Bringing agents onto the world wide web — Paul Klein IV, Browserbase

>> Hello. >> Hello. Very sleepy crowd in the computer use Very sleepy crowd in the computer use Very sleepy crowd in the computer use room. Have we all given up at this room. Have we all given up at this room. Have we all given up at this point? Like, what's going on? Uh thank point? Like, what's going on? Uh thank point? Like, what's going on? Uh thank you for coming in to my talk. My name is you for coming in to my talk. My name is you for coming in to my talk. My name is Paul Klein. I'm the founder of Paul Klein. I'm the founder of Paul Klein. I'm the founder of Browserbase, and I'm going to talk about Browserbase, and I'm going to talk about Browserbase, and I'm going to talk about bringing agents onto the World Wide Web. bringing agents onto the World Wide Web. bringing agents onto the World Wide Web. If you're in this audience in this If you're in this audience in this If you're in this audience in this track, you've done computer use, you track, you've done computer use, you track, you've done computer use, you tried operator when it came out, and tried operator when it came out, and tried operator when it came out, and you're probably like, "Why isn't this you're probably like, "Why isn't this you're probably like, "Why isn't this happening yet? This this seems obvious." happening yet? This this seems obvious." happening yet? This this seems obvious." Well, we'll address some of the Well, we'll address some of the Well, we'll address some of the high-level needs of computer use high-level needs of computer use high-level needs of computer use to really serve what I think is the to really serve what I think is the to really serve what I think is the largest category of AI agents, the largest category of AI agents, the largest category of AI agents, the agents that actually go out and do work agents that actually go out and do work agents that actually go out and do work on your behalf in the real world. We'll on your behalf in the real world. We'll on your behalf in the real world. We'll talk through some of the technical talk through some of the technical talk through some of the technical stuff, but really trying to focus on stuff, but really trying to focus on stuff, but really trying to focus on there's a huge model capabilities there's a huge model capabilities there's a huge model capabilities overhang in this category specifically, overhang in this category specifically, overhang in this category specifically, and you all here can hopefully solve it. and you all here can hopefully solve it. and you all here can hopefully solve it. So, thank you for coming. So, thank you for coming. So, thank you for coming. Well, of course, as we all know, the web Well, of course, as we all know, the web Well, of course, as we all know, the web wasn't built for agents. It was built wasn't built for agents. It was built wasn't built for agents. It was built for people, and that becomes very for people, and that becomes very for people, and that becomes very challenging as we're building systems to challenging as we're building systems to challenging as we're building systems to try and interact with it and automate try and interact with it and automate try and interact with it and automate it. So, when we're thinking about it. So, when we're thinking about it. So, when we're thinking about building agents that interact with building agents that interact with building agents that interact with people for systems, we have to really people for systems, we have to really people for systems, we have to really wonder, "Why wasn't it, you know, why wonder, "Why wasn't it, you know, why wonder, "Why wasn't it, you know, why was it built for us? And and why is it was it built for us? And and why is it was it built for us? And and why is it struggle?"

struggle?" struggle?" When you've done any sort of automation When you've done any sort of automation When you've done any sort of automation in the past, you've run into so many in the past, you've run into so many in the past, you've run into so many roadblocks. You know, the pages change, roadblocks. You know, the pages change, roadblocks. You know, the pages change, the web was built in a very context the web was built in a very context the web was built in a very context inefficient way. It's a lot of text, a inefficient way. It's a lot of text, a inefficient way. It's a lot of text, a lot of tokens. And when you're running lot of tokens. And when you're running lot of tokens. And when you're running any sort of browser agent or web agent any sort of browser agent or web agent any sort of browser agent or web agent right now, you get a broken browser that right now, you get a broken browser that right now, you get a broken browser that doesn't spin up. You know, you have doesn't spin up. You know, you have doesn't spin up. You know, you have pages that don't work. You have blockers pages that don't work. You have blockers pages that don't work. You have blockers or other sort of problems that really or other sort of problems that really or other sort of problems that really limit you. And I actually started my limit you. And I actually started my limit you. And I actually started my career doing web automation, maintaining career doing web automation, maintaining career doing web automation, maintaining these scripts every single day. It was these scripts every single day. It was these scripts every single day. It was very painful. So, in my world, agents very painful. So, in my world, agents very painful. So, in my world, agents have made a huge advancement and allowed have made a huge advancement and allowed have made a huge advancement and allowed me to write durable web automation me to write durable web automation me to write durable web automation scripts, but we still haven't gone to scripts, but we still haven't gone to scripts, but we still haven't gone to agents yet. agents yet. agents yet. And the question I want to ask is why? And the question I want to ask is why? And the question I want to ask is why? We We We we're sitting here in this room we're we're sitting here in this room we're we're sitting here in this room we're thinking computer use we saw a year ago thinking computer use we saw a year ago thinking computer use we saw a year ago has progress stalled? Like why why are has progress stalled? Like why why are has progress stalled? Like why why are web agents and browser extensions not as web agents and browser extensions not as web agents and browser extensions not as big as it could be? And I think it's big as it could be? And I think it's big as it could be? And I think it's really a comes down to a few things. really a comes down to a few things. really a comes down to a few things. You know, until recently the the bottom You know, until recently the the bottom You know, until recently the the bottom neck was the models. The models one year neck was the models. The models one year neck was the models. The models one year ago really weren't good at long context ago really weren't good at long context ago really weren't good at long context horizon tasks. But that's clearly been horizon tasks. But that's clearly been horizon tasks. But that's clearly been you know, solved in a major way. Models you know, solved in a major way. Models you know, solved in a major way. Models can do more and more complex tasks than can do more and more complex tasks than can do more and more complex tasks than ever and of course in AI you always have ever and of course in AI you always have ever and of course in AI you always have to update your priors. It's clear to me to update your priors. It's clear to me to update your priors. It's clear to me that anything I believe 6 months ago I that anything I believe 6 months ago I that anything I believe 6 months ago I have to revisit every single week have to revisit every single week have to revisit every single week because these models are progressing at because these models are progressing at because these models are progressing at an insanely fast pace. So I don't think an insanely fast pace. So I don't think an insanely fast pace. So I don't think it's the models.

