Rebuilding the web for agents — Liad Yosef, MCP Apps

AI Engineer · 20 min · 226 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

00:01[music]
00:12Hi everyone.
00:14Uh hello.
00:15Hello.
00:16So we're going to talk a little bit about the agentic web and more specifically what it means and how do we make the web ready for agents.
00:23Um one disclaimer um I built this talk yesterday so it might be outdated because uh things are moving really really fast in this uh in this space and um I think I need to introduce myself just to um to get some context.
00:36So I'm the co-creator and maintainer of a spec called MCP apps.
00:40MCP apps that's the underlying spec be behind uh CHP apps cloud apps copilot GitHub every chatbased app that you so that you see is based on the spec MCP apps in the MCP committee I'm also the co-founder of a company called Aura where we research the agentic
00:59human interactions and I built and led the agentic storefronts in Shopify so the topic of agentic web is really close to my heart um and a a little bit primer about MCP apps if you're not familiar.
01:13Um MCP apps are actually this is actually the spec that paved the way to Agentic Web.
01:18So MCP apps were released as a spec and as a standard few months ago um with the support of Claude first as the first client and then all clients followed.
01:28And if you play with any chatbased app and you pulled any uh visualization and interaction UI layer from an MCP server, you probably used MCP apps.
01:40And the good thing about MCP apps is that everybody wins because if the servers or the providers can send UI chunks into the chats, then the apps gain their brand and identity.
01:55they get they get to keep their UI and not being reduced to a database or just textbased uh information.
02:03The users gain the trust and familiarity because if you uh ask JGPT, hey, book some hotels for me and you see Booking.com, then you know it's Booking.com, right?
02:11You know who you're interacting with and the host or the chats gain uh access to the world of capabilities.
02:19Um you know, a few months ago people ask, yeah, why won't OpenAI just build everything out of scratch?
02:24But open eye will not negotiate uh deals with hotels or support users that want to change seat in a venue.
02:32Um we need those services and MCP apps actually solve for this uh for this last mile interaction.
02:41Yeah.
02:42So um uh this is the this is the benefit that we get and MCP apps solved our last mile of interaction.
02:51Right.
02:51So what does it mean the last mile?
02:53Last mile is as agents become better, as models become better, they can do things more autonomously, but we are still in the end of the chain as humans.
03:03We still need to be able to choose the hotel or to view a 3D model or to choose a seat at the venue.
03:11all those last mile of interaction and that can change and um this is an example from uh um oh sorry yeah so this is from like cloud um APR on MCPF this is how it looks like right so when you um when you're embedding apps inside chats
03:29you get this unified experience of how UI feels inside the same context of a chat and as you can see in the same conversation you get data from booking you get data from all trails you can get data from everything you um
03:43uh you want to interact with.
03:44Right?
03:45So this is this is already happening.
03:47MCP apps is already widely supported across every uh chat agent except Gemini, but that's coming.
03:53Um and this is already happening.
03:56And the interesting part is that it brings us to what we call the agentic web.
04:02So what's what is the agentic web?
04:05The agentic web is not a web of agents.
04:07Uh it's not the web as we know it today.
04:09It's a it's a shift.
04:11It's a change.
04:12It's a a change in in the way that we see websites and the way that we see browsers.
04:17Because up until now, we spent 20 years, two decades perfecting this experience.
04:26Meaning that if I want to somehow um do a project or fulfill a task or plan an anniversary, I have to open tabs in my browser and I have to go through these tabs and I have to convey my intent
04:39differently to each of those services, right?
04:41Meaning that if I just want to plan an anniversary, then I need to learn Google's UI and Amazon's UI and booking UI and the other booking UI and the other Amazon UI.
04:53All that just convey the same intent to those interfaces and every company perfected this user flow.
05:00But we don't need to do it anymore, right?
05:02We can just take these interfaces and break them into atoms because we don't need I don't need 99% of bookings dashboard.
05:11I don't need 99% of Airbnb's dashboard.
05:14I definitely don't need Jira's dashboard.
05:17But I do want to convey my intent to those services.
05:20So why can't I let my personal assistant use that?
