Josh Elman on Apple, AI Agents & Why Startups Can Still Win

Are we about to enter a world where everything is recorded and we have a transcript of everything?

You know, I think we are definitely in a world of building new norms, but these are all norms and trust that have to get built out. And a lot of it's not software, it's humans.

Do you think it'll be sort of adversarial where it's like my agent, yours, and like mine spewing lies and yours is like trying to ferret them out?

I mean, I tend to think that truth and authenticity win in the end. Doesn't mean we won't have a lot of gamesmanship in the middle as we're figuring these things out.

What's your view on how much this is, you know, OpenAI's to lose?

I look at this not as OpenAI and Anthropic. It's not a Netflix game to lose or win. It's they have to do an incredible job continuing to build their businesses. But there's so much more that we're all going to do.

Josh Elman, welcome to the Newcomer Podcast.

Eric, thanks so much for having me.

Man, we go way back. You were a venture capitalist at Greylock, invested in Discord and Musical.ly, which became TikTok. And then I had a friend who worked at one of your companies, Operator, which was an early chatbot company. You were also an investor in sort of the zeitgeisty consumer companies of the time in Silicon Valley world, Houseparty and Meerkat, if people remember those. Then you sort of, you know, you go to Robinhood, you go to Apple, which Apple, you're not really like staying in touch as well, I think, with reporters.

Yeah.

It's sort of like there's a level of safety. And then now you're back at Andreessen Horowitz. So it feels like, oh, I keep saying, welcome back to my world. Like you've come back online and very excited about consumer, which are always fun things.

Yeah.

to talk about. So great to have you on the podcast.

No, thanks so much. It's great for being here. And look, my whole career has been about trying to make products that people use and bring into their lives. And, you know, whether that was in my early career, you know, building products or becoming a VC. But about 8 years ago, I was like, man, the best way I want to go make consumer products is just to go work on them again. And so I went and did that. And I mean, Apple was an incredible place to work. But now I see this edge that consumer products can really come back and we can build these little startups that are growing into these next wave of great companies. And I just couldn't resist the opportunity to go and be part of this incredible new wave.

Yeah. And I might get into the sort of why each decision, but I wanted to just sort of get into the meat of one of the core sort of categories you saw that is now becoming reality, which is just sort of the chatbot. Like, what appealed to you about sort of chat-based companies? I don't know, how many years ago was it now?

It was like 2014, 2015.

Right, exactly. More than a decade, which is, yeah. What was appealing then? Do you think they could have succeeded or it really required LLM or what would— how would you look back at that period now with the technology that we have?

You know, when I think about chatbots, I go almost all the way back to like ELIZA, which is like this very old text-based system that you could just talk to and it could figure out with very simple, you know, regular expressions to kind of spit back some text. But it was always this sense of we're humans, we speak in language, how can we naturally use our language to communicate with a product and have it respond to us in a way we understand and then can go back and forth and be really interactive. You know, we saw this first wave of AI, or the beginnings of sort of what could be these chatbots that were happening in 2014, 2015, with the idea that maybe you could use a little bit of technology and also people on the other side, but simulate a great experience where everything could be done over chat. If you think back to like 2014, 2015, we had moved from a world Where SMS was kind of a thing for a long time, but it was like the only way we started communicating. And we were so actively communicating. Everybody's like, Messenger was huge, iMessage was huge, WhatsApp was getting very, very big.

And there was this intuition that obviously we have to meet the customer where they are. If they're on SMS, try to sort of give them a sort of shopping or other experience there.

That's right. And so Messenger had come out with M, that was the idea at Facebook that you could just type something. And they had, you know, you know, a lot of people behind the scenes that could answer things or AI that was doing it. With Operator, when I met Robin Chan and Garrett Camp that founded that, where Peter worked, it was this great idea that we should make every business textable. And the idea was, could we do 2 things? One is put a business, a phone in the hands of businesses so they would actually have people that could answer your questions. And then could we then build an AI and routing layer on top of it that could actually solve some questions, do some things? We actually ended up even building sort of a human force that could get the questions aggregated, figure out what could be done, you know, through software, what needed to be done, you know, with the— it was a human doing the research for you to be able to do it. And it was all just too early. The— what the AI could actually do at the time was very, very hard to automate. It really couldn't do much at all. And so we ended up having to more have humans on the other side to be able to do it. And that just completely tips over your economics, what the actual product is, what the network is. And so it ended up not being something that consumers would, you know, stick to and pay for. We found this thing with Operator that really fascinated me. People loved going back and forth with Operator to find a really hard-to-find item. And somebody in the Operator team, or with the help of their tools, would actually be able to go find that item. And then they would send the link back to the person. They would say, great, will you pay an extra, you know, percentage points or whatever to actually help make this purchase? And they'd be like, no, now that I found it, I wanna go—

You're giving me the answer I wanna hear.