it's the models. it's the models. And especially models are now much And especially models are now much And especially models are now much better at using interfaces. You know, better at using interfaces. You know, better at using interfaces. You know, we've seen this kind of capabilities we've seen this kind of capabilities we've seen this kind of capabilities improvement in computer use models. improvement in computer use models. improvement in computer use models. In the last year a lot of investment was In the last year a lot of investment was In the last year a lot of investment was made in RL environments for coding. And made in RL environments for coding. And made in RL environments for coding. And in the last 6 months months just as much in the last 6 months months just as much in the last 6 months months just as much investment has been made in RL investment has been made in RL investment has been made in RL environments for computer use. And environments for computer use. And environments for computer use. And computer use models are getting better computer use models are getting better computer use models are getting better and you can see this in the evals. When and you can see this in the evals. When and you can see this in the evals. When you train things on human trajectories you train things on human trajectories you train things on human trajectories in RL environments that model our real in RL environments that model our real in RL environments that model our real world, the real web, you can make better world, the real web, you can make better world, the real web, you can make better models. So the models are getting there, models. So the models are getting there, models. So the models are getting there, I promise. I promise. I promise. But the kind of so okay, so the models But the kind of so okay, so the models But the kind of so okay, so the models are good. Why why do agents still are good. Why why do agents still are good. Why why do agents still struggle to use the web? What are the struggle to use the web? What are the struggle to use the web? What are the what are the problems here? If if it's a what are the problems here? If if it's a what are the problems here? If if it's a model problem, Docus says if the models model problem, Docus says if the models model problem, Docus says if the models were good enough diffusion would just were good enough diffusion would just were good enough diffusion would just happen. Uh there's still a lot of work happen. Uh there's still a lot of work happen. Uh there's still a lot of work to be done here. to be done here. to be done here. And and to me it's it's no longer just And and to me it's it's no longer just And and to me it's it's no longer just the models. I'd argue that agents are the models. I'd argue that agents are the models. I'd argue that agents are missing the right harness and tools. And missing the right harness and tools. And missing the right harness and tools. And if you aren't familiar with an agent if you aren't familiar with an agent if you aren't familiar with an agent harness, if you haven't been on Twitter harness, if you haven't been on Twitter harness, if you haven't been on Twitter in the last few weeks, it's the it's the in the last few weeks, it's the it's the in the last few weeks, it's the it's the scaffolding and systems around your scaffolding and systems around your scaffolding and systems around your model that enable it to actually model that enable it to actually model that enable it to actually interact with the world. Uh a lot of interact with the world. Uh a lot of interact with the world. Uh a lot of talks and talk about harness engineering talks and talk about harness engineering talks and talk about harness engineering we're going to spend a lot of time on we're going to spend a lot of time on we're going to spend a lot of time on today, but I really think you can invest today, but I really think you can invest today, but I really think you can invest a lot in a harness and get a lot more a lot in a harness and get a lot more a lot in a harness and get a lot more out of the models and extract that out of the models and extract that out of the models and extract that overhang out of the models.

overhang out of the models. overhang out of the models. You know, Karpathy actually tweeted this You know, Karpathy actually tweeted this You know, Karpathy actually tweeted this back in November 2023 and I thought it back in November 2023 and I thought it back in November 2023 and I thought it was just so forward-looking that this was just so forward-looking that this was just so forward-looking that this what he described, these systems around what he described, these systems around what he described, these systems around an LLM, it's the harness. It's It's the an LLM, it's the harness. It's It's the an LLM, it's the harness. It's It's the the tools that it can access. And if you the tools that it can access. And if you the tools that it can access. And if you fast-forward 3 years later, a lot of fast-forward 3 years later, a lot of fast-forward 3 years later, a lot of what we're doing every single day is what we're doing every single day is what we're doing every single day is building towards this. A code building towards this. A code building towards this. A code interpreter for the LLM, audio and video interpreter for the LLM, audio and video interpreter for the LLM, audio and video input like screenshots, a browser, and input like screenshots, a browser, and input like screenshots, a browser, and other LLMs as subagents. All of these other LLMs as subagents. All of these other LLMs as subagents. All of these principles have held true. So, if you're principles have held true. So, if you're principles have held true. So, if you're ever wondering what should I build next, ever wondering what should I build next, ever wondering what should I build next, just go look at Karpathy's old talks. just go look at Karpathy's old talks. just go look at Karpathy's old talks. He's a pretty good predictor of the He's a pretty good predictor of the He's a pretty good predictor of the future. future. future. And you know, applying this to coding, And you know, applying this to coding, And you know, applying this to coding, we know that harnesses work really, we know that harnesses work really, we know that harnesses work really, really well with coding. Uh on the the really well with coding. Uh on the the really well with coding. Uh on the the graphic on the right, you can see graphic on the right, you can see graphic on the right, you can see Factory when it compared to Claude Code Factory when it compared to Claude Code Factory when it compared to Claude Code using the same model but using their using the same model but using their using the same model but using their kind of custom uh harness. And it turns kind of custom uh harness. And it turns kind of custom uh harness. And it turns out when you build a harness optimized out when you build a harness optimized out when you build a harness optimized for the domain that your agent is for the domain that your agent is for the domain that your agent is operating in, it can actually achieve, operating in, it can actually achieve, operating in, it can actually achieve, you know, above model results in that you know, above model results in that you know, above model results in that domain. Harness engineering is a real domain. Harness engineering is a real domain. Harness engineering is a real thing. And I'd say Cursor actually thing. And I'd say Cursor actually thing. And I'd say Cursor actually started this. Cursor was the first one started this. Cursor was the first one started this. Cursor was the first one that was doing a model engineering or that was doing a model engineering or that was doing a model engineering or harness engineering on top of the harness engineering on top of the harness engineering on top of the original LLMs. And a lot of what we've original LLMs. And a lot of what we've original LLMs. And a lot of what we've done at Browserbase with browser models done at Browserbase with browser models done at Browserbase with browser models has been, you know, harness engineering.