05:25So if I have a personal assistant, I can just take these atoms and it composes these atoms and say, "Yeah, I see that you have an anniversary coming.
05:33This is this is the view from Google that says that you have an anniversary coming."
05:36So I can book I can I can buy things for you.
05:39I can book your hotel.
05:40Now Claude knows me.
05:42So it it knows that I prefer to book hotels in uh in the nature, right?
05:47So it knows to pull the map from booking.
05:49I don't need to think about it.
05:50And what booking and Amazon and Google benefits from this is that they have this integration layer that they don't they don't need to develop because cloud has the context on for me, right?
06:01So booking doesn't need to develop integration with my calendar.
06:03Amazon doesn't need to develop integration with booking.
06:06So it's a win-win.
06:07This is the view that we're uh aiming for.
06:10And this is very different from what the things are going which is this which is basically browsing into different services just to stare at different text boxes that ask what I want to do there.
06:25But no one is going to do that in few in a year.
06:28No one is going to um browse to Amazon's agents and and Etsy's agents and Expedia's agent.
06:34I don't want to do that.
06:34I have my own personal assistant.
06:36I don't want I don't want to use your agent.
06:39I want to use my agent.
06:41Okay.
06:42Now, if we think about it, it means that websites as the source of truth are something that's going to go away because why would they need the website?
06:53Why would I need to open the tab?
06:55Why would they need the information that the website is working really hard to convey to me?
07:00And the immediate response to that is that yeah, we have browser agents.
07:03We have computer user agents.
07:04have very smart agents or assistants that can browse the web for us.
07:09But that doesn't make sense because that that's that's a faster horses solution.
07:13Why would I let an agent that doesn't know anything about the human perception limitations work with uh filters and pagenations and sorting and all of those UIs that we spent decades perfecting for us
07:27um just for it to fulfill its task.
07:29So Google has a um Google is very bullish on the other end of the spectrum which is co- browsing and and websites and they have a protocol called or standard called WebMCP.
07:39WebMCP says okay instead of the agent taking screenshots of the websites and trying to figure out what's happening let's have the website expose some tools in the website JavaScript tools and the agent will walk with it.
07:51And you can see here an example of Gemini in Chrome that um buys stuff for me or uh finds uh deals for me as I browse the web, which is very nice, but it's very simple.
08:02What happens if we push Salesforce dashboard to Gemini on Chrome?
08:07Why would I want that?
08:08Why would I want to browse to Salesforce dashboard with Gemini to click buttons for me in a dashboard that I don't even want to use?
08:16I mean, this is definitely not what we want.
08:20And the shift is that assistants are becoming our entry point to the web, right?
08:26We see it everywhere.
08:28Every major lab wants its app to be the everything app.
08:32And the shift is happening.
08:34And friends of mine that connect Jira's MCP to their IDs, they don't go to Jira's website anymore.
08:41Right?
08:41And my mother uses CHP for everything.
08:43If she could book a doctor's appointment using Chip, she won't go to that clinic's website anymore.
08:48and you have a long tail of websites that no one is going to see because it's much easier for me to connect to them using my personal assistant.
08:55So websites will become obsolete, browsers will become obsolete, and personal assistant will become our only gate to the web.
09:01And I know what it sounds like.
09:03It sounds like old men yelling at a lit literal cloud, right?
09:07I mean, um, yeah, they're not going to go completely obsolete, but we see it with the younger generation today.
09:14My 9-year-old goes to Chpt.
09:17It doesn't go to Google when it when it wants to search for stuff.
09:20I was uh I was traveling to Georgia, the country in Eastern Europe, and there was an older lady there that asked me to take a picture of her, like 70 year old, and on her phone, she had three apps,
09:30WhatsApp, camera, and Chip.
09:32That's it.
09:33The shift is happening.
09:35Um, and very smart people like David Ker from Sentry said a year ago, I will take this bet against anyone who thinks websites are going to be obsolete in 25 years.
09:46That was one year ago.
09:48Few days ago, David Kramer published designing for agents which says we have to recognize that interaction with products at century will no longer be exclusively through our web application
09:58and we have to design for API first as a surface.
10:02So everything is going headless.
