Inside the Cerebral Valley AI Summit: What to Expect Next Week

0:00This episode is brought to you by Forethought. Most companies build their customer experience in pieces. Sales in one system, support in another, and onboarding somewhere else. Forethought brings it all together. Forethought is an AI system made up of advanced agents that handle sales, onboarding, support, and retention. Each team manages its own agents. The customer sees one unified experience. Forethought powers over a billion interactions every month for brands like Scale AI, Cohere, Airtable, and Upwork. Learn more at forethought.ai.

0:42Welcome to the Cerebral Valley podcast. I'm Eric Newcomer. I'm here with Max. Hey, Eric. And James. Happy to be here. Looking forward to the conference next week in London. Next week. For our returning listeners, you know I'm Eric Newcomer, the author of the Newcomer newsletter. And I have Max Child and James Willsterman, the co-founders of Volley and my co-hosts of the Cerebral Valley AI Summit. Some of the highlights will be in this podcast feed. They'll all be on our YouTube channel. We're going to hear from Dara Khazr-Shahi, the CEO of Uber, Dylan Field, the CEO of Figma, Alex Kendall, the CEO of the self-driving car company Wave, Victor, the CEO of Synthesia, the CEOs of Lovable, Granola, Orbi, Crusoe, PhotoRoom.

1:28I'm not going going to list them all. Lots of cool CEOs next week. And so Max, James, and I assign ourselves to come up with three themes each that hopefully are not just agents are important and sort of talk through what we think some of the big themes are going to be in there for sort of think about what we want to hear from our panelists. Who wants to go first with one of their themes? Yeah, so I think a big theme, especially since we have Dylan from Figma speaking, will be this evolution of the role of the designer. And maybe this will apply to other roles as well. But I think we're already seeing it a bit at Volley.

2:09Designers being able to make much more interactive, fully executed, programmable prototypes in a way that really multiplies their impact, their speed to iteration. because they don't have to hand it off to a coder because they're yeah exactly they're cutting out that step in that early phase of design or even for like things like internal tools which don't have to be like fully perfect when they launch right we're allowing designers to kind of contribute directly to the engineering in a way that was never possible before and i think that's just got getting started and i think figma's leaning in there as well there was that old discussion back in the early days of streaming movies and tv where they were talking about at Netflix, like they were saying, you know, our goal is to become HBO before HBO can become Netflix basically.

2:57And, um, in the end, they kind of both won, but I guess you could probably say Netflix won a lot bigger based on the stock prices these days. But I think similarly we're seeing internally and a lot of other folks we talked to that there's this question of like, do designers become engineers faster than engineers become designers or that's great. You could include product people as sort of part of the design cadre of like, you know, do designers and product managers become able to create software products faster than people who can already create great software products are empowered and able to design and ship and sort of finish complete products in a way that they weren't in the past.

3:33And I think that we've seen just amazing gains in the productivity of our engineering team. Right. I mean, I think some people are like three to five X as effective as they were a year ago. But I think you're really seeing like zero to one games from the designers and the product folks who are now making websites and making mockups and making internal tools and pages and stuff that literally they could not have done any of six to 12 months ago. So there's a true kind of, yeah, step change happening. And I think obviously Figma and a lot of other folks are kind of on the cutting edge. And sort of lovable from the other end of it.

4:07And lovable, of course. So we're going to have both of them. Yeah. Do you want to explain lovable for our audience? I feel like a lot of people might not be familiar. It's for people like me. It's for no code. People who have no sense of code. I mean, literally, I think you have to imagine you're taking a screenshot of a website. You like sending that to lovable and saying like, build this with XYZ. You go back and forth with it. And it's just like, it's building with you. Is the preferred flow that you take a screenshot of a website, by the way? I don't think they're selling that, but I think people understate that.