has been, you know, harness engineering. has been, you know, harness engineering. But and I think that like building a But and I think that like building a But and I think that like building a good harness is an engineering problem. good harness is an engineering problem. good harness is an engineering problem. You don't have to be a lab to build a You don't have to be a lab to build a You don't have to be a lab to build a good harness. And And a lot of us in the good harness. And And a lot of us in the good harness. And And a lot of us in the room maybe aren't working at labs. Your room maybe aren't working at labs. Your room maybe aren't working at labs. Your company can make a great harness for company can make a great harness for company can make a great harness for your domain and actually improve model your domain and actually improve model your domain and actually improve model results. You don't just have to wait for results. You don't just have to wait for results. You don't just have to wait for the models to catch up. And once again, the models to catch up. And once again, the models to catch up. And once again, I believe the models are quite capable I believe the models are quite capable I believe the models are quite capable now. now. now. And And if you look at this, you can see And And if you look at this, you can see And And if you look at this, you can see that it's not just Cursor, it's not just that it's not just Cursor, it's not just that it's not just Cursor, it's not just Factory. You know, many, many different Factory. You know, many, many different Factory. You know, many, many different types of companies are building coding types of companies are building coding types of companies are building coding harnesses on top of models and over harnesses on top of models and over harnesses on top of models and over performing on the models' capabilities. performing on the models' capabilities. performing on the models' capabilities. Now, it's not clear yet if custom Now, it's not clear yet if custom Now, it's not clear yet if custom harnesses are going to beat out durable, harnesses are going to beat out durable, harnesses are going to beat out durable, you know, RL'd models. Uh but we're not you know, RL'd models. Uh but we're not you know, RL'd models. Uh but we're not going to debate that today. We know that going to debate that today. We know that going to debate that today. We know that adding a harness on a model improves adding a harness on a model improves adding a harness on a model improves results, whether or not, you know, results, whether or not, you know, results, whether or not, you know, Claude Code will be the best harness Claude Code will be the best harness Claude Code will be the best harness ever or not. I think that's a different ever or not. I think that's a different ever or not. I think that's a different conversation. But you should still have conversation. But you should still have conversation. But you should still have some sort of harness on your model and some sort of harness on your model and some sort of harness on your model and measure the performance versus baseline measure the performance versus baseline measure the performance versus baseline model. model. model. And what I really want to get back to is And what I really want to get back to is And what I really want to get back to is that there is a massive capabilities that there is a massive capabilities that there is a massive capabilities overhang in computer use. The models are overhang in computer use. The models are overhang in computer use. The models are good enough, but we haven't done the good enough, but we haven't done the good enough, but we haven't done the engineering work to solve it. And I love engineering work to solve it. And I love engineering work to solve it. And I love this great Brokman tweet where he says, this great Brokman tweet where he says, this great Brokman tweet where he says, "Whenever I don't use Codex for a task, "Whenever I don't use Codex for a task, "Whenever I don't use Codex for a task, I I ask myself why?" And it feels like I I ask myself why?" And it feels like I I ask myself why?" And it feels like the task is an outside of the the task is an outside of the the task is an outside of the capabilities of the model. The overhang capabilities of the model. The overhang capabilities of the model. The overhang is there. The actual work we can do is is there. The actual work we can do is is there. The actual work we can do is missing. And when you look at the amount missing. And when you look at the amount missing. And when you look at the amount of like task completion you can get with of like task completion you can get with of like task completion you can get with coding, it's so much higher than COA coding, it's so much higher than COA coding, it's so much higher than COA because we haven't actually really because we haven't actually really because we haven't actually really pushed the models far enough and given pushed the models far enough and given pushed the models far enough and given it the right tools. So, to me it the right tools. So, to me it the right tools. So, to me not only is this important because I not only is this important because I not only is this important because I think that non-coding is a much bigger think that non-coding is a much bigger think that non-coding is a much bigger opportunity than it is coding. If you

