Computer-use models will agentify the web, not APIs — Dhruv Batra, Yutori
Read full transcript 16 segments
-
Um, welcome. Let's get started. So, my Um, welcome. Let's get started. So, my name is Dadra. I want to talk to you name is Dadra. I want to talk to you name is Dadra. I want to talk to you about an argument. There's an argument about an argument. There's an argument about an argument. There's an argument online that says, roughly speaking, AI online that says, roughly speaking, AI online that says, roughly speaking, AI agents will be the drivers of actions on agents will be the drivers of actions on agents will be the drivers of actions on the web, not humans. that there will be the web, not humans. that there will be the web, not humans. that there will be more agents clicking buttons, reading more agents clicking buttons, reading more agents clicking buttons, reading things, buying things for us than human things, buying things for us than human things, buying things for us than human eyeballs on the web. Um, you drive one eyeballs on the web. Um, you drive one eyeballs on the web. Um, you drive one step deeper into the argument and you step deeper into the argument and you step deeper into the argument and you ask how will this happen? Because today ask how will this happen? Because today ask how will this happen? Because today the web is extremely hostile to the web is extremely hostile to the web is extremely hostile to automated traffic and the answer usually automated traffic and the answer usually automated traffic and the answer usually is the web will be identified. um you is the web will be identified. um you is the web will be identified. um you drive one step deeper into what does drive one step deeper into what does drive one step deeper into what does that mean and how will that happen and that mean and how will that happen and that mean and how will that happen and the answer usually is with APIs that the answer usually is with APIs that the answer usually is with APIs that agents will call via some set of agents will call via some set of agents will call via some set of protocols and the set of protocols grows protocols and the set of protocols grows protocols and the set of protocols grows over time. I mean it was initially over time. I mean it was initially over time. I mean it was initially supposed to be MCP servers then web MCP supposed to be MCP servers then web MCP supposed to be MCP servers then web MCP and for payments there are 20 different and for payments there are 20 different and for payments there are 20 different competing protocols and you know every competing protocols and you know every competing protocols and you know every single company wants to introduce their single company wants to introduce their single company wants to introduce their own protocol along the way but okay own protocol along the way but okay own protocol along the way but okay there will be APIs is the is the there will be APIs is the is the there will be APIs is the is the argument. My claim today and argument argument. My claim today and argument argument. My claim today and argument and what I hope to convince you today is and what I hope to convince you today is and what I hope to convince you today is that this last bit is wrong. I think the that this last bit is wrong. I think the that this last bit is wrong. I think the first two I generally agree with. This first two I generally agree with. This first two I generally agree with. This last bit that suddenly the web will last bit that suddenly the web will last bit that suddenly the web will provide you APIs for accessing things is provide you APIs for accessing things is provide you APIs for accessing things is is just delusional.
-
is just delusional. is just delusional. And my claim is computer use agents and And my claim is computer use agents and And my claim is computer use agents and computer use models will identify the computer use models will identify the computer use models will identify the web not APIs and more specifically the web not APIs and more specifically the web not APIs and more specifically the long tail of the web. Um the head of the long tail of the web. Um the head of the long tail of the web. Um the head of the distribution the most popular website distribution the most popular website distribution the most popular website perhaps will give you the API but the perhaps will give you the API but the perhaps will give you the API but the long tail will not. Um and the story long tail will not. Um and the story long tail will not. Um and the story starts uh usually with these sorts of starts uh usually with these sorts of starts uh usually with these sorts of use cases. I I agree that they also use cases. I I agree that they also use cases. I I agree that they also frustrate me when people show a use case frustrate me when people show a use case frustrate me when people show a use case of find me business class flights from of find me business class flights from of find me business class flights from Miami to Palma de Mayorca for August and Miami to Palma de Mayorca for August and Miami to Palma de Mayorca for August and somehow they expect that behind the somehow they expect that behind the somehow they expect that behind the scenes it'll be a computer use agent scenes it'll be a computer use agent scenes it'll be a computer use agent clicking buttons on flights.google.com. clicking buttons on flights.google.com. clicking buttons on flights.google.com. I find this example, this is my own I find this example, this is my own I find this example, this is my own example. I I'm not dunking on anybody example. I I'm not dunking on anybody example. I I'm not dunking on anybody else. I find this example bizarre. Why else. I find this example bizarre. Why else. I find this example bizarre. Why would you possibly do it this way? would you possibly do it this way? would you possibly do it this way? Aren't you aware that there are already Aren't you aware that there are already Aren't you aware that there are already aggregators? You send them a variable, aggregators? You send them a variable, aggregators? You send them a variable, they will send you a JSON. Your LLMs can they will send you a JSON. Your LLMs can they will send you a JSON. Your LLMs can do tool calling. Why would you click do tool calling. Why would you click do tool calling. Why would you click buttons? The purpose of those clicking buttons? The purpose of those clicking buttons? The purpose of those clicking buttons is to generate this behind the buttons is to generate this behind the buttons is to generate this behind the scenes. So, we're on the same page scenes. So, we're on the same page scenes. So, we're on the same page there. The next step usually is, okay, there. The next step usually is, okay, there. The next step usually is, okay, let's roll this out to some other case.