I wanna go find the cheapest price online. And so, you know, it's this early days of figuring out the products, the modality for users, and what we could actually convince people that it could actually be a sustainable business.

The AI skeptics worry that there's still, you know, people behind the scenes answering questions. Do you have a sense of, you know, I mean, obviously there are a number of these apps that are getting started, but sort of personal agent chat-based apps, do you think there are still people filling in the gaps or they're mostly true fully LLM answers at this point?

I mean, I think the products that we're seeing today are really powered by LLMs, powered even by computer use agents that are actually navigating the web and pulling up webpages in doing things for you. And I think where the technology is now there that it can actually do most of this for you. I think the, the real question is, at what point do you still want an escape hatch to a human?

Hmm.

Something goes wrong. Like, when things go right, it's amazing when this happens and it just works. But when things go wrong, are the computers truly able to troubleshoot, or do we still need people to help us troubleshoot? And I think that's gonna be one of the very interesting challenges, 'cause at scale, Things that can go wrong a little bit can, you know, percentage points can really add up. And so I do think that that's going to be a pattern we'll see and learn over time.

Yeah, in some ways the LLMs are more powerful than the humans you can designate now. It's like, oh, they can use your computer in a way that would have been hard to hand over to some customer service agent. On the other hand, yeah, they— that with that power, they can buy plane tickets that you can't refund and burn money on accident. Um, you know, I guess clearly I'm referencing Instinct, which is sort of a buzzy Silicon Valley company right now. You all are invested in Talon, which is sort of an enterprise personal agent. What— yeah, give us, I guess, the, the high-level read on like the personal agent space, uh, as you see it right now.

I mean, this is, this is one of those things when, when I— what part of what actually got me really excited to come back to venture was when I first saw OpenClaw. And what, what I saw with OpenClaw was, one person in almost in his living room with some of the story wrote this incredible piece of software that wrapped the models but did a lot of things for users and on top of what the models can do. And it saved context and it identified skills and it found new things and it created this paradigm where you could just text it and you were like, whoa, I'm actually interacting not with a model. ChatGPT and other things had taught us what interacting with a model was like. This taught us what interacting with an actual agent that was solving problems and getting things done for you was like. And so now we're seeing this incredible consumerization bringing this both to our work, like Town, bringing this to people like Muse from Facebook and Instinct, that it makes it really easy just to almost send a quick message and have it go take care of something. We're just nascently learning what these things can mean for their lives. And I think we're going to see every chatbot begin to not just give you answers and have a good conversation but be able to complete tasks for you too. I think this would be really important. The real question that I have is, as we're all using these things and getting things done, this tips a lot of things on the marketplace side. You know, we've seen these examples already, booking reservations.

Oh, it's all destroying Resy.

Well, but here's this challenge, which is, let's say there's one reservation slot that opens up and 5 of us want it for that time. Who makes the decision? Is it first in? Whose agent is first? If we're all using Facebook's Muse or Meta's Muse, which person does it decide is going to actually get that slot? And if there's 5 different agents from 5 different companies and then 10 of us are running it on our own home computers, like, is it all just another first in? How is Resy going to manage that? How do limited space get managed? So we're going to have massive implications.

There's a, there's a word that I think resolves pricing. You know, it turns out it's like, oh, you have a lot of people who want this Same thing, and right now we're just giving it based on sort of who comes first, and they can treat it irresponsibly.

Yeah, and so, you know, I mean, obviously a lot of people just propose it could be auctions and reverse auctions. I mean, those are always fun, but then that's just the highest bidder. Sometimes you want it to be people who really have reason to be there, you know. If it's a very special— if you want this restaurant because it was where you got engaged and you're celebrating your anniversary, a way to actually be able to express something like that. And have that be part of the equation too, and have that work on the other side. So I think there's so many different mechanics as we move to this world of personal agents where we have, you know, we could always do this ourselves, but I think when you automate it, it can happen at hyper scale.