10:04Salesforce recently went headless which is very important because Salesforce main differentiator from its competitors is its UI.
10:12It's the way that it conveys everything to the user.
10:14But it went headless.
10:14Cloudflow went headless.
10:16Sentry went headless.
10:17Um this is a a tweet by Cloudfl CEO saying that agents traffic surpassed uh human traffic in the web.
10:25And we have this spectrum of interaction because if I if my agent is autonomous, I don't need this last mile of UI, right?
10:31If I have an open claw, I just send and I just say yeah organize my emails.
10:35It comes back said good.
10:37But if I want to book a hotel for my honeymoon, I probably need this less mile of interaction.
10:41So we call it the nearly headless web, right?
10:44It's not completely headless.
10:46Um the agents will interact with your web head with your um website headlessly, but then it will bring back these UI resources for you.
10:55So user experience as we know it is being split.
10:59Now it's the agent experience of your website plus how a user experiences the agent that experiences your website.
11:06But 90% is going to rely on agent experiences which means your website needs to be agent ready for everything.
11:15It needs to be agent ready for your agent.
11:17If I'm booking I need to make bookings data ready for bookings agent.
11:20I need to make it ready for my customers chpt or my customers open claw and I need I need to be ready for my human customer who doesn't have an agent but still wants to browse to booking.com
11:31right so I need to to be agent ready and what we find found out is that agent ready means a lot of things so we heard a lot about AEO SEO geo how to get discovered but discover discovery is only the first
11:44the first step because once an agent knows about you it still needs to know what you are and how to interact with you and how authenticate to you and how to headlessly pay you, right?
11:53And actually when when we built uh it's an anecdote, but when we built analytics for our product, we asked Cloud Code, what's the best analytic service?
12:02And it recommended Postto, which is an analytic service.
12:05And we said no, we prefer Mixpanel because we know Mixpanel.
12:07We worked with Mixpanel for a decade, we know how to work with it.
12:10And Cloud Code insisted on Postto because it said Postto has better MCP and API and I can integrate to it better.
12:17So we don't have brand loyalty, right?
12:18We went with postto hog if that's what cloud code recommend but it made us think about mixpan mixpan spent a decade perfecting their UIX and developer experience we just left it just because cloud code prefers posto and it will always happen her for example
12:31if it uses your product and it needs to spin up a browser to use your product it will remember that and next time it won't go to your product anymore
12:39right so at aura which is an agentic web research lab we started researching uh we raised some raised some capital and we started researching what does it mean for the web to be ready for agents
12:52and we have these um readiness benchmarks which is interesting so you can run it's free you can go to aura.ai AI you can run any website you get this
13:01score and benchmarks according to a lot of protocols and best practices and you have agentic feedback.
13:07So the agent actually returns the feedback about your uh your website and we have this leaderboard of how companies and products rank according to this benchmark and we started mapping the web and everything was nice.
13:24But we hit one insight or one uh unexpected result.
13:31We found out anyone here heard about lm.txt txt lm.txt that's like the de facto standard to be agent ready.
13:40You say yeah if you if your website publish an lm.txt agent know how to interact with you um and you have o.md and pricing.mmd and x42 and a lot of standards and we found out that almost 50% of the website that we tested that we ran
13:54published lms.txt but none of the agents that we ran on this website actually used llm.txt txt.
14:02Actually, almost all the agents went straight to the docs page and then the homepage.
14:08And the 40% that did use lm.txt use it only because the docs pointed out that there's a file called nmst txt that they need to to use.
14:16So then it hit us.
14:17We said it doesn't make a lot of sense for us as humans to define to agents what they need.
14:24No one is doing it anymore.
14:26Even OpenAI, they don't publish best practices for tools anymore because they say that every time they publish best practices, the models become better and these practices become obsolete.
14:35Six months ago, the best practice for an MCP server was yeah uh have three paragraph descriptions so agents will know how to interact with you and now three lines are enough.
14:44So best practices become we need the agents to define what the agent need, right?
14:48We need agents feedback on this website.
14:51So we built Aura's journey and Aura's journey that's a really cool that's also free.
14:58You can go to journey.ora.ai and you can run on any website any intent with any agent and see the path of the agent as it tries to interact with the website.