opportunity than it is coding. If you opportunity than it is coding. If you look at this in recent slide, there's so look at this in recent slide, there's so look at this in recent slide, there's so many use cases that are in the many use cases that are in the many use cases that are in the non-coding domain that can benefit from non-coding domain that can benefit from non-coding domain that can benefit from computer use. It's a problem worth computer use. It's a problem worth computer use. It's a problem worth investing in, and the wrong answer is to investing in, and the wrong answer is to investing in, and the wrong answer is to sit around and just wait for the models sit around and just wait for the models sit around and just wait for the models to get better. You can actually solve to get better. You can actually solve to get better. You can actually solve this today. this today. this today. Solving overhang is an engineering Solving overhang is an engineering Solving overhang is an engineering problem, and this is the work that we problem, and this is the work that we problem, and this is the work that we can do within our companies and within can do within our companies and within can do within our companies and within our agents, especially within the our agents, especially within the our agents, especially within the computer use domain, to build reliable computer use domain, to build reliable computer use domain, to build reliable web agents. web agents. web agents. And when I think about browser agents And when I think about browser agents And when I think about browser agents that work, it it really comes down to that work, it it really comes down to that work, it it really comes down to three different types of things. They're three different types of things. They're three different types of things. They're multimodal, they're harness engineered, multimodal, they're harness engineered, multimodal, they're harness engineered, and they have reliable infrastructure. and they have reliable infrastructure. and they have reliable infrastructure. And I'll go through each of these three. And I'll go through each of these three. And I'll go through each of these three. First, they're multimodal. You no longer First, they're multimodal. You no longer First, they're multimodal. You no longer have to use a single model to actually have to use a single model to actually have to use a single model to actually interact with the task. And we see this interact with the task. And we see this interact with the task. And we see this with coding agents all the time. with coding agents all the time. with coding agents all the time. Sometimes you'll use a smarter model for Sometimes you'll use a smarter model for Sometimes you'll use a smarter model for a more complex page. Sometimes a a a more complex page. Sometimes a a a more complex page. Sometimes a a dumber model for a simpler page. And dumber model for a simpler page. And dumber model for a simpler page. And maybe you're using a combination of maybe you're using a combination of maybe you're using a combination of coding and computer use to actually coding and computer use to actually coding and computer use to actually power your agent. This is a really power your agent. This is a really power your agent. This is a really important insight. It turns out important insight. It turns out important insight. It turns out automating the web isn't always just automating the web isn't always just automating the web isn't always just clicking the button on the screen. It clicking the button on the screen. It clicking the button on the screen. It might be intercepting the network might be intercepting the network might be intercepting the network requests and writing a coding agent or requests and writing a coding agent or requests and writing a coding agent or having coding write a script to actually having coding write a script to actually having coding write a script to actually replay those network requests. The most replay those network requests. The most replay those network requests. The most reliable browser agents that we see in reliable browser agents that we see in reliable browser agents that we see in production right now are often writing production right now are often writing production right now are often writing code alongside using the browser to code alongside using the browser to code alongside using the browser to actually automate a task. If you've done actually automate a task. If you've done actually automate a task. If you've done any sort of personal automation work in any sort of personal automation work in any sort of personal automation work in your life, you might see Cloud code your life, you might see Cloud code your life, you might see Cloud code output a script more often than using, output a script more often than using, output a script more often than using, you know, Cloud in Chrome because that's you know, Cloud in Chrome because that's you know, Cloud in Chrome because that's a very context-efficient way to automate a very context-efficient way to automate a very context-efficient way to automate a repeatable task.

a repeatable task. a repeatable task. There's also harness engineering. It There's also harness engineering. It There's also harness engineering. It turns out that sure we can, you know, turns out that sure we can, you know, turns out that sure we can, you know, write scripts and use models, but doing write scripts and use models, but doing write scripts and use models, but doing these things repeatedly, you want to these things repeatedly, you want to these things repeatedly, you want to benefit from things like memory and benefit from things like memory and benefit from things like memory and skills. We launched something if we skills. We launched something if we skills. We launched something if we so-called browser.sh, which actually so-called browser.sh, which actually so-called browser.sh, which actually publishes skills for websites. So, publishes skills for websites. So, publishes skills for websites. So, before your agent even goes to the before your agent even goes to the before your agent even goes to the website, it can observe what types of website, it can observe what types of website, it can observe what types of tasks it can do. WebMCP is very useful tasks it can do. WebMCP is very useful tasks it can do. WebMCP is very useful for this. It's part of pulling in for this. It's part of pulling in for this. It's part of pulling in existing knowledge to optimize a existing knowledge to optimize a existing knowledge to optimize a website. Your agent doesn't have to website. Your agent doesn't have to website. Your agent doesn't have to discover something in the first place if discover something in the first place if discover something in the first place if it's done it before. It can use its it's done it before. It can use its it's done it before. It can use its memory and its skills to actually make memory and its skills to actually make memory and its skills to actually make it better. And you should think about it better. And you should think about it better. And you should think about trying to build in skills to your trying to build in skills to your trying to build in skills to your agents. If your agent is using CLIs to agents. If your agent is using CLIs to agents. If your agent is using CLIs to control websites like the Playwright control websites like the Playwright control websites like the Playwright CLI, you could actually give it skills CLI, you could actually give it skills CLI, you could actually give it skills and context to be more effective there. and context to be more effective there. and context to be more effective there. And this results in much more optimized And this results in much more optimized And this results in much more optimized token usage. As if you're throwing token usage. As if you're throwing token usage. As if you're throwing everything on the page to a model, everything on the page to a model, everything on the page to a model, you're going to get sub-par results and you're going to get sub-par results and you're going to get sub-par results and it's going to cost you a lot. The right it's going to cost you a lot. The right it's going to cost you a lot. The right harness should not only present the harness should not only present the harness should not only present the right tools, but present an optimized right tools, but present an optimized right tools, but present an optimized amount of tokens that are compressed to amount of tokens that are compressed to amount of tokens that are compressed to get exactly the right repeatable result get exactly the right repeatable result get exactly the right repeatable result every single time.