-
let's roll this out to some other case. let's roll this out to some other case. Um, I want to know does my favorite Um, I want to know does my favorite Um, I want to know does my favorite restaurant or does this restaurant have restaurant or does this restaurant have restaurant or does this restaurant have gluten-free items on the restaurant's gluten-free items on the restaurant's gluten-free items on the restaurant's menu? menu? menu? And so I assume that there's going to be And so I assume that there's going to be And so I assume that there's going to be a similar endpoint somewhere for that a similar endpoint somewhere for that a similar endpoint somewhere for that restaurant website or via an aggregator. restaurant website or via an aggregator. restaurant website or via an aggregator. Oops. Um that accepts a query that I can Um that accepts a query that I can filter through where I can ask for give filter through where I can ask for give filter through where I can ask for give me your menu items that are gluten-free. me your menu items that are gluten-free. me your menu items that are gluten-free. Um I want to for those of you who can Um I want to for those of you who can Um I want to for those of you who can already see I want to rid you of the already see I want to rid you of the already see I want to rid you of the delusion that such an endpoint exists. delusion that such an endpoint exists. delusion that such an endpoint exists. Uh, and I want to show you just for the Uh, and I want to show you just for the Uh, and I want to show you just for the sake of, you know, being on the same sake of, you know, being on the same sake of, you know, being on the same page. What do restaurant pages look page. What do restaurant pages look page. What do restaurant pages look like? Um, there are three that are on like? Um, there are three that are on like? Um, there are three that are on the screen. This is the first one. This the screen. This is the first one. This the screen. This is the first one. This is what we imagine a prototypical. This is what we imagine a prototypical. This is what we imagine a prototypical. This is easy mode. You go to a web page. is easy mode. You go to a web page. is easy mode. You go to a web page. There's no API, but at least it's text. There's no API, but at least it's text. There's no API, but at least it's text. It's in German. Sure, a different part It's in German. Sure, a different part It's in German. Sure, a different part of the world. Models can speak of the world. Models can speak of the world. Models can speak languages. I see text and I see languages. I see text and I see languages. I see text and I see pricings. Maybe this is easily pricings. Maybe this is easily pricings. Maybe this is easily scrapable. Okay, this is easy mode. The scrapable. Okay, this is easy mode. The scrapable. Okay, this is easy mode. The medium mode is a medium mode is a medium mode is a a a page like this where of course the a a page like this where of course the a a page like this where of course the thing that you're looking for is thing that you're looking for is thing that you're looking for is slightly hidden away. You press the menu slightly hidden away. You press the menu slightly hidden away. You press the menu button, it actually takes you to a PDF.
-
button, it actually takes you to a PDF. button, it actually takes you to a PDF. Okay, fine. If your agent has to do Okay, fine. If your agent has to do Okay, fine. If your agent has to do this, it needs a PDF reader mode. Fine. this, it needs a PDF reader mode. Fine. this, it needs a PDF reader mode. Fine. Like, you know, I can highlight these. Like, you know, I can highlight these. Like, you know, I can highlight these. Okay, so this is at least text. No Okay, so this is at least text. No Okay, so this is at least text. No problem. Like, you know, I will dump problem. Like, you know, I will dump problem. Like, you know, I will dump this into chat GPT or whatever and it'll this into chat GPT or whatever and it'll this into chat GPT or whatever and it'll do it. Here's the hard mode. This is do it. Here's the hard mode. This is do it. Here's the hard mode. This is what a hard mode website for a what a hard mode website for a what a hard mode website for a restaurant looks like, where there's restaurant looks like, where there's restaurant looks like, where there's just pictures of the people who created just pictures of the people who created just pictures of the people who created the restaurant, the restaurant, the restaurant, um, of where they're located, of menu um, of where they're located, of menu um, of where they're located, of menu items, and you're like, where is the items, and you're like, where is the items, and you're like, where is the menu? Uh, does anybody speak Spanish? menu? Uh, does anybody speak Spanish? menu? Uh, does anybody speak Spanish? No estraa. See? Okay, let's let's check. No estraa. See? Okay, let's let's check. No estraa. See? Okay, let's let's check. That's our menu. Um, you click there. Oh That's our menu. Um, you click there. Oh That's our menu. Um, you click there. Oh my god, these are What am I staring at? my god, these are What am I staring at? my god, these are What am I staring at? Okay. Uh, you click on it. Oh, it's a Okay. Uh, you click on it. Oh, it's a Okay. Uh, you click on it. Oh, it's a gallery of of individual. What is this? gallery of of individual. What is this? gallery of of individual. What is this? This is a This is a pixelated. This is This is a This is a pixelated. This is This is a This is a pixelated. This is There's no text here. I am looking at There's no text here. I am looking at There's no text here. I am looking at JPEGs embedded in a gallery which JPEGs embedded in a gallery which JPEGs embedded in a gallery which contain the PDF items. You download this contain the PDF items. You download this contain the PDF items. You download this and you put it into chat GPT and it's and you put it into chat GPT and it's and you put it into chat GPT and it's having it's struggling doing OCR on this having it's struggling doing OCR on this having it's struggling doing OCR on this thing. Okay, so this is what the web thing. Okay, so this is what the web thing. Okay, so this is what the web looks like. You're telling me this will looks like. You're telling me this will looks like. You're telling me this will give you an API endpoint that you can give you an API endpoint that you can give you an API endpoint that you can pass glutenfree items and it'll tell you pass glutenfree items and it'll tell you pass glutenfree items and it'll tell you what that is.