I'm super interested in that intuition, particularly, you know, I don't know, the journalist in me that likes sort of an honest layer. Like right now with a restaurant, I could claim That it's my anniversary or something, and it might not. But, but with agents with potential layers in between, there might be some sort of validator where it's like, yeah, I sort of work for them, but I'm also sort of a neutral layer and I've seen the pictures and this is their anniversary. Like, do you think it'll be sort of adversarial where it's like my agent, yours, and like mine's spewing lies and yours is like trying to ferret them out? Or there is this middle layer that's trying to adjudicate whether it was my anniversary or not?

I mean, I tend to think that truth and authenticity win in the end. And in most platforms, that actually works. Doesn't mean we won't have a lot of gamesmanship in the middle as we're figuring these things out, but I tend to think that that authenticity really works. I mean, Airbnb is such a great example. It's Andreessen Horowitz's company. Like, that truly the authenticity makes it able for you to trust somebody in your home, trust that you're using somebody else's home in such an incredible way. And that authenticity will come to all of these things, and the agents can very much vouch for that and represent that. And rogue agents Ultimately, if they are like very good at faking things and using AI to simulate anniversary photos or whatever, those will eventually get figured out. But we might have some mess along the way. Now, I think, I think about this a lot with ticketing too, and Spotify is doing some really cool things where they are helping, you know, tickets, you know, deliver tickets to super fans. You know, if you listen to somebody the most on Spotify, you deserve at some point to be one of the people who gets good access to their their tickets for a show versus who can, you know, hit the bots.

Man, with Taylor Swift, my wife and I will be like 1%, 2% Spotify Taylor Swift listeners. The gamesmanship that would occur there— you already have Swift fans who just, you know, play her music just to support her and drive up the bill or whatever for Spotify. If you actually got rewarded off that, it would be a disaster. But it's a fun situation, certainly for less prominent artists.

Yeah. And I think— and so I think this idea of authenticity, connection, more than just money is how these things will actually play out. But there's so much we have to figure out.

And are there companies that represent sort of layers to this that you're optimistic about?

Or, you know, I think it's so early that we're, we're just figuring out, you know, I think the core platforms already like Resy and things like that will actually have a lot of opportunity to continue to grow what they do as they become not just sort of a place for booking, but the trusted representative of the restaurants.

Mm-hmm.

And, you know, for example, they're very much on the restaurant side. And so restaurants can rely on them to help them stay full, stay active, make sure to make sure that, you know, the right people are coming. That's a real power that they can have. And I think we'll see more marketplaces develop that bring in those capabilities over time and then connect very well into agents. You know, this agent-to-agent network, whether it's connecting people, connecting people to businesses, connecting agents to other agents, Maybe even, you know, having money that flows from agent to agent based on the agent's capabilities. There's so much opportunity here, and we're just beginning to imagine and sort of tangibly see what that might be.

OpenAI, and probably Anthropic too, don't want to give up this space, I imagine. You know, Meta is a model company that moved very quickly to it. What's your view on how much this is, you know, OpenAI's to lose basically versus something that's an opportunity for startups?

I mean, I think we're at a moment where we're getting to redefine so much of how the internet works that this is a massive industry change, economy change, not just a single point for anyone to win. When we look at the original version of the internet, there are so many winners— Google, Amazon, Facebook, Meta, Twitter. I mean, so many things got very, very large. When we look at the mobile economy, Uber and Instacart and DoorDash. And again, so many companies grew. So I look at this not as OpenAI and Anthropic's game to lose or win. It's they have to do an incredible job continuing to build their businesses. They are the model foundations that so many companies are built on. Many are also now using open source, but that is a platform that they move forward. ChatGPT is obviously an incredible product used by, I think it's like a billion people per week. But there's so much more that we're all gonna do. And so this, there's room for them to be great and winners, and there's room for startups to find their wedge and be winners. And there's room for companies like Facebook or Meta and Google to continue to move into that world and be very, very relevant too.

Are you bullish on personal shopping agents specifically? I mean, coming out of Operator.

Yeah.