4:36Like So much of what's great about AI in every domain is that it's really good at ripping things off. And so Lovable works really well when you're like, I like the New York Times. I like the Substack design. Why don't you borrow from that? Here's my color palette. I mean, you can give it your logos and I'll put them in and I'll take your color palette, but it's really sometimes good at being derivative. Yeah. Well, you got to speak to your personal experience, right? Like you, you tried to build a full website on Lovable, right? And you've never, you should explain this whole experience. What happened?

5:09I was trying to make like sort of a news aggregation site, basically, like get the top headlines in tech news. And so I could build like a comment function and, you know, it was pretty cool. But then I got hung up. You got stuck. I moved this button and like, no matter how many times in ways I'd ask it and, you know, know, you'd be taking your conversation and dumping it into Anthropic and ChatGPT. And still, it just like, for whatever reason, it just like, it got stuck. So I had sort of something visually cool, but then it didn't turn into anything. So you sort of had this last mile problem. I think this is another, I don't know if this is a theme of the conference.

5:47We didn't put it. That's not one of our official themes or it could be because that's good because we have the self-driving car companies and they are the, the sort of test case of the sort of, you have to get the last 1 % and you get the last 5%. And a lot of these products look really good. And then they, yeah, I like it. Last mile problem. That's a good one. The last, the last mile problems, overall theme of the conference. I think, yeah, I think lovable really gives you that last mile problem experience where you feel like you could build 98 % of the website, but it doesn't work without the last 2%.

6:19And you're like, ah, God, I was so close. And the question is, with the next generation of models or the next generation of tools or just lovable itself becomes better, you know, do we cross that last mile? You know, I don't know, James, you've experienced this from the engineering side of things. Someone was tweeting about this idea that coding models would allow finally everyone to complete those like 90 side projects that they have sitting around on their computers that they have just been, you know, tinkering on for decades, but never get finished. And Sam Allman responded and said, like, yeah, so excited to see all those projects get finished.

6:57But I actually think the opposite is happening to me, at least, where it's like 10xing the number of side projects that I have that are unfinished. You're like, now I have 900 unfinished sunbodies. Yeah, now I have 900 unfinished sunbodies. That's pretty sweet. Yeah, so I definitely think this is an issue, but I don't know, every few months or weeks even, it seems like the models get more capable and it allows me to make more progress. So it's less that I'm worried that this is like a never-ending. Someday you'll have 100 projects come out. They're all just like hitting. They're all getting close.

7:28You're building them in parallel. Right, right, right. And then one day I'll just, the last mile will just be completed for all of them. I mean, that is sort of how, you know, AI people can sort of talk about, you know, the singularity or AGI. It's like, oh, we're going to have drug discovery. We're going to have this boom, you know, which, you know, I don't think we're necessarily closed off to. But yeah, there is a sort of like mode of all at once. Yeah, true. The takeoff, fast takeoff. Yeah, fast takeoff. Exactly. Yeah, yeah, yeah. All right. I have a theme that's sort of related to your design one.

8:00Yours is really from like the employee perspective. This is from maybe corporate strategy, but the theme is text box versus product. Like I think chat GBT has introduced this idea that like a company, you know, you can just have the company in a text box and in some ways like lovable is that in some ways Figma, which is like a very intentionally designed product now feels the threat of like, oh, we need our almost text box solution. And so I think there's a real question of the moment is, do people want to interact with their tech products by typing to them? Or is that just sort of the kludgy interim solution before people build great product design around the capabilities of AI?

8:44I mean, I think this actually comes back a little bit to some of the voice discussion we had last time, but our sort of general thesis at our company, Volley, is there's really only like two ways people actually interface with the world, interact with the world, right? And it's basically like with your hands, you know, obviously touching a screen, typing on a keyboard, whatever, playing with a game controller, or by talking essentially, or outputting words, right? And obviously, you know, you're using typing as the sort of the current way we put words into a computer. But like, we believe the future will be talking into a computer rather than typing into a computer.