-
what that is. what that is. That's an example number one. Let's take That's an example number one. Let's take That's an example number one. Let's take a look at another examples. Uh this is a look at another examples. Uh this is a look at another examples. Uh this is much more, you know, enterprise focused. much more, you know, enterprise focused. much more, you know, enterprise focused. Um um I am a business. I want to sell to Um um I am a business. I want to sell to Um um I am a business. I want to sell to a school district. There's about 15 to a school district. There's about 15 to a school district. There's about 15 to 20,000 school districts in the US. And I 20,000 school districts in the US. And I 20,000 school districts in the US. And I want to ask the question, is this school want to ask the question, is this school want to ask the question, is this school district where I give you, you know, my district where I give you, you know, my district where I give you, you know, my district.gov, is this school district district.gov, is this school district district.gov, is this school district procuring a laptop right now? Right? procuring a laptop right now? Right? procuring a laptop right now? Right? It's a simple question. And I want a It's a simple question. And I want a It's a simple question. And I want a similar endpoint. I want to say RFP similar endpoint. I want to say RFP similar endpoint. I want to say RFP status open and topic laptop. and you status open and topic laptop. and you status open and topic laptop. and you think this will exist. Uh let me tell think this will exist. Uh let me tell think this will exist. Uh let me tell you what these websites look like. you what these websites look like. you what these websites look like. Again, um this is what easy mode looks Again, um this is what easy mode looks Again, um this is what easy mode looks like. This is, you know, Ithaca public like. This is, you know, Ithaca public like. This is, you know, Ithaca public schools. Um okay, there's some portal. schools. Um okay, there's some portal. schools. Um okay, there's some portal. Um maybe there's, you know, uh some, you Um maybe there's, you know, uh some, you Um maybe there's, you know, uh some, you know, enrollment is maybe the wrong know, enrollment is maybe the wrong know, enrollment is maybe the wrong place to go. This is what slightly place to go. This is what slightly place to go. This is what slightly harder mode looks like. Um this is a harder mode looks like. Um this is a harder mode looks like. Um this is a different public school. You look at the different public school. You look at the different public school. You look at the menu under finance there's purchasing menu under finance there's purchasing menu under finance there's purchasing under purchasing there's certain under purchasing there's certain under purchasing there's certain solicitations.
-
solicitations. solicitations. Um scanned PDF again. Okay. Very nice. Um scanned PDF again. Okay. Very nice. Um scanned PDF again. Okay. Very nice. Uh so no text in here but this is how Uh so no text in here but this is how Uh so no text in here but this is how they tell you about what they have have they tell you about what they have have they tell you about what they have have purchased. Um, and for ultimate boss purchased. Um, and for ultimate boss purchased. Um, and for ultimate boss level, I want to show you a different level, I want to show you a different level, I want to show you a different school district where in order to find school district where in order to find school district where in order to find information, you have to file a request information, you have to file a request information, you have to file a request for access of in information under for access of in information under for access of in information under Freedom of Information Act. And then Freedom of Information Act. And then Freedom of Information Act. And then what they do is they scan your email what they do is they scan your email what they do is they scan your email that you sent to them that you sent to them that you sent to them and they will scan that email, put it on and they will scan that email, put it on and they will scan that email, put it on a Google Drive and then attach PDFs a Google Drive and then attach PDFs a Google Drive and then attach PDFs associated with your request. associated with your request. associated with your request. These are the people you're telling me These are the people you're telling me These are the people you're telling me will give you an MCP server. will give you an MCP server. will give you an MCP server. The the amount of delusion here is off The the amount of delusion here is off The the amount of delusion here is off the chart. like the chart. like the chart. like if there was ever a time to to say go if there was ever a time to to say go if there was ever a time to to say go touch grass like I think this is it. touch grass like I think this is it. touch grass like I think this is it. Okay. Uh so this is what my claim here is that the web we forget my claim here is that the web we forget is massive. It is extremely big. The the is massive. It is extremely big. The the is massive. It is extremely big. The the number of websites out there uh somewhat number of websites out there uh somewhat number of websites out there uh somewhat you know the active websites are you know the active websites are you know the active websites are somewhere 200 million. the total number somewhere 200 million. the total number somewhere 200 million. the total number of websites are sitting in a in a of websites are sitting in a in a of websites are sitting in a in a billion infrastructure changes very billion infrastructure changes very billion infrastructure changes very slowly. Um you can imagine as an slowly. Um you can imagine as an slowly. Um you can imagine as an engineer you know getting unfettered engineer you know getting unfettered engineer you know getting unfettered access to software systems and letting
-
