1015: Browsers and UIs Are Dead. Everything Is Chat ![rw-book-cover](https://wsrv.nl/?url=https%3A%2F%2Fmegaphone.imgix.net%2Fpodcasts%2F5197fe5a-42f7-11f0-affd-87d9985a1760%2Fimage%2Fc86a54acd72683732c4773e25bf0ae14.png%3Fixlib%3Drails-4.3.1%26max-w%3D3000%26max-h%3D3000%26fit%3Dcrop%26auto%3Dformat%2Ccompress&w=512&h=512) šŸ”— **Source:** [1015: Browsers and UIs Are Dead. Everything Is Chat](https://share.snipd.com/episode/c44eb092-97d6-4a0b-9229-a8fd07f1ed2c) ## Highlights - **Chat Is Another Interface Not The End Of The Web** - Chat interfaces will become another interface layer, not a complete replacement for every UI. - Wes argues for coexistence: sometimes no UI, sometimes voice, sometimes clickable WebMCP tools exposed as actions on sites. Transcript: Wes Bos And I'm doing my talk on, like you said, agentic interfaces and like generative interfaces. Because there's a lot of talk going on right now saying that like either like the web is dead or the web is going to be headless or chat is the final interface or these little components That are inlined inside of your your chat interface are like going to replace websites and i'm in like a weird spot because i i so heavily agree and disagree at the same same point and that's Kind of be what what my talk is going to be about as well as as web mcp which is we had a whole episode on web mcp and i don't know if this is necessarily about the tech anytime you use the word Mcp everybody's like skills are better i'll give me a cli and that's all i need and then everyone's like no mcp is better and like i don't give a shit about the like actual specification Because whatever ends up being the spec, I don't think we're there yet. Yeah, I'll just prompt it to convert it to that end spec. I don't necessarily care about the specs here. I more care about like, what does user interface look like, from a standpoint of both agents using it as well as us and the people using it. Speaker 3 Yeah, I'm really interested to see your talk because I've done so very little research on any of this stuff. But I did just see yesterday, Hyundai announced that they were having a new future style interface that they're developing that included things like telling your car to turn my heated Seats on and stuff and it was all very llm based ui instead of a car and i like my hyundai a lot yeah i love my my ui for all that thing and i'm not necessarily one of these people that's like You've got to have tactile buttons for everything i know that's a really strong opinion but I can't imagine that everything being a chat interface is the right car yeah especially in Wes Bos A car where like you have to pay attention to the road turn my turn signal on car yeah you're like in a thread of chats yeah so that's what my talk is about is like sometimes you want no ui at All. I think about my smart home. In a lot of cases, I shouldn't even have to tell it when to turn on the lights. It should just be learning from context. Sometimes you want a voice UI as well. Then when you get to the actual UI, you want to view a shopping list of what you have. Sometimes you want to be able to just like type into a chat and like sort of batch process that. But then sometimes you want to be able to click the buttons. And that's what MCPUI is. And like right now they're just like inlining these like little components inside of your chat. And it's kind of slow but like not knocking it for that but i i kind of feel like like the web mcp which is what this is is it allows you to expose the tooling of your website as mcp tools so like You visit a website like i have my example is like i have a grocery app and you can like add an item, mark as, as whatever, reorder, add a store, you know, all the, all those typical things. So you expose all of those tools via WebMCP. Yeah. And then like an agent, like I have Claude hooked up to the WebMCP browser. Then I'm able to just use the website normally. And then I am also able to just like prompt it from my clot or chat GPT or whatever it is that I want. And then it can also be hooked up to other things as well. Because like, you might be looking at being like, that's just like a chat bar you added to your website. And like nobody wants a chat bar added to every website ever. That's not a good experience. I wanna use my agents and hook it up to that website, or I wanna hook it up to my Google Calendar, and my groceries, and my fridge camera, or all that, whatever it is. So I think that that's the moves, because I'm calling