every single time. every single time. And finally, the infrastructure here is And finally, the infrastructure here is And finally, the infrastructure here is extremely important because when you're extremely important because when you're extremely important because when you're running browser agents in production, running browser agents in production, running browser agents in production, you want an environment that's going to you want an environment that's going to you want an environment that's going to work everywhere, every time. And I think work everywhere, every time. And I think work everywhere, every time. And I think a lot of work still needs to be done a lot of work still needs to be done a lot of work still needs to be done here. This is a lot of what our company here. This is a lot of what our company here. This is a lot of what our company does because computer use environments does because computer use environments does because computer use environments are pretty complex to scale up. You are pretty complex to scale up. You are pretty complex to scale up. You know, it's funny when Open Cloud came know, it's funny when Open Cloud came know, it's funny when Open Cloud came out, everyone started buying Mac minis, out, everyone started buying Mac minis, out, everyone started buying Mac minis, which to me feels like an infrastructure which to me feels like an infrastructure which to me feels like an infrastructure problem, right? You're running Open problem, right? You're running Open problem, right? You're running Open Cloud on a Mac mini in your house Cloud on a Mac mini in your house Cloud on a Mac mini in your house because that's the best way to run Mac because that's the best way to run Mac because that's the best way to run Mac OS that you can SSH into and then end up OS that you can SSH into and then end up OS that you can SSH into and then end up like solving the CAPTCHAs cuz of your like solving the CAPTCHAs cuz of your like solving the CAPTCHAs cuz of your home IP address. That is not something home IP address. That is not something home IP address. That is not something you can do when you're building you can do when you're building you can do when you're building thousands of agents for customers in thousands of agents for customers in thousands of agents for customers in production. I've yet to see a SOC 2 production. I've yet to see a SOC 2 production. I've yet to see a SOC 2 compliant Mac mini setup at scale, but compliant Mac mini setup at scale, but compliant Mac mini setup at scale, but please tell me afterwards if you found please tell me afterwards if you found please tell me afterwards if you found one. I'm very curious about it. The one. I'm very curious about it. The one. I'm very curious about it. The infrastructure problem that needs to be infrastructure problem that needs to be infrastructure problem that needs to be solved here is also an engineering solved here is also an engineering solved here is also an engineering problem. And most importantly, the problem. And most importantly, the problem. And most importantly, the consistency in this infrastructure is consistency in this infrastructure is consistency in this infrastructure is important. When your agent is running important. When your agent is running important. When your agent is running across a website multiple times, you across a website multiple times, you across a website multiple times, you want it to see the same inputs and want it to see the same inputs and want it to see the same inputs and outputs, the same page layout, the same outputs, the same page layout, the same outputs, the same page layout, the same size. If your infrastructure renders a size. If your infrastructure renders a size. If your infrastructure renders a page in a mobile layout one time and page in a mobile layout one time and page in a mobile layout one time and then like in a desktop layout the second then like in a desktop layout the second then like in a desktop layout the second time, it's going to have inconsistent time, it's going to have inconsistent time, it's going to have inconsistent results. Consistency in the results. Consistency in the results. Consistency in the infrastructure is the nice base layer on infrastructure is the nice base layer on infrastructure is the nice base layer on top of your harness and on top of your top of your harness and on top of your top of your harness and on top of your models to actually get good results with models to actually get good results with models to actually get good results with this.

I also think we have to improve the web I also think we have to improve the web itself. So, there's a whole other side itself. So, there's a whole other side itself. So, there's a whole other side of this problem that's very interesting, of this problem that's very interesting, of this problem that's very interesting, which is like, how are we going to make which is like, how are we going to make which is like, how are we going to make it so the web works well with agents? it so the web works well with agents? it so the web works well with agents? And I think this is arguably the harder And I think this is arguably the harder And I think this is arguably the harder challenge because we're not just challenge because we're not just challenge because we're not just engineering on our own systems anymore. engineering on our own systems anymore. engineering on our own systems anymore. We have to be evangelists to the web and We have to be evangelists to the web and We have to be evangelists to the web and to the broader world that, "Hey, you to the broader world that, "Hey, you to the broader world that, "Hey, you want agents to come to your website." want agents to come to your website." want agents to come to your website." So, accessibility is the first thing I So, accessibility is the first thing I So, accessibility is the first thing I want to talk about. There's been a lot want to talk about. There's been a lot want to talk about. There's been a lot of really cool stuff here. Now, when you of really cool stuff here. Now, when you of really cool stuff here. Now, when you look at what best-in-class browser look at what best-in-class browser look at what best-in-class browser agents are doing, they're not just agents are doing, they're not just agents are doing, they're not just consuming the raw DOM and HTML of the consuming the raw DOM and HTML of the consuming the raw DOM and HTML of the page anymore. They're looking at sub, page anymore. They're looking at sub, page anymore. They're looking at sub, you know, subsections of that like the you know, subsections of that like the you know, subsections of that like the accessibility tree, the area tags. These accessibility tree, the area tags. These accessibility tree, the area tags. These are labeled components of a page that are labeled components of a page that are labeled components of a page that can help show your agent where it needs can help show your agent where it needs can help show your agent where it needs to click and why. to click and why. to click and why. Chrome just added WebMCP, which I think Chrome just added WebMCP, which I think Chrome just added WebMCP, which I think is really, really cool. Websites can now is really, really cool. Websites can now is really, really cool. Websites can now publish MCP servers within their page publish MCP servers within their page publish MCP servers within their page that your agent can take advantage of that your agent can take advantage of that your agent can take advantage of without pre-installing the actual MCP. without pre-installing the actual MCP. without pre-installing the actual MCP. It can now issue tool calls to a website It can now issue tool calls to a website It can now issue tool calls to a website like submit the registration form in a like submit the registration form in a like submit the registration form in a way that's not only context context way that's not only context context way that's not only context context context-efficient, but is website context-efficient, but is website context-efficient, but is website approved and blessed.