access to software systems and letting access to software systems and letting your favorite coding agent rip and your favorite coding agent rip and your favorite coding agent rip and generate an API endpoint even if that generate an API endpoint even if that generate an API endpoint even if that technological problem is solved which I technological problem is solved which I technological problem is solved which I agree seems like on a horizon it should agree seems like on a horizon it should agree seems like on a horizon it should eventually be solved you're not going to eventually be solved you're not going to eventually be solved you're not going to get unfettered access to these get unfettered access to these get unfettered access to these institutions and these institutions institutions and these institutions institutions and these institutions change very slowly there are still change very slowly there are still change very slowly there are still places that are faxing each other you're places that are faxing each other you're places that are faxing each other you're you're not going to be able to suddenly you're not going to be able to suddenly you're not going to be able to suddenly change this overnight. change this overnight. change this overnight. Okay, so at this point of time you might Okay, so at this point of time you might Okay, so at this point of time you might be thinking fine, I'm stuck with the be thinking fine, I'm stuck with the be thinking fine, I'm stuck with the infra as it is. Um, so there's not going infra as it is. Um, so there's not going infra as it is. Um, so there's not going to be APIs available, but I have coding to be APIs available, but I have coding to be APIs available, but I have coding agents. Why don't I just throw them at agents. Why don't I just throw them at agents. Why don't I just throw them at the HTML? There is, after all, if the the HTML? There is, after all, if the the HTML? There is, after all, if the browser is doing it, it's a piece of browser is doing it, it's a piece of browser is doing it, it's a piece of code. My coding agent should be able to code. My coding agent should be able to code. My coding agent should be able to do it as well. Fair fair point. I, you do it as well. Fair fair point. I, you do it as well. Fair fair point. I, you know, I also thought this way two years know, I also thought this way two years know, I also thought this way two years ago. Let me tell you what what the web ago. Let me tell you what what the web ago. Let me tell you what what the web actually looks like. Um, actually looks like. Um, actually looks like. Um, if I ask you the question, what was the if I ask you the question, what was the if I ask you the question, what was the final score of this game between final score of this game between final score of this game between Minnesota Timberwolves and the Brooklyn Minnesota Timberwolves and the Brooklyn Minnesota Timberwolves and the Brooklyn Nets? Here's what the web page looks Nets? Here's what the web page looks Nets? Here's what the web page looks like. This is a modern website. You like. This is a modern website. You like. This is a modern website. You know, it's nba.com.
-
know, it's nba.com. know, it's nba.com. You and I can see the score 125 versus You and I can see the score 125 versus You and I can see the score 125 versus 109. Okay. Uh, behind the scenes, when 109. Okay. Uh, behind the scenes, when 109. Okay. Uh, behind the scenes, when you actually load the page and read it, you actually load the page and read it, you actually load the page and read it, this is what initially gets loaded. this is what initially gets loaded. this is what initially gets loaded. There's an empty placeholder initially There's an empty placeholder initially There's an empty placeholder initially and you wait a few hundred milliseconds and you wait a few hundred milliseconds and you wait a few hundred milliseconds to a few seconds depending on your to a few seconds depending on your to a few seconds depending on your network connection and your browser network connection and your browser network connection and your browser makes an asynchronous call later to makes an asynchronous call later to makes an asynchronous call later to fetch the information. So the browser fetch the information. So the browser fetch the information. So the browser makes a call to an endpoint uh that it's makes a call to an endpoint uh that it's makes a call to an endpoint uh that it's extracting that information from that extracting that information from that extracting that information from that endpoint responds with the JSON that endpoint responds with the JSON that endpoint responds with the JSON that contains the answer. If you just read contains the answer. If you just read contains the answer. If you just read the HTML when you load it, the answer is the HTML when you load it, the answer is the HTML when you load it, the answer is not in the HTML and so your your chatbot not in the HTML and so your your chatbot not in the HTML and so your your chatbot doesn't have access to that either. So, doesn't have access to that either. So, doesn't have access to that either. So, okay, you go fine. These are just okay, you go fine. These are just okay, you go fine. These are just details. I just have to add some weights details. I just have to add some weights details. I just have to add some weights and sleeps and sure. Okay, fair enough. and sleeps and sure. Okay, fair enough. and sleeps and sure. Okay, fair enough. I will take you to another example. Um, I will take you to another example. Um, I will take you to another example. Um, I ask you the question on a product web I ask you the question on a product web I ask you the question on a product web page, is the 25mm Osmium cube in stock page, is the 25mm Osmium cube in stock page, is the 25mm Osmium cube in stock or out of stock? You're on a on this or out of stock? You're on a on this or out of stock? You're on a on this product website. Um, you scroll down, product website. Um, you scroll down, product website. Um, you scroll down, there is a drop-down. Uh, that dropdown there is a drop-down. Uh, that dropdown there is a drop-down. Uh, that dropdown is telling you and I as humans, you is telling you and I as humans, you is telling you and I as humans, you know, three things are sold out, one know, three things are sold out, one know, three things are sold out, one thing is in stock, even though it thing is in stock, even though it thing is in stock, even though it doesn't actually say in stock, like doesn't actually say in stock, like doesn't actually say in stock, like there's no text there that says in there's no text there that says in there's no text there that says in stock, but you understand that grayed stock, but you understand that grayed stock, but you understand that grayed out means sold out. And you know, out means sold out. And you know, out means sold out. And you know, sometimes the sold out won't actually be sometimes the sold out won't actually be sometimes the sold out won't actually be there as a text. Sometimes it'll just be there as a text. Sometimes it'll just be there as a text. Sometimes it'll just be uncclickable grayed out. Okay, surely uncclickable grayed out. Okay, surely uncclickable grayed out. Okay, surely this information must be in the code this information must be in the code this information must be in the code somewhere. You go and read the HTML and somewhere. You go and read the HTML and somewhere. You go and read the HTML and it turns out there is an option
