use it locally, too. I think Stage Chan is criminally underrated. But we built it because we wanted our customers to have the best way to control a browser with code, either writing it as a human or AI generating it. And Stage was the biggest bottleneck. We're kind of like raptoring, continuing to improve our infrastructure, find the next bottleneck and make it better. Stage hand was one of those. But then once you have the code to control a browser, once you have the primitives, the APIs, you need the intelligence layer. So we launched a model gateway product. And the special part about our model gateway product is that it's optimized for the domain of browsing the web. So with browser's model gateway, we can help intelligently choose the model that makes sense for your task. If you're just browsing hacker news, you probably can get away with a pretty dumb model. But if you're doing your taxes or interacting with iframes, you might want a more sophisticated complex model. Our model gateway can route that for you. And if you're taking actions multiple times, maybe you don't need to use inference.

9:09 Again, tokens are getting really damn expensive. Our model gateway can cache the actions you take on the web using a proprietary caching strategy that actually won't blow up the cache keys if the page subtly changes. This enables cache reliability specialized for the domain of web browsing. And I'm really proud of the growth of this product so far. But agents are doing more than just calling APIs or interacting with pages. They're also writing code. And that code needs a place to run. Our browserbased functions product is a sandbox for your agents to deploy code to. And what's special about this is that it's extremely performant because the code runs next to the browser, not far away.

9:46 Roundtrip latency from your code to the browser and back is the bane of performance. And if we can colllocate that code alongside the browser, we can let your agents run code securely but also performantly. All of these come together to give you the primitives to build best-in-class AI agents. And there's a lot of companies building best-in-class AI agents on browserbased today. I want to talk about three of them. And we're actually customers of these too. So, RAMP is one of my favorite software products I use out there because when I use RAMP, I don't have to think about receipts or finances. I get emails of my receipts all the time, but oftent times there's receipts that don't show up in my email.

10:23 Receipts I have to go to some weird portal to go get or have to go use my confirmation code to pull it from my flight. Well, RAM's mission is to save time and money for their customers. So, how can we save them more time? Well, let's actually use an agent to go take a browser and get those receipts that you might have to go get manually. This has been incredibly impactful for the team because they've now done 5 million receipts every single month using their agent product powered by browserbase saving an ton of time for their customers and making my life a little bit easier. Checker has been using browserbase to actually help verify identities faster. You see they previously had a legacy playright implementation and a whole team maintaining it. But with browserbase, they were actually able to move their infrastructure completely over to our product and let their teams spend more time making sure that their customers are onboarding people in a safe way as opposed to fixing broken connectors.

11:13 When you're modernizing your web browsing stack, browser base is one of the choices our customers make. And we're incredibly proud to be a customer of Checker, but also, you know, powering them ourselves. And this last use case is especially important to me because my wife is here and she's a nurse practitioner. And I see how often that she spends time actually doing paperwork and filling in charts or talking to insurance companies versus helping providers help provide healthcare services. And Commir uses browser base actually to help replace that. So they're using a browser agent to go to an insurance website actually integrate and handle the form filling and all that painful stuff so providers can spend more time with patients and do less paperwork. This is incredibly meaningful and these vast amount of use cases kind of show how many different industries are going to benefit from products like browserbased and increases in model capabilities. We think that we can make the world better by letting people do the stuff they actually should integrating with systems or grab grabbing PDFs and dragging them around.