I think people debate, it's like maybe that's one of the things humans wanna do at the end of the day. Everything else has been taken from me, but the thing I wanna do is, you know, fantasize about my next vacation or whatever. On the other hand, you know, it's, it's a lot of money. It's something that it's nice that, you know, there were travel agents. You know, I don't know, it's not one or the other, but do you have a general sense of how bullish are you on these like personal shopping agents?

You know, when I think about this category of shopping and travel, it's a combination of the, the, the joy is not just in the activity or the purchase, but in the exploration. in the discovery, in the seeing lots of different options, sometimes physically getting to, you know, for clothing, getting to try it on, for travel, getting to explore different places where you might want to go, making all of those trade-offs and trying to figure out which activities you'd want to do. Like, the joy is in the entire journey of planning or exploring or shopping. So I very much believe that we're going to have agentic experiences that will help us do these things. faster and better, communicate natural language, get to much more specific things. You know, I would like to stay at a place that has a view of the water and has a kitchen and has all these things. And right now, a lot of the search engines that come back, you have to do so much sifting to get options that even meet your criteria, let alone then picking which one really feels right to you. And if you can express these things better in natural language and then have something that actually you know, uses AI to look at the results and give you back the things that match it. Now you have a great option set to pull from. And then you might say, widen this a little bit, show me some things I wasn't thinking of. And the AI can do a great job giving you more options you hadn't considered. So I think the best shopping, travel type experiences are going to be full of this planning and discovery and very visual and full of, you know, videos or demos and, you know, Put myself, you know, in clothing, put myself into this and use AI to imagine myself wearing it and what the fit will be like.

Right.

And there's so much in that journey.

And you can imagine it like a good, you know, so much of what investors are thinking about right now is service industries that used to be inaccessible, now they can access. And so you think of like a great real estate agent or a travel agent, they're like texting you. It's like, here's a cool house. You know, it's not necessarily It can be spontaneous. It can be where you are, and it's sort of a joyful moment that doesn't necessarily say you need to transact right now. And so these agents are going to have all this possibility to make it a more pleasant and sort of ongoing experience.

That's right. And it really becomes this constant journey of discovery and learning new things. And, you know, we're still humans. We still make our own decisions for our lives. It's just great to have tools that will help us discover new things, get them done faster, You know, if I know I have a task I just need taken care of, it's great when the agent can do it for me. But if I don't know what I want to get done, the way that AI and the chatbot tools help me explore the space and brainstorm ideas and think about the parameters I should care about is such a change that you used to only be able to get with a human. Now you can get much further so that when you go to have the human interactions, it's even more powerful.

So there's another huge trend. I mean, probably more that's hit more people, which is just TikTok. Obviously, you were an investor in Musical.ly, which gets acquired by TikTok and helps become the product we know and love today. I mean, it's sort of a lean-back experience in some ways, so it doesn't, I guess, call for as many, you know, VC tweets. But, you know, the time spent on it is enormous. I think another dynamic is just It is. I mean, you know, between Meta and TikTok, they've sort of like gobbled up a lot of the opportunity.

And YouTube with Shorts.

Right. And YouTube. Yeah, definitely. But there aren't like new startups to bet on and therefore less sort of energy from the investor class to talk about. I don't know. Are they entrenched? Like, where is the opportunity for startups and like, where is this sort of like short video trend headed?

I mean, if we had talked in 2015 when we invested in Musical.ly, I think we did. Um, you know, everybody was like, well, YouTube, YouTube's already so big on the internet.

Vine had failed right around then.

Vine had failed. You know, Dubsmash had come and everybody was making fun videos, but it wasn't sticky. Musical.ly just looked like teenagers making little music videos in their bedrooms dancing to a song. And, you know, I have to give real credit to the team and especially, um, Alex, who's, uh, one of the co-founders. Who painted a vision even then of what TikTok could become, which is this incredible feed of things that were personalized and interesting to you. And that as you swipe through one at a time, you were able to kind of be immersed in that moment and get to know it. And that there were really going to be creators that have become, you know, these creators and these identities that have become these real rich relationships that you form. We're seeing some of that on YouTube as well. But with Musical.ly, he was like, this short form can really help them share their lives in a way that is even more impactful than what we were seeing everywhere else. That this ability just to swipe and watch and the feed learns about you and really delivers personal things to you. He had that vision all the way back then and in selling to ByteDance was able to really go into the company and realize it. Look, now we're at a point where between Reels and Shorts and TikTok and Douyin, which is in China and the same from ByteDance as well, this is now a foundational paradigm of what people do to spend time. Often like, do I want to sit and lean back and watch a big screen TV or just on my phone watch a few things and have a good moment? It's been so amazing to do that. And yeah, I don't think there's room necessarily for, I'm going to be another short video network, because I think we're very well served by the products that we have. However, there's things that spin off the short video network that are actually, you know, incredible new businesses themselves. I'm not sure how much you've been following microdramas.