9:22But I do think basically, if you say, Hey, humans only have two ways to do anything in the world, right. With their hands or with their mouths. Except for some reason, uh, driving a car, which I have hands. I mean, the steering wheel. Okay. We use your feet occasionally for, you know, other things, but I've always found that kind of bizarre that we trust our feet, uh, as the, with only one task, the most important task. Anyway, the point I would be making, besides the pedal-driven Photoshop that James is imagining here, was that to date, most creative software has either been point and click, drag and drop, move a mouse or move your fingers touching, or it's been typing in a box.

10:05And I just think that the headroom for what you can do typing in a box is probably 10 or 100x what we have done to date. because it is one of the major ways humans interface with the world. It's how we're interfacing right now on this podcast, right? I'm much more bearish. Like what we're not doing on this podcast, we're using Riverside right now, which is like Zoom effectively. It's like, we didn't fire this up and be like, what tools do we want? We need a producer to be able to be in the background. We need to be able to toggle a camera. We want, you know, it's just like using a text box is like deriving from first principles, like everything you want for, you know, I was creating cute images for my wife last night on chat.

10:44And it's like, you have to ask it like, oh yeah, what are the different like color, like, you know, graphic types that you probably do. And it's like, you have to reinvent it every time. I just think the text box is not ideal for most of the use cases and great product design. I don't think it's ideal for many use cases that we have developed to date. And I think those aren't going away anywhere. I don't think that like the concept of a Photoshop Figma type experience or, you know, things like that are going to disappear. But to me, it seems clear that like anytime you want a piece of information or you want, you know, a fairly simple action carried out on your behalf, it's much easier to just put that request out in the world as texts, you know, oh, explain to me this thing, go research this thing, go do this simple thing.

11:32You know, even something as simple as like, you know, DoorDash ordering agent, which we don't have today. But if you were like, oh, yeah, go order the pepperoni pizza I always order from this place, like, you know, and have it at my house and under an hour like that is still easier than the 95 clicks it takes to do it in DoorDash if that actually worked. Right. So I think there's a sort of a U curve of complexity where if you get really, really high complexity tasks, then of course, you're going to want to use your hands and drag and drop and touch things and so on. But there's lots of low complexity tasks or even medium complexity tasks that you could do fairly easily with text or with your voice and it's almost like you can do it faster that way because yeah it's much faster yeah like you don't have to like learn the interface it would be much faster if siri actually worked and i was driving home from work and i was like hey like order me the you know i get like for for dinner or whatever you know and they'll be like oh yeah the foot from this place it's like you know it's 15 bucks i'll be like yeah done right if that worked that would be better than using doordash i think and i mean taking the riverside podcast recording application as an example here there aren't that many buttons like we kind of just send out a link and join oh my god no they're way more than anything like i feel like i do use them although i think like if there was a little voice agent or text box here where you could be like you know accessing the deeper features of riverside then you might actually use them the beauty is it's suggesting functionality that i wouldn't even necessarily have imagined you know right right Yeah, I mean, presenting those affordances on screen visually is still valuable.

12:58We're not saying get rid of that. We're just saying doing stuff is easier. Isn't it a pretty big coincidence that we've decided that text is a great way to create products just as that's the most straightforward way to build them? I don't know. It's not a coincidence. It's literally how humans communicate, right? I mean, you use Slack. I know you use Slack because we're on Slack together, right? You text message your friends. We text message each other all the time. writer your product is words in a box you're literally your product is an email newsletter right i know i know how many times a day do you use text in a box in some way or another right i've been seeing a resurgence of that uh graphic of like craigslist where oh yeah yeah the un the unbundling of craigslist unbundling of craigslist oh my god yeah like and then you know do we is is that going to happen with chat gpt i think that sort of gets to some of the the questions that we're asking here i'm curious if you guys have any ideas of like what do you think would be unbundled from chat gpt is it going to be like this therapist use case or like is it is it just the most popular use cases or is it the most valuable use case like what startups would you actually build to unbundle chat gpt that's a good question i mean obviously you would you would want something that's reasonably popular.