it clicks and clankers. Because sometimes I wanna click, ([TimeĀ 0:01:02](https://share.snipd.com/snip/8ade7185-d70d-4354-99e8-aa7c8ebe47e3)) - **Expose Actions Via WebMCP For Agent Control** - Expose website actions as WebMCP tools so agents can perform them without a clumsy chat bar UI. - Wes describes a grocery app exposing add, reorder, mark, and store actions to let agents operate your site directly. Transcript: Wes Bos And like right now they're just like inlining these like little components inside of your chat. And it's kind of slow but like not knocking it for that but i i kind of feel like like the web mcp which is what this is is it allows you to expose the tooling of your website as mcp tools so like You visit a website like i have my example is like i have a grocery app and you can like add an item, mark as, as whatever, reorder, add a store, you know, all the, all those typical things. So you expose all of those tools via WebMCP. Yeah. And then like an agent, like I have Claude hooked up to the WebMCP browser. Then I'm able to just use the website normally. And then I am also able to just like prompt it from my clot or chat GPT or whatever it is that I want. And then it can also be hooked up to other things as well. Because like, you might be looking at being like, that's just like a chat bar you added to your website. And like nobody wants a chat bar added to every website ever. That's not a good experience. I wanna use my agents and hook it up to that website, or I wanna hook it up to my Google Calendar, and my groceries, and my fridge camera, or all that, whatever it is. ([TimeĀ 0:03:52](https://share.snipd.com/snip/0d1aadae-16da-4fab-982c-165c4fd4738f)) - **Clicks And Clankers Mixed Interaction Model** - Use a mixed interaction model called "Clicks and Clankers" where users sometimes click and sometimes prompt. - The hosts emphasize combining clickable components inline with agent prompting for different tasks and user preferences. Transcript: Wes Bos So I think that that's the moves, because I'm calling it clicks and clankers. Because sometimes I wanna click, sometimes I just want to prompt. Yeah. And sometimes I wanted to just figure out what the UI should be for me. Right? And sometimes just having a visual, even if you are prompting, can change things. Speaker 3 I mean, we saw this in our calendar challenge, where even though my calendar did win the calendar challenge, there was no UI for it in terms of a calendar. There wasn't a UI. And then like a lot of people said they needed that. Yeah. Yeah. Yeah. I think, yeah. Scott Tolinski I mean, I think in an ideal world, it's cool that, okay, let's say everything does happen in chat, but I think that's too idealistic thinking. I think there's a time and a place for the MCP apps and a time and a place for a full-on UI. Think like power users, think just like maybe you don't want to have individual MCP apps, but with WebMCP, you expose all of the actions that you can perform, and then an agent does it On your behalf. ([TimeĀ 0:05:11](https://share.snipd.com/snip/756f243f-dd79-43a5-8181-6ae32cceafe2)) - **Shopping Needs Visual Discovery Not Just Chat** - Shopping often relies on discovery and visual browsing, which a pure chat agent may ruin. - Wes prefers filtering, sales and visual recommendation browsing and doubts chat alone can replicate that pleasure. Transcript: Wes Bos Right. Like it's all algorithm based. And I just want to like sit there and consume or like if I'm shopping, like a lot of people are saying like the future of shopping, say I need shoes. And then the boom, the shoes are shipped to your door. And like, that just reminds me of the Soylent. Were you around for when Soylent happened? Yeah. All these tech bros were like you just drink this goo and you don't have to think about eating anymore. I'm like no I enjoy looking at like what's on the website. I like going to filtering for what's on sale and I like that. ([TimeĀ 0:07:00](https://share.snipd.com/snip/13fe414b-ce1f-430f-b157-3e90943dbe29)) - **Agent Interfaces Could Rebuild Walled Gardens** - Centralized chat-based agents risk consolidating commerce behind a few guardians who control surfacing and monetization. - The hosts worry big platform agents could favor paid partners or insert dark-pattern upsells into responses. Transcript: Wes Bos Navy and it's like, then they're going to. Well, that's another thing is like, are we, oh yeah, now the open web is now guarded by these four companies that you have to pay