approved and blessed. approved and blessed. More and more work can go into More and more work can go into More and more work can go into accessibility and we've seen things like accessibility and we've seen things like accessibility and we've seen things like LLMs.text, LLMs.text, LLMs.text, skills.md, agents.md all being published skills.md, agents.md all being published skills.md, agents.md all being published alongside our websites. We need to see alongside our websites. We need to see alongside our websites. We need to see more of that to build the agent-first more of that to build the agent-first more of that to build the agent-first web. web. web. I think authentication is actually an I think authentication is actually an I think authentication is actually an even bigger problem here, too, because even bigger problem here, too, because even bigger problem here, too, because once your agent can actually go to a once your agent can actually go to a once your agent can actually go to a website, how can it log in on your website, how can it log in on your website, how can it log in on your behalf? There's been a lot of different behalf? There's been a lot of different behalf? There's been a lot of different paradigms here. Maybe you're just giving paradigms here. Maybe you're just giving paradigms here. Maybe you're just giving your agent your password, but doing that your agent your password, but doing that your agent your password, but doing that securely can be very challenging. Maybe securely can be very challenging. Maybe securely can be very challenging. Maybe you're creating a service account for you're creating a service account for you're creating a service account for your agent where it has some limited your agent where it has some limited your agent where it has some limited access and you constantly have to give access and you constantly have to give access and you constantly have to give it new permissions. You know, it new permissions. You know, it new permissions. You know, authentication for agents is the next authentication for agents is the next authentication for agents is the next thing to be solved once you solve the thing to be solved once you solve the thing to be solved once you solve the harnessing capability problems, and harnessing capability problems, and harnessing capability problems, and doing that securely where you can have a doing that securely where you can have a doing that securely where you can have a human loop approve certain actions on a human loop approve certain actions on a human loop approve certain actions on a website is going to be a major challenge website is going to be a major challenge website is going to be a major challenge for unlocking computers for the for unlocking computers for the for unlocking computers for the enterprise. enterprise. enterprise. The biggest gate to building agents that The biggest gate to building agents that The biggest gate to building agents that actually can work in prod is going to be actually can work in prod is going to be actually can work in prod is going to be the systems it has access to. And the systems it has access to. And the systems it has access to. And authentication is something that needs authentication is something that needs authentication is something that needs to be solved in our industry to make to be solved in our industry to make to be solved in our industry to make this possible. I've seen a lot of really this possible. I've seen a lot of really this possible. I've seen a lot of really cool stuff come out. WorkOS just cool stuff come out. WorkOS just cool stuff come out. WorkOS just launched OffMD, which is a new way for launched OffMD, which is a new way for launched OffMD, which is a new way for your agent that goes to a website to your agent that goes to a website to your agent that goes to a website to find how to sign up on that website and find how to sign up on that website and find how to sign up on that website and get its own accounts. And if you're get its own accounts. And if you're get its own accounts. And if you're building software now, you should think building software now, you should think building software now, you should think about what is my agent-first sign up and about what is my agent-first sign up and about what is my agent-first sign up and login flow look like because agents are login flow look like because agents are login flow look like because agents are going to be using your software whether going to be using your software whether going to be using your software whether you like it or not. It's best to let you like it or not. It's best to let you like it or not. It's best to let them use it securely.

them use it securely. them use it securely. Finally, I want to talk about trust. Finally, I want to talk about trust. Finally, I want to talk about trust. The web was built to stop bad bots, but The web was built to stop bad bots, but The web was built to stop bad bots, but now there's good agents and bad bots. now there's good agents and bad bots. now there's good agents and bad bots. How do we delineate between the two? And How do we delineate between the two? And How do we delineate between the two? And the CAPTCHA has been the tool in our the CAPTCHA has been the tool in our the CAPTCHA has been the tool in our tool chest for a very long time, but as tool chest for a very long time, but as tool chest for a very long time, but as we all know, CAPTCHAs are not as we all know, CAPTCHAs are not as we all know, CAPTCHAs are not as effective as we think against agents. effective as we think against agents. effective as we think against agents. And trying to identify these good agents And trying to identify these good agents And trying to identify these good agents is very important. There's been a lot of is very important. There's been a lot of is very important. There's been a lot of cool frameworks and work done on things cool frameworks and work done on things cool frameworks and work done on things like web bot off and in more like web bot off and in more like web bot off and in more authenticated ways to say this is my authenticated ways to say this is my authenticated ways to say this is my agent, it's coming from me, and you can agent, it's coming from me, and you can agent, it's coming from me, and you can follow me along on the web, but I still follow me along on the web, but I still follow me along on the web, but I still don't think we've solved the issue yet. don't think we've solved the issue yet. don't think we've solved the issue yet. And And And a big unlock to agents accessing the web a big unlock to agents accessing the web a big unlock to agents accessing the web alongside authentication is actually how alongside authentication is actually how alongside authentication is actually how can we trust these agents? And I think can we trust these agents? And I think can we trust these agents? And I think there needs to be almost like a Verisign there needs to be almost like a Verisign there needs to be almost like a Verisign moment for web agents where who can be moment for web agents where who can be moment for web agents where who can be the certificate issuer in saying my the certificate issuer in saying my the certificate issuer in saying my agent is trusted in this agent vendor is agent is trusted in this agent vendor is agent is trusted in this agent vendor is trusted? Nobody's come out and done that trusted? Nobody's come out and done that trusted? Nobody's come out and done that yet. I think those are really really big yet. I think those are really really big yet. I think those are really really big opportunities. opportunities. opportunities. So, building reliable browser agents is So, building reliable browser agents is So, building reliable browser agents is is not a model problem. It's an is not a model problem. It's an is not a model problem. It's an engineering problem that all of us can engineering problem that all of us can engineering problem that all of us can solve, but doing that engineering is is solve, but doing that engineering is is solve, but doing that engineering is is a full-time job. And if you are working a full-time job. And if you are working a full-time job. And if you are working in the space, I'd love to meet you. But in the space, I'd love to meet you. But in the space, I'd love to meet you. But if you aren't and you just want to build if you aren't and you just want to build if you aren't and you just want to build something that works, I might feel something that works, I might feel something that works, I might feel ideas. You really don't have to reinvent ideas. You really don't have to reinvent ideas. You really don't have to reinvent the wheel here. There's been a lot of the wheel here. There's been a lot of the wheel here. There's been a lot of stuff happening and it's a consortium of stuff happening and it's a consortium of stuff happening and it's a consortium of companies that are continuing to push companies that are continuing to push companies that are continuing to push the world forward on what's possible the world forward on what's possible the world forward on what's possible when you want to automate the web.