14:21I do think just off the top of my head, I think the therapy use case is probably one that will be unbundled because I think that there's probably going to be some sort of like AMA compliant version of a GPT therapist at some point, you know, some sort of like actual doctor endorsed or doctor founded or, you know, psychologist approved version of the sort of GPT therapist. And I think that that will probably unbundle some of that use case. I do think like you and I are obviously sort of in the early days on this, but we both have kids. And I think that childhood education use cases from GPT will definitely get unbundled in some way where like a lot of parents are talking to GPT now being like, explain gamma rays and x-rays to my kid because they just watched Incredible Hulk or whatever.

15:02Right. And there's like definitely some sort of very focused educational, I think, use case. I just think everything that's useful, everything that's useful, either people will train it better or they will build a product around it that suggests what features are hidden inside of ChatGPT. Instead of having this thing you need to pull out, it will market to you. Oh, here are the different things that you can do in this category. And there's value in the marketing alone, right? Because the marketing alone, just to have somebody yelling at you, you can do this cool thing here. I wouldn't have thought about it unless you're telling me.

15:39The problem is ChatGPT, it's like, do anything. That's too broad. I do think we're talking about separate things. I guess I would say I think a lot of these unbundlings, like James and I are doing an unbundling of GBT where we're building Dungeons & Dragons, GBT wrapper in some sense. You still want the text input and the voice input to be a big way that people play the game. right you might want them to be able to click on things you want to be able to show characters on screen you want to be able to make cool art of the monsters and environments dnd is an argument for the text box rather than the product as i'm sure i mean it i'm just saying that i think more i think that's sort of the first order reaction and that then it's like oh we need to go to move back even with the educational use case i still think you even if you have a bunch of stuff on screen like my daughter's learning to read right now you want to show the different sounds and how she's sounding them out and she has to be able to tap, you know, the versus the bar or whatever.

16:34But like you also want in a, in a sane world, her to be able to say, you know, rat versus cat and have it be like, yeah, you said rat. That was correct. You know? So, you know, and she can't type yet, but there is obviously both text input and touch input, which is kind of the point I was trying to make up. Yeah. And I like your overall frame. The point is that chat GPT has unlocked what was a very natural way of communicating that we couldn't with machines before. Right. Yeah. It just unlocks a lot of new opportunities, more exploration in a lot of ways, because you can just ask the thing if it can do the thing that comes in your head versus before you had to like dig in through menus, as James said.

17:10Can I share? Can you guys see? Yes. The Anthropic use case map here. I actually haven't seen this. This is super interesting, This is, oh, is this their agent guidebook? This is how they built their multi-agent. Oh, I was going to read this today, actually. Yeah. I'm excited. Then at the bottom, They said embedding plots showing the most common ways people are using the research feature. So not necessarily just generic clod. But there was some really interesting stuff here like translating documents between languages, developing advanced automated trading systems and strategies, researching pharmacological and performance handing substance innovations.

17:49So I think to me, all of these are like good candidates, right, for being kind of unbundled. Unbundled. to spin it a different way these are all valuable wonky things that companies have a lot of incentive to do the hard way yes yes and in some ways what people aren't doing are the cool but chill consumer things that aren't worth figuring out how to do and they need a company to really tell people oh you should be doing this here it's good for this thing you haven't thought of and like if it analyzed sports betting odds gets its own little thing here like anything that involves like potentially making money.

18:27Monetizing. It's like, oh, I've discovered something fundamental about human beings here and there. Yeah. I mean, I think that like the sports betting odds is a good example because the people who need that or want it like really badly want it, but it's, it's objectively just not that large of an industry. Right. I mean, like sports betting is, you know, there's lots of money in sports betting. Don't get me wrong, but it's not like the size of like, you know, enterprise software or whatever silly stuff we talk about on the show. So I just think that it's an interesting example of like the market size of sports betting odds analysis is probably not that big.