-
it turns out there is an option it turns out there is an option selector. It actually doesn't say any of selector. It actually doesn't say any of selector. It actually doesn't say any of the things that I'm seeing on screen. It the things that I'm seeing on screen. It the things that I'm seeing on screen. It doesn't say sold out. It doesn't say doesn't say sold out. It doesn't say doesn't say sold out. It doesn't say available. So what's going on? Uh turns available. So what's going on? Uh turns available. So what's going on? Uh turns out behind the scenes uh the browser out behind the scenes uh the browser out behind the scenes uh the browser makes a call gets a JSON object which is makes a call gets a JSON object which is makes a call gets a JSON object which is the variable which is product count that the variable which is product count that the variable which is product count that contains a variable called quantity. contains a variable called quantity. contains a variable called quantity. That quantity is you know 10 sometimes That quantity is you know 10 sometimes That quantity is you know 10 sometimes zero sometimes. That's just you know how zero sometimes. That's just you know how zero sometimes. That's just you know how many things can the back end support. many things can the back end support. many things can the back end support. Right now, some of those quantities are Right now, some of those quantities are Right now, some of those quantities are zero and there's a different rendering zero and there's a different rendering zero and there's a different rendering script that anytime there's zero grays script that anytime there's zero grays script that anytime there's zero grays it out and makes it uncclickable. it out and makes it uncclickable. it out and makes it uncclickable. Fundamentally, what is happening here is Fundamentally, what is happening here is Fundamentally, what is happening here is this information that you are seeing on this information that you are seeing on this information that you are seeing on screen is not written somewhere as pure screen is not written somewhere as pure screen is not written somewhere as pure text. It is calculated. It is rendered. text. It is calculated. It is rendered. text. It is calculated. It is rendered. Um, and for people who work in the Um, and for people who work in the Um, and for people who work in the browser industry, they understand this. browser industry, they understand this. browser industry, they understand this. But you know often people who are coming But you know often people who are coming But you know often people who are coming from an AI background like me uh we from an AI background like me uh we from an AI background like me uh we didn't always understand this. The didn't always understand this. The didn't always understand this. The browser is a rendering engine. You are browser is a rendering engine. You are browser is a rendering engine. You are seeing pixels on screen. It's think of seeing pixels on screen. It's think of seeing pixels on screen. It's think of it as a game engine. Um and you're it as a game engine. Um and you're it as a game engine. Um and you're asking can I not read the source code of asking can I not read the source code of asking can I not read the source code of the game and predict exactly what the the game and predict exactly what the the game and predict exactly what the pixels are going to be? Well yes pixels are going to be? Well yes pixels are going to be? Well yes eventually. But right now you're asking eventually. But right now you're asking eventually. But right now you're asking for an exact inversion of that process.
-
for an exact inversion of that process. for an exact inversion of that process. Fundamentally the web was built for Fundamentally the web was built for Fundamentally the web was built for human eyes. Pixels are the source of the human eyes. Pixels are the source of the human eyes. Pixels are the source of the truth because the consumers of the truth because the consumers of the truth because the consumers of the websites are humans. That is what it was websites are humans. That is what it was websites are humans. That is what it was built for. And so there is a implication built for. And so there is a implication built for. And so there is a implication here that we have to we have to grapple here that we have to we have to grapple here that we have to we have to grapple with which is machines will need vision with which is machines will need vision with which is machines will need vision to operate those things because the web to operate those things because the web to operate those things because the web world was built for for human world was built for for human world was built for for human consumption. Um in a way this is the consumption. Um in a way this is the consumption. Um in a way this is the bitter lesson uh for web agents that the bitter lesson uh for web agents that the bitter lesson uh for web agents that the more you end up writing scaffolds around more you end up writing scaffolds around more you end up writing scaffolds around existing websites the it doesn't existing websites the it doesn't existing websites the it doesn't actually generalize to the long tail of actually generalize to the long tail of actually generalize to the long tail of the web. The thing that generalizes is the web. The thing that generalizes is the web. The thing that generalizes is the thing that it was designed for which the thing that it was designed for which the thing that it was designed for which is the most general solution just pixels is the most general solution just pixels is the most general solution just pixels in. in. in. This is what um you know we and some This is what um you know we and some This is what um you know we and some others in the in the area have been others in the in the area have been others in the in the area have been working on. Um we have a model called working on. Um we have a model called working on. Um we have a model called navigator. Um the first version of the navigator. Um the first version of the navigator. Um the first version of the model went out in November last year. model went out in November last year. model went out in November last year. The first version acted purely like a The first version acted purely like a The first version acted purely like a human screenshot in button clicks and human screenshot in button clicks and human screenshot in button clicks and scrolls out. Um I will tell you in a scrolls out. Um I will tell you in a scrolls out. Um I will tell you in a second that is not where you should second that is not where you should second that is not where you should settle on but it is a general solution.