when you want to automate the web. when you want to automate the web. And I think there's like a few things And I think there's like a few things And I think there's like a few things here that are really important for a here that are really important for a here that are really important for a great solution, right? It has to be a great solution, right? It has to be a great solution, right? It has to be a scalable platform that serves your scalable platform that serves your scalable platform that serves your infrastructure needs. You know, you can infrastructure needs. You know, you can infrastructure needs. You know, you can want to run run one agent, but also want to run run one agent, but also want to run run one agent, but also thousands of agents. And And the thousands of agents. And And the thousands of agents. And And the challenge is that those different levels challenge is that those different levels challenge is that those different levels of scale is very very important. You of scale is very very important. You of scale is very very important. You want browser agents that are model want browser agents that are model want browser agents that are model agnostic. As a developer, I don't want agnostic. As a developer, I don't want agnostic. As a developer, I don't want to be locked into a single model to be locked into a single model to be locked into a single model provider. As models continually change provider. As models continually change provider. As models continually change and get better, I want to be able to and get better, I want to be able to and get better, I want to be able to move my agent around. That's why you move my agent around. That's why you move my agent around. That's why you need model agnostic infrastructure. You need model agnostic infrastructure. You need model agnostic infrastructure. You need somebody to solve agent identity. need somebody to solve agent identity. need somebody to solve agent identity. Somebody who's going to go out and Somebody who's going to go out and Somebody who's going to go out and negotiate with the, you know, anti-bot negotiate with the, you know, anti-bot negotiate with the, you know, anti-bot providers of the world and say, "We are providers of the world and say, "We are providers of the world and say, "We are the platform for trusted agents and we the platform for trusted agents and we the platform for trusted agents and we are the ones that can help broker the are the ones that can help broker the are the ones that can help broker the access for your agents as you use the access for your agents as you use the access for your agents as you use the web." And finally, you need web." And finally, you need web." And finally, you need observability. When you're building observability. When you're building observability. When you're building these agents that go to any website in these agents that go to any website in these agents that go to any website in the world, you need to see where they're the world, you need to see where they're the world, you need to see where they're going and why and how that you can make going and why and how that you can make going and why and how that you can make sure that it's improving every sure that it's improving every sure that it's improving every iteration. Every agent you run should iteration. Every agent you run should iteration. Every agent you run should get better every single time. You need get better every single time. You need get better every single time. You need screen recordings, logs, network screen recordings, logs, network screen recordings, logs, network activity. And you need to feed that back activity. And you need to feed that back activity. And you need to feed that back into your agent so it can self-improve. into your agent so it can self-improve. into your agent so it can self-improve. We published something called Auto We published something called Auto We published something called Auto Browse earlier this year. That's a Browse earlier this year. That's a Browse earlier this year. That's a really interesting way to see how is my really interesting way to see how is my really interesting way to see how is my agent able to improve itself over agent able to improve itself over agent able to improve itself over multiple loops. And the the feed-in of multiple loops. And the the feed-in of multiple loops. And the the feed-in of data to that from observability is data to that from observability is data to that from observability is extremely important to make your agents extremely important to make your agents extremely important to make your agents get better over time.

get better over time. get better over time. And And that's, you know, what we're And And that's, you know, what we're And And that's, you know, what we're building here at BrowserBase. You know, building here at BrowserBase. You know, building here at BrowserBase. You know, we power browser agents, web data we power browser agents, web data we power browser agents, web data extraction, and really all these use extraction, and really all these use extraction, and really all these use cases across the entire web to make your cases across the entire web to make your cases across the entire web to make your agents work well. And what I've been agents work well. And what I've been agents work well. And what I've been extremely surprised by in building this extremely surprised by in building this extremely surprised by in building this company is the plethora of use cases. Of company is the plethora of use cases. Of company is the plethora of use cases. Of course, there are the, you know, large course, there are the, you know, large course, there are the, you know, large AI native companies that use companies AI native companies that use companies AI native companies that use companies like BrowserBase to power the browser like BrowserBase to power the browser like BrowserBase to power the browser agents, but there's also all these agents, but there's also all these agents, but there's also all these little companies across the world that little companies across the world that little companies across the world that can benefit from automation. And my core can benefit from automation. And my core can benefit from automation. And my core belief with this company is that solving belief with this company is that solving belief with this company is that solving computer use accelerates the diffusion computer use accelerates the diffusion computer use accelerates the diffusion of AI to the real economy. And as much of AI to the real economy. And as much of AI to the real economy. And as much as I love our bubble here in San as I love our bubble here in San as I love our bubble here in San Francisco, the real economy is companies Francisco, the real economy is companies Francisco, the real economy is companies like the logistics company in Singapore, like the logistics company in Singapore, like the logistics company in Singapore, the bank in South Africa, or the lumber the bank in South Africa, or the lumber the bank in South Africa, or the lumber factory in Mexico. These people are factory in Mexico. These people are factory in Mexico. These people are built on PHP websites with forums and built on PHP websites with forums and built on PHP websites with forums and human beings clicking buttons every human beings clicking buttons every human beings clicking buttons every single day. That's a huge opportunity single day. That's a huge opportunity single day. That's a huge opportunity for you to go solve to build browser for you to go solve to build browser for you to go solve to build browser agents for them, and hopefully you can agents for them, and hopefully you can agents for them, and hopefully you can use the right infrastructure to power use the right infrastructure to power use the right infrastructure to power those things. those things. those things. And and that's why we built BrowserBase And and that's why we built BrowserBase And and that's why we built BrowserBase agents, by the way. This is our new agents, by the way. This is our new agents, by the way. This is our new product we launched yesterday, because product we launched yesterday, because product we launched yesterday, because we want to give everyone a we want to give everyone a we want to give everyone a battery-included agent and harness for battery-included agent and harness for battery-included agent and harness for everything they need to automate the everything they need to automate the everything they need to automate the web. The goal here is that you shouldn't web. The goal here is that you shouldn't web. The goal here is that you shouldn't reinvent the wheel. You shouldn't have reinvent the wheel. You shouldn't have reinvent the wheel. You shouldn't have to figure all this out and optimize it.