Microdrama— this isn't the sort of— I saw the AI vegetable thing. That's a different trend.

I mean, the AI vegetable thing is a small part of it, but There are companies like Reel Shorts, Drama Fever, DramaBox that are creating short stories that are maybe 20, 30, 60 minutes in length, chunked into multiple episodes. And you might on TikTok see, or on Instagram Reels, see the first episode or 2, or the first trailer for this wonderful story. Some of them are like soap operas. Some of them are little fantasy things. Some of them are mystery and horror, but they kind of get you in. And instead of just kind of, again, flipping one show to the next, that sort of becomes this feed of constant attention switching. This is watching a little bit longer. This might not quite be an 8-hour Netflix series that's costing, you know, hundreds of millions to produce, but this is a thing that is cheaper to produce, but it still requires great storytelling.

And are they mostly AI-generated?

You know, right now these got big even before the AI wave. So a lot of them were actually just filming people and doing things, you know, a little bit on the cheap, but we're creating real shows that, that made sense, or, or using animation. We're seeing a lot more of them now use AI in order to do this, um, and, and to create these videos. There's like one called Shorticle that I think is, is primarily AI-generated that is— they're great stories. Like, you go to watch the stories, and at least me, I always get hooked, and I'm enjoying, um, you know, just what I'm watching. And again, it's— they're short, they're, they're easy to consume, but you kind of go on the whole narrative arc, um, and I think with AI, we're democratizing creation of these so much that we're going to see massive new platforms for storytellers, not just for a short video that's a little bit gimmicky, but a real story. And so we'll still have our TikToks and our YouTubes to kind of watch the kind of clips and the things that we watch there. But when we want to go through something that's a little bit more story mode, maybe instead of Netflix and Hulu, we'll still go watch Game of Thrones on HBO and we'll still go watch You know, Squid Game on Netflix. But when you just want to watch something that's kind of in your zone and in, in your area and just be able to watch it on your phone and just flip through it, these microdrama platforms, I think, are going to become so much bigger.

They need to be able to sort of build a standalone app to be sort of venture-worthy, or do you think a studio model, like if they were just on TikTok, could they?

So, so most of these already are standalone apps because what you'll do is on TikTok, you'll see the beginning of the show. or a trailer for it.

Like a bigger hook.

Yeah, you'll be hooked and be like, go finish this, go download this. And they actually all monetize mostly through subscriptions. And so, I mean, there's billions of dollars already flowing into this category if you go dig into like the data or other things. Many of them are actually Chinese companies that have kind of figured out the model. It looks a little bit like mobile gaming and they crank through these stories. But I think with AI, we're gonna be able to elevate the storytelling, elevate the creativity, And, and really, and create platforms that allow anybody with a story they want to tell to be able to put it into one of these platforms. And just like TikTok let anybody be a creator of a short video, this will let anybody who's a storyteller put something out there that can get a larger and larger audience.

Are you optimistic about AI-generated feature films? You know, the shorter content, you can test it more. You can see, oh, are people responding to it? Lean into it, build the next segment. Like, do you think we'll have movies that are AI-generated?

I mean, the long-term answer is sure. I tend to think we always try to jump to the endgame. And what's beautiful about startups and the startup ecosystem and the venture ecosystem is that we don't always need to boil the ocean on day one.

Hmm.

We build something that meets today and that can deliver a great experience today, but it may not be the biggest thing. I remember the early days of YouTube, one of the best quotes that was told to me was we're not— you know, I was like, how are you ever going to compete with like network TV? It was like, we're not aiming to beat the best hour of network TV. That, you know, if you still want to watch an incredible— you know, I think Grey's Anatomy was a really popular show at the time. Grey's Anatomy, you should still watch Grey's Anatomy. But we're trying to beat the worst hour of TV. Because if we're more interesting to find things on YouTube than your worst hour of TV, you'll start to spend that hour on YouTube instead of on TV. And so I think that's very much true, is again, you don't need to beat Netflix and Hulu today, but you need to beat the, I don't really know what to watch, I'm just kind of surfing around, I wish I had a little bit better story. Be that better story. Now you start to fill from the bottom, and then as you build new franchises, you build stories, as you build characters, it can move all the way up.