-
settle on but it is a general solution. settle on but it is a general solution. It lets you do things like this. Um I It lets you do things like this. Um I It lets you do things like this. Um I give you an e-commerce website and I give you an e-commerce website and I give you an e-commerce website and I tell you there is this discount code. tell you there is this discount code. tell you there is this discount code. Please go. The discount code is Please go. The discount code is Please go. The discount code is applicable under certain constraints. applicable under certain constraints. applicable under certain constraints. Maybe it's only on a product. It's only Maybe it's only on a product. It's only Maybe it's only on a product. It's only on a certain set of dates. Maybe it's on a certain set of dates. Maybe it's on a certain set of dates. Maybe it's only um with this minimum cart only um with this minimum cart only um with this minimum cart threshold. I describe that in natural threshold. I describe that in natural threshold. I describe that in natural language and I tell you tell me if this language and I tell you tell me if this language and I tell you tell me if this discount code is valid or not. There is discount code is valid or not. There is discount code is valid or not. There is no API for this. The reason the way to no API for this. The reason the way to no API for this. The reason the way to do it is just like a human can. You go do it is just like a human can. You go do it is just like a human can. You go to that website, you find the product to that website, you find the product to that website, you find the product that is described, you add it to cart, that is described, you add it to cart, that is described, you add it to cart, you apply the discount code and you you apply the discount code and you you apply the discount code and you check whether the claim of 22% off was check whether the claim of 22% off was check whether the claim of 22% off was met or not. And that is you know met or not. And that is you know met or not. And that is you know screenshot in button click out. This screenshot in button click out. This screenshot in button click out. This trajectory took you know 20 30 40 steps trajectory took you know 20 30 40 steps trajectory took you know 20 30 40 steps depending on the sophistication of the depending on the sophistication of the depending on the sophistication of the task. If you can do it on your browser task. If you can do it on your browser task. If you can do it on your browser uh this model can accomplish it in uh this model can accomplish it in uh this model can accomplish it in principle. In practice of course there principle. In practice of course there principle. In practice of course there are accuracy gaps and so on. But in are accuracy gaps and so on. But in are accuracy gaps and so on. But in principle this task is solvable whereas principle this task is solvable whereas principle this task is solvable whereas in a lot of earlier cases even in in a lot of earlier cases even in in a lot of earlier cases even in principle that task may not be solvable.
-
So my claim is the web was built for So my claim is the web was built for human eyes. Machines will need vision. human eyes. Machines will need vision. human eyes. Machines will need vision. But of course they do not need to be But of course they do not need to be But of course they do not need to be limited to human ways. Just because for limited to human ways. Just because for limited to human ways. Just because for the long tail you need to have a the long tail you need to have a the long tail you need to have a capability does not mean that is the capability does not mean that is the capability does not mean that is the only way you should do it. Um here is an only way you should do it. Um here is an only way you should do it. Um here is an example showing that the next version of example showing that the next version of example showing that the next version of the model that we trained can also write the model that we trained can also write the model that we trained can also write JavaScript on demand. So this is a JavaScript on demand. So this is a JavaScript on demand. So this is a Chrome extension. On the right you see Chrome extension. On the right you see Chrome extension. On the right you see an action that says execute js value an action that says execute js value an action that says execute js value default text select. On the left, you default text select. On the left, you default text select. On the left, you saw that the model filled out multiple saw that the model filled out multiple saw that the model filled out multiple form fields simultaneously. The reason form fields simultaneously. The reason form fields simultaneously. The reason why it could do that is because it wrote why it could do that is because it wrote why it could do that is because it wrote a little bit of a function. It can read a little bit of a function. It can read a little bit of a function. It can read the code when necessary. It can write the code when necessary. It can write the code when necessary. It can write code when necessary because the browser code when necessary because the browser code when necessary because the browser after all is an engine that can execute after all is an engine that can execute after all is an engine that can execute code, but it has a sort of formal code, but it has a sort of formal code, but it has a sort of formal verification system built in. It it is verification system built in. It it is verification system built in. It it is seeing the screenshot that is the source seeing the screenshot that is the source seeing the screenshot that is the source of the truth. So, it knows whether it of the truth. So, it knows whether it of the truth. So, it knows whether it succeeded or not. So click buttons when succeeded or not. So click buttons when succeeded or not. So click buttons when you have to, write code when you have to you have to, write code when you have to you have to, write code when you have to and look at the result uh through pixels and look at the result uh through pixels and look at the result uh through pixels because that is the that is the source because that is the that is the source because that is the that is the source of of truth.