19:01So maybe there's like, you know, one company that could could really execute well on that. But it's also possible that just like individual bettors will fine tune Claude and, you know, figure out how to develop cutting edge strategies. I think Claude could be strong in all these. All right. Next theme. Max, you're up. Yeah, next theme, a lot less sexy, but I do think something that I keep thinking about with Granola, which is one of our speakers at the conference, which is a meeting notes recording and summarization analysis tool, is this idea of like, is there a sort of Microsoft Office suite for the AI era or not, right?

19:39And I warned you, this wasn't a sexy topic. When you said not sexy, I was like, oh man, I underestimated it. I'll try to, I'm a consumer founder, so I'll try to make this as fun and interesting as possible. But we all grew up with Word, Excel, and PowerPoint. And they're not cool. Some of the most valuable things in the world, obviously. Yeah, exactly. Outlook, right. Yeah, I mean, they're literally worth like a trillion dollars if you added them together, right? Right. And they created 40 years of lock-in for Microsoft, you know, and one of the most durable businesses of all time. Right. Okay.

20:08So finally, for the first time in our lives, I feel like people are starting to do interesting things where they're taking a different layer of work. Right. And they're saying, hey, instead of I'm going to do documents as sort of the fundamental layer of work, I'm going to do like conversations and interactions and sort of insights from, you know, from those meetings and those discussions. Right. And I guess the question is, yeah, like, is there a different slice of work where AI can capture, you know, every meeting you have, every email you sent, every time you talk to someone, every, you know, and, and then also suck in all these documents and kind of create like a higher level abstraction of how we do work.

20:48And I think that is kind of what Granola is going for. It's also what Notion's going for. It's also what Glean's going for. It's also what OpenAI is going for. Right. That's what I was going to say. That's what I was going to say. ChatGPT should be that. I mean, if word is everything, right. I mean, I'm in ChatGPT, I'm doing the thinking. And if writing is sort of like an output or a part of the thinking process, capturing some of that in a document makes so much sense. I think I'm pretty bullish on open AI. Open AI. Yeah. Well, the interesting thing is James and I went through a discovery process at our company where we were like, Hey, we need one of these tools that like sucks in everything, every document that volley has ever created or interacted with.

21:28So Dropbox and all these word files and, you know, notion notes we've written to each other and so on and so forth. And we realized like no one really has all the pieces of the puzzle right now. Like open AI just announced that they're going to have almost everything, but they don't have notion docs and we use notion. And then like Glean has like almost everything, but they're missing something else. And then, you know, like granola is very early. They really just have meeting notes. So like no one's put together the sort of coherent, like, Hey, you only need to sign up for this, which is sort of what Microsoft Office did in the good old days.

22:00And so I think OpenAI wants to do that. And I think Notion and Glean both want to do that too. And it's unclear to me who's going to get there first, because I think OpenAI has got a lot of other stuff on their plate. I was thinking this was going to be one of my themes. But since we're talking about it, I do think there's actually kind of that war kind of brewing of who owns the context, right? Like there's a reason Notion doesn't want to be able to be exported into ChatGPT. There's also like this kind of counter force or counter trend where everyone's trying to come up with like the protocol or, or kind of open source version of how you, how you provide context into models, like the model context protocol, right?