I'm curious for your reflections on Meerkat, Houseparty, and then sort of Clubhouse is a different phenomenon. To me, I'd bucket them as sort of like things that were— they tried to sort of build energy in a sort of an ecosystem and will them to being. People have, you know, some people wonder if Instinct has Clubhouse vibes and that, you know, VCs are very excited about it. They're obviously, you know, what, like Foursquare and Twitter launched at South by. This sort of had this sort of insider community is going to will into being it. What's your reflection? Like, will there be another, like, the venture community, the Silicon Valley community of insiders sort of gets excited and that catapults it, or is that past? Or what's your reflection on that?

I mean, I think there's 2 different things, 'cause I think with things like Meerkat that was so tied to Twitter, you know, and Clubhouse, that definitely started from a vibe of people who were very much in the tech and Twitter ecosystems and able to grow. And they were all about bringing people together live. And I think what both of these companies showed is how magical that moment is, how magical that experience is when we're having live, but also how hard it is to scale when you're only live. Most of the products we've talked about, TikTok and Instagram and everything else, are very much async. And so how do you combine a product that is async and live to be a large platform? What has turned out is that the platforms that are large and async can add live components and have those work rather than a fully live network becoming the next big thing. And I think that's always been the challenge with, with some of those companies. To separate your other part of your question, I sense you're talking about kind of the Silicon Valley energy of technology and technologists discovering something new and bringing that to the world. And I actually think the magic's back. Like, I think with AI, the beginning people who really understood AI were the tinkerers, were the ones who would not just get ChatGPT and be like, I can type in what I type into Google. Ooh, that's a little bit better result. But I can type in really complex incantations and start to get really thoughtful, different responses. You know, remember all those prompts that were shared in the early days of people with ChatGPT? Imagine that you're a detective and you're trying to solve this problem. Well, here's all the evidence I have. Can you please help me figure out what's going on or what are the 3 steps that I should do next? You know, with Google, if you ever typed that into Google, it would like throw up in there, you know, before they really added their AI mode. And with ChatGPT, it would give you a great answer and the more and more complex prompts. And as technologists were figuring that out and then they were figuring out what they could build on top of these APIs. Mm-hmm. Things like Open Claw were very much a tech phenomenon. We're discovering things again that reminds me of the very earliest days of the tech ecosystem where technologists, you know, builders, tinkerers get this first. And now we're at this incredible wave. Part of why I came back to VC is turning from that moment to how do we consumerize this? How do we truly make it available to most people? And I look at Instinct as one of the coolest phenomenons of a startup that is like taking a bunch of this complex technology. It's one of the first that really was like, how do I make it super easy?

Mm-hmm.

I'll give Poke some credit. They discovered, you know, putting this just through texting, through iMessage made their, you know, it made it so much easier to interact even than going to a complex website and other things. And so we're at this great wave where sure, it gets exciting in the tech community, but that's 'cause we've already been tinkering for so long. The rest of the world hasn't been on that journey in the same way. You know, many of them are very heavy, users of AI now, ChatGPT, other things. There's also a lot of resistance to AI because we're telling such bad narratives about how this could replace all jobs or kill everybody as opposed to enable the next incredible wave of innovation, which I very much believe in.

Well, yeah, one of the reasons I'm excited you're back is to me in Silicon Valley, you represented like, you were very optimistic. And, you know, I feel like my journey as a reporter, it's like, I was more skeptical in the beginning. The information where I really came up was more skeptical than most because, you know, we're in the age of like gadget reporters and all that. And then at some point, you know, while I'm at Bloomberg, Silicon Valley became— or the press and the mood generally just became sort of oppressively negative.

Yeah.

And then by the time I'm launching Newcomer in 2020, I'm like, oh, we should have more optimism. And now we're almost in this world where I'd say the public, honestly, is the most negative of all.

Yeah.