12:10 That's why we're doing this. But there's still more work to do because you don't need to be an AI expert to actually build this yourself. We think that a lot of these teams are sophisticated in AI. We want everyone to be able to build these types of agents to touch more use cases. That's why we built the browserbased agents product. It's a managed agent with all of our primitives built in. And we can help you pick the right tools, the right model, and actually will optimize over time, continuing to get better by looking at the traces it's generated. And this is a very convenient product where customers can use it in one prompt. But it's also best-in-class. You see, with a great harness, browserbased agents is able to match frontier level intelligence at half the cost because it's not just using inference every time. It's using the caching and the routing and the token efficient primitives. And we think we're going to continue to match frontier level intelligence and continue to reduce costs because you shouldn't be spending so much on tokens. You should be spending more time with your customers. But there's still so much of the web that 93% we can do it efficiently, but how do we access more of it? We want to make sure that all these agents that you're building are first class citizens of the web. Because another history lesson, for the longest time, there were bad bots and the web now can't differentiate between bad bots and good agents. They blow them up just like that. We want to find a way to help the web know who is trusted, know who should be able to use their website, know what where that agent is coming from. This is an important problem that requires a coalition of companies to solve it. And that's why we partner with some of the best This new technology called webbot off allows us as browser base to sign our agents say where they're coming from and companies like Versell recapture and cloudflare are working with us to help identify and accept our traffic from our verified agents to make sure that they are trusted on the web and not treated like bad bots. And you can kind of think about webbot off almost like an identification card for your agent where us at browserbase are issuing that to you and spending all the time on your behalf to continue to get more websites to adopt it, integrate with it and trust it. We think webbot off is one of the most important things to solve to enable AI to access the entire web. And we're so thankful for some of our partners in the room who are actually pushing that forward. But what if your agents can like access the website? They still need to log in as you. And that's a whole another class of problems because when you think about authentication, most systems on the internet actually live behind authentication. The login page is often the next gate to allow your agent to actually take an action online. So we think about all the sensitive things that are required for that. Your passwords, your sessions, your 2FA, your one-time password, your cookies, your location. All these things are necessary to let agents to log in. And it's a pretty sensitive thing. We can't solve this problem in isolation at browserbase. We need to work again with partners like we did for webbot off.

14:59 That's why we've been partnering with teams at octa clerk and one password to actually rethink how agents should integrate and authenticate on the web because we're only one side of the equation. We're the agents. We also need to work with the greater websites out there to accept and welcome agents in and give them the right guardrails to access these systems reliably and securely. All in all, we're bundling this together in this thing we call the browserbased agentic identity suite. We can manage your secrets. We can give your agents an email. We can help securely store your cookies and even manage the authentication connection.

15:34 Our goal is that you don't have to think about identity and authentication ever again for your agents. Because we believe at browserbase that if a human can do a task, an agent should also be able to do that task because agents are delegates on our behalf. And if we can do this responsibly with the right tracking, trust, and guard rails, we believe agents will tackle more and more of the work that we have. And that's especially important because you see, like I said earlier, we're not just an infrastructure company here at Browserbase. We're in the business of accelerating diffusion. And we want to help take AI and bring it to the web we already have. Because if we don't do this, AI will only be able to integrate with people who build APIs and MCPs. And it could be a walled garden of, you know, gatekeepers and tollkeepers saying who's allowed to access what. We want to arm the rebels at browser base and let you work with the entire web. work just like people do because if we don't, we may leave entire communities and the real economy behind because I know the DMV, my local barberh shop, or all these other small businesses won't be launching an MCP or API anytime soon, but they should benefit from your agents transacting with them, working with them, or using them to actually further the goals of the people controlling those agents. And furthermore, we think that this is a better way overall because we think that our customers are the ones that are actually going to do a lot of this work because we can't do this without you. You're the ones that are bringing forward the use cases or the techniques or the ideas for more tasks to help people with. If we all want to solve this diffusion mission together, we need you to come to us and talk to us about what you want to go solve and who you're working with. I meet with companies all over the world.

17:19 the lumber factory in Mexico, the shipping terminal in Singapore, all of them want to make people's lives easier using AI agents, but they don't know how to start. But we think we can do it together because we want to build a future where human and agents work alongside each other. And if we're able to do this well, we can unlock so much growth and productivity for the world and save us all a lot more time instead of looking at screens, actually doing the things that are important for us, which is spending time with people, being creative, building community. AI is going to take away the boring work.

17:52 And it's our job in this room to make that happen. So today at Navigate, I'm extremely excited to have a full lineup of speakers where we're going to talk about how we bring these challenges forward and tackle them. And my call to action for all of you is to think ambitiously because everything you think can be possible with AI is now possible today as long as we build it together. Thank you all so much.

18:23 Perfect.