for. Scott Tolinski If you don't have an app, then your shopping might not surface in the chat interface. Or you get the free version, but they're going to break it to you. ([TimeĀ 0:08:22](https://share.snipd.com/snip/d2d10ee4-f159-4358-884b-9c6c51dc6b64)) - **Built A Personal Expense Categorizer For FreshBooks** - Wes built a personal tool to auto-categorize expenses because FreshBooks lacked AI categorization. - His tool pulled expenses, classified categories, and re-pushed them, including handling Canadian HST tax boxes. Transcript: Wes Bos Like the example I give is categorizing my expenses in FreshBooks. FreshBooks has not integrated any AI stuff at all. And I'm just begging them, like, how do you not know what category this is? Like I had to build my own thing that pulled my expenses out, pulled the categories, categorize them and push them back into FreshBooks. And, or like if it was a Canadian vendor, then it had to check a box for adding HST, Canadian tax, right? That's such busy work that I hate doing. ([TimeĀ 0:10:17](https://share.snipd.com/snip/fc7af83a-4b11-4daa-a96c-9d9ba93ce610)) - **Combine Components And LLM For Generated UIs** - Generative UI should combine developer-provided components with LLM layout to balance brand knowledge and personalization. - Wes notes using a few vetted components plus LLM assembly (Open UI-style) can yield better personalized UIs than pure LLM HTML. Transcript: Wes Bos And this like agentic ui generative ui and the one thing we didn't even talk about is this like generated ui is like do you want to choose what the ui is or do you want the llm to just make a Whole bunch of html for you and the answer is probably somewhere in between you want to give it a couple components that you've made but then it can figure out how to piece them together. There's a whole bunch of like specs. Open UI is one of them right now. Probably somewhere in between there. But like the question is like, can the user or the LLM make a better UI than the company that knows their product really well? ([TimeĀ 0:13:36](https://share.snipd.com/snip/9efab591-11d8-4913-81f1-05787581aaba)) - **Use Agents To Generate Contextual Smart Home Dashboards** - Let agents generate personalized UIs from rich context instead of forcing users to design dashboards. - Wes suggests smart-home systems should auto-generate dashboards and automations from sensors, camera alerts, and historical patterns. Transcript: Wes Bos And as a user, I don't know what I want my smart home dashboard to be. I want you to tell me because you have all the information. You know what cameras have triggered alerts. You know the humidity in my house. You know when I turn lights on and off. And if if i want either automations or like a dashboard for my smart home i want the we have so much context about my house in my home assistant yeah i want the the ai to just programmatically Make a dashboard for me given what it knows about what has happened recently and whatnot. You know, like, oh, hey, we figured, oh, you always turn on your lights at 9 a.m. We'll do it for you or maybe we'll pop a button up here so you can just tap it while you're walking by. ([TimeĀ 0:15:03](https://share.snipd.com/snip/553bd373-fdaf-4b86-af5c-ef3428d8ab47)) - **Agents Excel At Automating Repetitive Busywork** - The killer agent use case is surfacing and automating repetitive, boring tasks rather than abstract creativity. - Examples: auto-turning lights at habitual times and surfacing recent camera alerts or humidity anomalies. Transcript: Wes Bos Has happened recently and whatnot. You know, like, oh, hey, we figured, oh, you always turn on your lights at 9 a.m. We'll do it for you or maybe we'll pop a button up here so you can just tap it while you're walking by. And I think that's going to be the killer use case is just like adding on UI or generating what you see in like, like a smart, like we've had this with algorithms, but like even a step further. Scott Tolinski Right. I think if it gets it right, that's amazing. Yeah. Can I get it right? We'll see. I mean, that's, that's the future from Iron Man. That's Jarvis. Well, he has this AI that he can literally talk to normally and like it knows his vitals, it knows everything and um to to what you're saying is it has all the data and is making the right Choices but if that's that's the big if is will it be able to make those choices and can we reliably depend on it and ([TimeĀ 0:15:36](https://share.snipd.com/snip/62faa41b-fefb-4f82-991b-0bc261e60e74))