Uh, yeah, I don't know what— yeah, what do you— I guess to ask directly about the public piece of it, like, why, why are people— do you believe that people are so negative about it? I mean, a tech response has been sort of, you know, their revealed preferences and then, you know, stated preferences. People use TikTok and Instagram all the time, which are algorithmically generated. They like these stories you're talking about, which AI is creating, obviously. Billions of people are using chatbots, but then Americans say they're very negative on AI. Like, what do you make about that in terms of building the next generation of consumer apps?

Yeah, I mean, I think if there's any one thing that I learned from Apple, it's we have such a responsibility in our industry to be grounded in building technology that benefits humanity. And we have to think about it, and when we think about it from that perspective, we make choices and we focus and we tell our stories about how we try to deliver things that benefit humanity, whether that's helping our health. I mean, the Apple Watch is one of the most incredible health devices that's ever been made, and the number of people that it's prevented from various, you know, falls or, you know, heart issues or other things, it's, it's extraordinary. But even just the number of people who rely on their iPhones to get them through their day. And so how do we think about technology from the standpoint of benefiting humanity? humanity. For some reason, with AI, a lot of the early tech bravado has seemed to be about how this— well, it's going to be so good, it's going to eliminate jobs. Well, when I go to anybody outside of the tech industry and I'm like, eliminate jobs, that is like red alert warning and scary, right?

Right.

And it should be, because that's not the way that we should think about what we're trying to do. It's empower people to do their jobs better. And If there are certain tasks that are less, you know, that are easier to automate, then that helps you move to higher-order thinking. And so I think moving from this world where we'll deliver it with proof, we'll deliver it with products that truly benefit people's lives, that's one part. The other part is a lot of the early things people figured out how to do with AI were cheat on homework and create sloppy things that didn't sound like they were human-written and just spew them out. And again, if you're going to put something out, even if you use AI to help you write it, It should stand for you and you should stand behind every word. I certainly use AI to help me write things now and build decks and everything else, but I never put it out if I don't control every single word. And if I decide I like the words that I iterated back and forth with AI, fantastic. And if I don't, I change words here and there and change sentence phrasing and move paragraphs around. And we just got to a point where people got lazy, and AI should not make you lazy, like to cheat on your homework and to spew out something that you didn't take seriously enough, but it should empower you to do better. And I feel like more of that, how we think about it, how we build for it, how we pitch it to people, and then to your point, they'll use it however they want. But there's a start from there.

I mean, part of the opportunity I think that we see with AI right now is these models are capable of so much that people haven't sort of really defined, built the data around, marketed. And so then People are just shouting AI, and it feels like, you know, sort of like, well, what do you want me to do? And it's like, well, it actually takes time to build the companies and figure out each individual use case, which is some of the exciting opportunity ahead of us, which is it's almost people yearn for like the storytelling of like, what am I supposed to do with this? And it's too much for like OpenAI and Anthropic to deliver it all because they just sort of want to be— it's a Swiss Army knife. I don't know, figure out what you do with it.

In some ways, I joke sometimes that Giving you a blank open edit field is an incredibly powerful thing, but in fact, sort of the worst thing in the world to use for onboarding and getting to know a product. And we have not built the tools that help the people who really get this bring other people along on the journey. The product experiences have gotten so simple that you just type in. Now, Google got there, and I'm going to give Google a ton of credit. Over time, people learned to search and people learned how to search better, and people showed each other. But not every product needs to be Google, or certainly not at the beginning when we're doing this. And even if you look at early Google, they used to have a lot more things on their website to teach people. And over time, it got sparer and sparer as people got familiar with it. And I feel like we keep starting there with a lot of these AI products and, you know, ask anything because it can do anything. I don't know what to ask, as opposed to really taking people on a journey.

I'd love some reflections from your time at Apple. You know, on the one hand, it seems like Siri has gotten much better. Uh, I'm not like a deep user, but that is, I think, the word. Um, I am an iPhone user. I, I'm, yeah, getting into voice, but not, not fully switched over. And then on the other hand, I mean, I feel like, you know, Apple intelligence as sort of a marketing frame sort of veered from some of what you said, which is I think Apple is traditionally good at What are we doing for you? And like, who cares what technology we're using for it? It's like, it's helpful for you, you should use it. I don't know, how— what did you, yeah, learn from sort of your Apple experience?