-
of of truth. of of truth. And of course because these things are And of course because these things are And of course because these things are are machines you can string them into are machines you can string them into are machines you can string them into multi- aent systems. So you can have an multi- aent systems. So you can have an multi- aent systems. So you can have an orchestrator that is launching multiple orchestrator that is launching multiple orchestrator that is launching multiple navigators in parallel each with a cloud navigators in parallel each with a cloud navigators in parallel each with a cloud sandbox instance. They are clicking sandbox instance. They are clicking sandbox instance. They are clicking buttons on multiple websites. So you can buttons on multiple websites. So you can buttons on multiple websites. So you can accomplish things that would be accomplish things that would be accomplish things that would be superhuman because no human would be superhuman because no human would be superhuman because no human would be able to paralyze over that many able to paralyze over that many able to paralyze over that many instances. instances. instances. Um around here usually you know in this Um around here usually you know in this Um around here usually you know in this flow of an argument is when people start flow of an argument is when people start flow of an argument is when people start asking are computer use models actually asking are computer use models actually asking are computer use models actually good enough uh for these tasks. Um good enough uh for these tasks. Um good enough uh for these tasks. Um there's actually a perception online uh there's actually a perception online uh there's actually a perception online uh that um it's not clear whether progress that um it's not clear whether progress that um it's not clear whether progress on computer use has been fast and there on computer use has been fast and there on computer use has been fast and there are questions about why has progress are questions about why has progress are questions about why has progress been slow. That's not the reality I'm been slow. That's not the reality I'm been slow. That's not the reality I'm seeing and that's not the reality that seeing and that's not the reality that seeing and that's not the reality that the numbers back up. Um this is a the numbers back up. Um this is a the numbers back up. Um this is a popular benchmark um online mind to web popular benchmark um online mind to web popular benchmark um online mind to web no benchmark is perfect. The point isn't no benchmark is perfect. The point isn't no benchmark is perfect. The point isn't uh that this is the the right solution uh that this is the the right solution uh that this is the the right solution but on the x-axis are are release times but on the x-axis are are release times but on the x-axis are are release times of different models on the y-axis is of different models on the y-axis is of different models on the y-axis is performance which is human eval on on performance which is human eval on on performance which is human eval on on this. So a human went in looked at the this. So a human went in looked at the this. So a human went in looked at the trajectory decided whether it was trajectory decided whether it was trajectory decided whether it was correct or not and basically this correct or not and basically this correct or not and basically this particular version of the benchmark is particular version of the benchmark is particular version of the benchmark is saturated. The the last the model that saturated. The the last the model that saturated. The the last the model that we just released navigator N 1.5 is we just released navigator N 1.5 is we just released navigator N 1.5 is sitting at 97% human eval eight sitting at 97% human eval eight sitting at 97% human eval eight trajectories out of 300 are incorrect at trajectories out of 300 are incorrect at trajectories out of 300 are incorrect at this point of time you should just this point of time you should just this point of time you should just retire the benchmark build something retire the benchmark build something retire the benchmark build something harder. There's about you know 30 to 50 harder. There's about you know 30 to 50 harder. There's about you know 30 to 50 steps of interaction that are happening.
-
steps of interaction that are happening. steps of interaction that are happening. The next step is to go for something The next step is to go for something The next step is to go for something larger. So at least in numbers what I'm larger. So at least in numbers what I'm larger. So at least in numbers what I'm seeing we're seeing steady progress in seeing we're seeing steady progress in seeing we're seeing steady progress in computer use agents becoming being able computer use agents becoming being able computer use agents becoming being able to do more and more things and this is to do more and more things and this is to do more and more things and this is the model that I showed pixels in button the model that I showed pixels in button the model that I showed pixels in button clicks and code out clicks and code out clicks and code out around this time usually is when people around this time usually is when people around this time usually is when people start asking questions okay so they're start asking questions okay so they're start asking questions okay so they're getting better but aren't these things getting better but aren't these things getting better but aren't these things slow because after all you're looking at slow because after all you're looking at slow because after all you're looking at a screen you're you're clicking a button a screen you're you're clicking a button a screen you're you're clicking a button there are lots of buttons to click and there are lots of buttons to click and there are lots of buttons to click and you know these things are expensive if you know these things are expensive if you know these things are expensive if you're running them for hundreds of of you're running them for hundreds of of you're running them for hundreds of of things. Um that claim I think is largely things. Um that claim I think is largely things. Um that claim I think is largely true. There is some truth to it. Um but true. There is some truth to it. Um but true. There is some truth to it. Um but I think people forget how how much you I think people forget how how much you I think people forget how how much you can optimize these things out. Um so can optimize these things out. Um so can optimize these things out. Um so this is our results compared to the this is our results compared to the this is our results compared to the Frontier models. Opus 4.7, GBD 5.5 on a Frontier models. Opus 4.7, GBD 5.5 on a Frontier models. Opus 4.7, GBD 5.5 on a couple of browser use benchmarks. Um couple of browser use benchmarks. Um couple of browser use benchmarks. Um we're slightly better, but I think we're slightly better, but I think we're slightly better, but I think that's within statistical uh noise in that's within statistical uh noise in that's within statistical uh noise in terms of accuracy. That improvement I terms of accuracy. That improvement I terms of accuracy. That improvement I wouldn't beat the drum on. What I would wouldn't beat the drum on. What I would wouldn't beat the drum on. What I would emphasize is latency per step and cost emphasize is latency per step and cost emphasize is latency per step and cost per task. In terms of latency, this is a per task. In terms of latency, this is a per task. In terms of latency, this is a smaller footprint model. That's why it's smaller footprint model. That's why it's smaller footprint model. That's why it's a lot faster than some of the trillion a lot faster than some of the trillion a lot faster than some of the trillion parameter plus models. And there are parameter plus models. And there are parameter plus models. And there are corresponding cost savings. So if you corresponding cost savings. So if you corresponding cost savings. So if you have something like on these data sets have something like on these data sets have something like on these data sets something like 20 30 steps of something like 20 30 steps of something like 20 30 steps of interaction, you're looking at 80 cents interaction, you're looking at 80 cents interaction, you're looking at 80 cents per task versus $230. And that makes a per task versus $230. And that makes a per task versus $230. And that makes a big difference. And so the models are big difference. And so the models are big difference. And so the models are getting cheaper uh in that sense and you getting cheaper uh in that sense and you getting cheaper uh in that sense and you can launch them at at scale.