15:09So for example atio.com choose cl codec code and we run it on the website and you can see in real time how the agent goes and what it tries to look for in the website
15:22and we we did it tens of thousands of times just to understand what agents really look for when they look when when they try to interact with with websites.
15:33And the cool thing is not running just one harness is running multiple harnesses.
15:39You see here cloud code and Eve which is Versel's harness um and CH GPT on the same website with the same intent.
15:45Okay.
15:45So this is cloud code.
15:46This is haiku.
15:48This is Eve.
15:49See how different the agent journey looks like.
15:52And this is CHP.
15:55So CHP could find results better.
15:58See you can just see the the the journey across across the website.
16:02And we need to understand why.
16:03We need to understand why does it happen?
16:06Why websites publish o.md files but agent don't look for o.md files and what do they look for?
16:13So in aura um you can actually uh go to any business business business question.
16:19So for any domain you have like business goals or or questions and you can see the paths that the agents are taking.
16:26And the uh interesting part of it is that okay now we know how agents interact with website.
16:33What's the next big big milestone for the agentic web and for those of you who heard the the previous talk?
16:39Um it's discovery, right?
16:42But it's not the discovery that we think of.
16:44It's not SEO, it's not GEO, it's not AEO because we have to remember that every revolution came with its discovery layer, right?
16:53The web revolution came with search, um mobile came with app stores, social came with feeds.
16:59What is the discovery layer for aic resources?
17:02What is the discovery layer for MCPS?
17:03What is the discovery layer for open API.json?
17:06What do we even look for?
17:07I mean, maybe Airbnb has a better MCP than than Booking.com, but Booking.com has better API than Airbnb.
17:13So, how do we do it?
17:15Um, web search, classic web search, it's not enough because um it it's based on human SEO and human uh page rank and it's not flexible enough.
17:27custom registries like per agents or per chats, they're not enough because they're closed and they're uh um and and they require every app to to submit itself to these reg registries
17:38and a central registry like an MCP registry.
17:41That's not enough because who will do the curation?
17:43What's the governance model?
17:44How do we decide which resource comes to that registry?
17:48There are emerging standards around it.
17:50There's AI catalog.json JSON which is a standard by entropic openai, Google, MCP and A2A which standardizes how a website exposes itself to agents and there's the agentic resource discovery standard
18:03which is by all these companies and more which is basically standardized how uh a discovery layer or or a directory uh expose itself to agents.
18:13So we built or directory right because we're a research lab for the agentic web.
18:17We built this and in or directory we take all the domains that we scanned or scanned the results and we put it in a directory that agents can actually query right
18:27um we expose the AI catalog.json JSON files.
18:30So you can see here for example for monday.com you can see that we generate this JSON file that basically tells agents yeah this is the MCP server for Monday this is the API server from monday.com
18:42and you can just the agent can just query that right so for every domain it can just goes to go to aic catalog.json JSON and we expose the registry the directory itself.
18:53So for every entry on the directory we can uh we can tell the agent yeah for example versel these are the agentic resources um and this is and this is how you access them and obviously
19:05this directory is fully ard or aentic resource discovery compliant.
19:09So any agent can just query that directory with any with any query.
19:13Right?
19:15So this is uh this is aura.
19:16directory uh if you want to take a look all of it is part of the aura multiverse journey directory and the ranker.
19:23And we found out some very other interesting insights.
19:26For example being agent ready and being human accessible is very similar because LLMs when they come to your website they're like um users or people with vision disabilities.
19:38they don't see your website.
19:39They need other signals to understand how to walk with your website.
19:42And making your website human accessible helps agent accessibility and vice vice versa.
19:48So to wrap up, these are MCP apps.
19:51This is what I've been working on in the past few months.
19:54And this was the last piece for the agentic web.
19:57This actually brought the agentic web.
19:59And now agents start to roam the web.
20:02And agents need different things.
20:03They need different roads.
20:04They need different infras.
20:06But we don't want to rebuild the web for agents.
20:08We we want to make the web agent accessible.
20:11So let's just make sure the web is prepared for when agents are coming.
20:19Thank you very much.
20:22[applause]