to figure all this out and optimize it. to figure all this out and optimize it. You should benefit from the platform You should benefit from the platform You should benefit from the platform scale that we've seen millions and scale that we've seen millions and scale that we've seen millions and millions of sessions every single month, millions of sessions every single month, millions of sessions every single month, and understand how we've solved the edge and understand how we've solved the edge and understand how we've solved the edge cases for you, so you don't have to cases for you, so you don't have to cases for you, so you don't have to solve them on your own. solve them on your own. solve them on your own. I have a quick little demo here. The way I have a quick little demo here. The way I have a quick little demo here. The way it works is instead of having to pull it works is instead of having to pull it works is instead of having to pull our tools together, you can actually put our tools together, you can actually put our tools together, you can actually put in a prompt, and we will stand up the in a prompt, and we will stand up the in a prompt, and we will stand up the harness, the runtime, the sandbox, the harness, the runtime, the sandbox, the harness, the runtime, the sandbox, the code execution, the fetch, the search code execution, the fetch, the search code execution, the fetch, the search tools, and the models to actually tools, and the models to actually tools, and the models to actually accomplish a task for you. And what's accomplish a task for you. And what's accomplish a task for you. And what's beautiful is as this agent is running, beautiful is as this agent is running, beautiful is as this agent is running, it's looking at its steps, and it's it's looking at its steps, and it's it's looking at its steps, and it's remembering what it can do, and learning remembering what it can do, and learning remembering what it can do, and learning from it, so it can do them again in the from it, so it can do them again in the from it, so it can do them again in the future. future. future. The future for you is not having to The future for you is not having to The future for you is not having to reinvent the wheel every single time. reinvent the wheel every single time. reinvent the wheel every single time. It's actually being able to use an agent It's actually being able to use an agent It's actually being able to use an agent that's purpose-built for browsing the that's purpose-built for browsing the that's purpose-built for browsing the web, and pull it in as a sub-agent of web, and pull it in as a sub-agent of web, and pull it in as a sub-agent of your larger agentic system. your larger agentic system. your larger agentic system. This is not the main thing you should be This is not the main thing you should be This is not the main thing you should be focusing your time on. You should be focusing your time on. You should be focusing your time on. You should be focusing your time on actually solving focusing your time on actually solving focusing your time on actually solving customer problems, not trying to rebuild customer problems, not trying to rebuild customer problems, not trying to rebuild the best in class browser agents. the best in class browser agents. the best in class browser agents. The optimization feature is quite cool. The optimization feature is quite cool. The optimization feature is quite cool. It's going to look back and actually It's going to look back and actually It's going to look back and actually understand, hey, how can I do this understand, hey, how can I do this understand, hey, how can I do this better after looking back at this this better after looking back at this this better after looking back at this this is this data feedback loop that I've is this data feedback loop that I've is this data feedback loop that I've talked about before. And I think it's talked about before. And I think it's talked about before. And I think it's what makes agents really, really what makes agents really, really what makes agents really, really special.

special. special. I kind of want to end with this last I kind of want to end with this last I kind of want to end with this last note. note. note. You know, You know, You know, based on the attendance in the room, I based on the attendance in the room, I based on the attendance in the room, I do think a lot of people have stepped do think a lot of people have stepped do think a lot of people have stepped back from computers because they've had back from computers because they've had back from computers because they've had so much challenges over the past year so much challenges over the past year so much challenges over the past year making browser agents work in making browser agents work in making browser agents work in production, but I can tell you first production, but I can tell you first production, but I can tell you first hand from our customers we see it hand from our customers we see it hand from our customers we see it working. And actually, I think 1 year working. And actually, I think 1 year working. And actually, I think 1 year from now this room is going to be from now this room is going to be from now this room is going to be overfilled with people because the overfilled with people because the overfilled with people because the models are getting better, the models are getting better, the models are getting better, the techniques are getting better, and the techniques are getting better, and the techniques are getting better, and the tools are getting better. It's just on tools are getting better. It's just on tools are getting better. It's just on us to build better things. Thank you all us to build better things. Thank you all us to build better things. Thank you all for having me today. I really appreciate for having me today. I really appreciate for having me today. I really appreciate it. it. it. >> [applause]