-
can launch them at at scale. can launch them at at scale. Um so hopefully at a at a high level Um so hopefully at a at a high level Um so hopefully at a at a high level I've sort of convinced you that there is I've sort of convinced you that there is I've sort of convinced you that there is something off with this argument. You something off with this argument. You something off with this argument. You know this is where where what my goal know this is where where what my goal know this is where where what my goal was. This is where I started. AI agents was. This is where I started. AI agents was. This is where I started. AI agents are going to be the drive primary are going to be the drive primary are going to be the drive primary drivers of action. um that seems sort of drivers of action. um that seems sort of drivers of action. um that seems sort of an uncontestable statement because you an uncontestable statement because you an uncontestable statement because you know the underlying intelligence of the know the underlying intelligence of the know the underlying intelligence of the models is becoming larger and larger. Um models is becoming larger and larger. Um models is becoming larger and larger. Um there is a certain gain of productivity there is a certain gain of productivity there is a certain gain of productivity and efficiency and you know just ease of and efficiency and you know just ease of and efficiency and you know just ease of life that you get. So makes sense. But I life that you get. So makes sense. But I life that you get. So makes sense. But I think this hypothesis that suddenly think this hypothesis that suddenly think this hypothesis that suddenly overnight 30 years of infrastructure overnight 30 years of infrastructure overnight 30 years of infrastructure that was built layer upon layer for that was built layer upon layer for that was built layer upon layer for human consumption will in what two five human consumption will in what two five human consumption will in what two five 10 years be reinvented is I think a 10 years be reinvented is I think a 10 years be reinvented is I think a fantasy. Um but but this is ultimately fantasy. Um but but this is ultimately fantasy. Um but but this is ultimately what you want right? Ultimately you want what you want right? Ultimately you want what you want right? Ultimately you want an endpoint that um you know some higher an endpoint that um you know some higher an endpoint that um you know some higher level entity can go to and I say I want level entity can go to and I say I want level entity can go to and I say I want you to do X. There is some task maybe I you to do X. There is some task maybe I you to do X. There is some task maybe I give you uh that task description give you uh that task description give you uh that task description natural language maybe I have some natural language maybe I have some natural language maybe I have some programmatic description with parameters programmatic description with parameters programmatic description with parameters and there's you know some endpoint I and there's you know some endpoint I and there's you know some endpoint I should be able to do it. So how will the should be able to do it. So how will the should be able to do it. So how will the web be identified? My answer is just web be identified? My answer is just web be identified? My answer is just extrapolate this trend. This is where we extrapolate this trend. This is where we extrapolate this trend. This is where we stand today. Accuracies are getting stand today. Accuracies are getting stand today. Accuracies are getting higher, benchmarks are falling, higher, benchmarks are falling, higher, benchmarks are falling, latencies are getting smaller, costs are latencies are getting smaller, costs are latencies are getting smaller, costs are falling. And that trend I think will falling. And that trend I think will falling. And that trend I think will continue. And the way we will identify
-
continue. And the way we will identify continue. And the way we will identify the web and in in in some sense this is the web and in in in some sense this is the web and in in in some sense this is depressing but this is true. We will depressing but this is true. We will depressing but this is true. We will just pile on another layer of mess on just pile on another layer of mess on just pile on another layer of mess on top of the mess that the web is which is top of the mess that the web is which is top of the mess that the web is which is which is to say you know you will issue which is to say you know you will issue which is to say you know you will issue a task behind the scenes. There will be a task behind the scenes. There will be a task behind the scenes. There will be hundreds of browsers that are pretending hundreds of browsers that are pretending hundreds of browsers that are pretending uh and clicking buttons just like humans uh and clicking buttons just like humans uh and clicking buttons just like humans would. Um and they will give you a would. Um and they will give you a would. Um and they will give you a result in a structured format. It will result in a structured format. It will result in a structured format. It will cost less than a penny. It will run in cost less than a penny. It will run in cost less than a penny. It will run in your browser. Sometimes it will, you your browser. Sometimes it will, you your browser. Sometimes it will, you know, run in less than 100 milliseconds. know, run in less than 100 milliseconds. know, run in less than 100 milliseconds. And at some point you will say, "Yeah, And at some point you will say, "Yeah, And at some point you will say, "Yeah, that's an API. Like why do I care?" And that's an API. Like why do I care?" And that's an API. Like why do I care?" And that's how we'll we'll identify the web. that's how we'll we'll identify the web. that's how we'll we'll identify the web. So thank you.
Summary
The discussion centers on the future of AI agents interacting with the web, moving beyond human clicks to automated actions. While many believe APIs will facilitate this, the speaker argues that AI agents will instead "identify" the web, particularly its less popular "long tail," through their own models rather than relying on readily available APIs for every service.