22:38What are we calling this theme? Well, the theme is the war for context or something. Open versus close for the eighth time. Yeah, true. I think Google are the, I don't know if they've fixed this recently but uh when i tried to integrate google docs and drive and gmail into open ai a few weeks ago it was like a 60 step process that also involved setting up a google cloud project in order to expose the information and and open ai has a document of like how to do this and it's it's literally like 60 steps i want that so bad google doesn't want this in chat gpt notion doesn't want it in chat gpt at the same time some companies like linear are you know intentionally leaning into mcp because they do kind of want you to be able to edit and interact with your tasks through mcp model context protocol it's like oh that's the new way a neutral way that everyone yeah it's like each other yeah sorry that wasn't for me that was for the poor listener who doesn't know sure sure sure um no mcp will probably be something that comes up a lot at the conference too that's uh that's been a hot topic for sure do you want to explain it for for the audience james sure um i guess it was created by anthropic and it's designed to expose context to agents where they are in operating right so if you think of like cursor has mcp connectors I could go build an MCP server myself and put a bunch of context accessible to that MCP server.

24:12I could expose it in cursor. So now, for example, all of our Volley developers could have access to like a style guide or a Volley recommended documentation or things like that. Right. So it's basically a way to share data into these LL11 agents. Lock in and allows you to be open source to have an agent one place and have all your data somewhere else. Yeah. Yeah. And it also starts to allow you to expose tools to those agents too, right? So your MCP server could expose tools from linear, for example, that would allow the agent to go update the ticket or close it out or kind of interact. So yeah, I think it's just one agreed upon or neutral protocol, right?

24:56For, for exposing this stuff to agents. I mean, it does feel like everyone's kind of gotten wise from the last few generations of these corporate battles over who owns, who owns the data. We don't need antitrust. We just need, you know, good competitive incentives. Well, I feel like everyone's like, no, no, no, no, no. I'm not going to give you my data. Why don't you give me your data? And, uh, it's, and it's like, no, no, no. Trust me. Trust me. Trust me. Trust me. Trust me. Just give me your data and we'll send you more traffic and it'll all be good. And that's basically what happened, you know, what Google did and what Facebook did and whatever.

25:29And so like, I think everyone's getting pretty smart now where like notion isn't sharing their data with with open ai and to your point it takes 60 steps to get google docs into open ai now and so like everybody's like hmm no how about you give me what you know and i'll be the one box the tool rule them all for everyone right all right uh let's move on to the next theme i'm only one one for three down how many themes have we covered here one of mine was kind of vibe coding which we covered with the design thing so i sort of just have one more so it's okay you have one more left Yeah, you go for it.

26:00This one, I didn't need to give a clever name. The term itself is ominous, but surveillance state. I think, you know, we've got granola, which obviously requires a level of everybody's getting recorded. So we have this question of, are we all getting listened to all the time? And then the other big one that's relevant right now is self-driving cars. You know, I mean, I'm very hostile to Waymo's getting attacked. But you did see, I think it was Taylor Lorenz did an interesting YouTube video where she was talking about it. I like self-driving cars, but they are recording protesters everywhere. You know, it is just like another piece of the sort of we're constantly being monitored.

26:41I have never been, I'm not a big privacy person. So it's not like I'm like up in arms about it, but I've always not really cared about privacy. I am, as you guys know, an overshare, but it's surprising to me that privacy is not a larger part of the criticism of AI because so much of this stuff requires like ubiquitous surveillance. There's also that New York Times court order that requires OpenAI to store and save every chat you've ever interacted with on the platform. Which is pretty wild, I think. I think it's a massive overreach from the judge, but whatever. Requiring them to store every single thing everyone types into ChatGPT for the purposes of the lawsuit.

27:23I don't know. I think that's bullshit. That said, I was just using ChatGPT last night to create Pokemon. And I'm like, how is this legal? Like, not even new Pokemon. Just like, hey, make an image of a Pokemon. And it's like, sure, happy to. I know all the Pokemon. You know, it's just like, isn't that the most blatant copyright? I know you can draw, but like... I think golden age for IP theft, baby. Somebody should have to sort of keep all this stuff they're creating because it's like, oh man, there are going to be lawsuits for like ages. Well, that's fair. Yeah, I think I'm more worried about like the law enforcement angle to it.