I mean, I had such an incredible time getting to work at Apple. Obviously, I was an Apple fanboy since I was a kid. I edited my high school newspaper on my Macintosh SE. I took a Mac LC II to college with me. So getting to work at Apple was kind of like living a a lifelong dream. And getting to go work on Apple and Siri and Apple Intelligence, figure out how do we tell the story of Apple's place in this AI wave, was, was just such a privilege over the last couple years, you know. And getting to launch Siri AI and kind of this latest version of Apple Intelligence, which we announced at WWDC— that happened to be my last week at Apple, was kind of going out with that big announcement— was, it was really a dream come true. And now I think it's available for everybody. I'm one of the people recommending Update the operating system now. It's fantastic.

To get what?

To get iOS 27 or macOS 27 or iPadOS 27 or the other operating systems. Because I think really you have, you know, incredible work that went into the operating system last year for stability and resilience, but also a ton of work to really make Siri deliver on a lot of the promise that it always did. And I think You know, Siri was never like a chatbot. Siri was always supposed to be an assistant that helps you get things done on your device. Set a calendar appointment, set a reminder, send a text to somebody. But like most of the early things that we were talking about earlier, like Messenger M and other early AI attempts, you had to basically know the commands to tell it. And if you knew the commands, it worked really well. But if you veered too far in your language or were trying to put things together that didn't quite work, or you didn't quite get the statement right, it couldn't deliver on that. And if you asked it questions outside of its domain, it couldn't deliver on that. When ChatGPT came out, it obviously showed a whole new paradigm of being able to have the LLM reason over everything. But to reason over everything and then still figure out how to understand all the data on your device, your personal context, and take actions in the exact steps that it needs to do for you was a hard thing even then for the LLMs to put together.

Hmm.

And so I think the Apple journey was one of like, Everybody was like, well, just can't you be ChatGPT and Siri all at once? And it was like, no, it actually took a lot more time and care and work to really build it and not just build it to a level that could work sometimes because that was actually not that hard to do. And that was part of when I think some of the initial Siri stuff got announced, things worked and there was a system that could work, but it wasn't at the quality level that Apple needed to be to bring this really mainstream. And so, you know, I think it was really impressive to watch a company not just say, okay, well, that didn't work, shut it down. It was put your heads down, work harder, figure this out, actually deliver on that promise that you made. And now you truly have a Siri that understands information on your device. You can take a number of actions across the apps on your device. You know, I use it all the time. When somebody texts me their email address, I just say, hold the Siri button, and I say, add that email address to their contact. And it just works. You know, when I—

How do you teach the user to have confidence again, or that it's possible?

I mean, I think this is going to be a really interesting journey and one that I am probably going to— or I'm sad to miss out on, but incredibly strong team behind it. How do they go tell those stories again and bring people back to Siri? You know, it's called Siri AI to kind of give you the sense that, hey, this is a little bit of a different version of Siri. It's Siri with Apple intelligence.

Mm-hmm.

It actually brings you all this personal context and actions, and you can try it again, and it's new. It teaches you a bunch of examples. But I'm sure there's going to be a huge learning curve and a huge trust curve for people to come back to. But if they do, I think they're going to have this great experience. And then I still think this is a platform. This is an ecosystem. Siri is not going to be everything to everybody. You'll use ChatGPT for other things, or you might use Muse or Instinct or Town for things in your, you know, town and your work life and other personal things because you are still, you know, get to use lots of tools on your device, but you'll have this base level on your device that's gonna be amazing.

Was there some impulse to sort of to AI, like as a verb? I mean, I never understood the text summary feature or like that was one that was very visible to me where it's like you'd get a text and Apple AI would say like, oh, it's like texts are short or I don't know.

You know, I think one of the really cool things with the operating system is obviously Siri and the Assistant is a very important part of it. But there's so many other things you do every day that AI can help. You know, some of that is sometimes you get 30 notifications and they stack up. And actually having a summary that gives you a little bit of a lens into what's in there before you tap and expand can be helpful. You know, text, it has to be like 200 characters or more or something before they actually get summarized. And if someone sends you a diatribe, it's sometimes nice to get the quick glance. And you can always again tap to expand it. And so I think what what Apple did really well with Apple Intelligence was not just was focus on the assistant, which took longer than it needed to, but also focus on a supporting cast of other features around the operating system that are just better with AI. Whether that's writing tools right in the moment to help you rewrite things, whether that's in the Photos app being able to you know edit your photos or clean them up, or you know expand them now, which just got launched, or this really cool feature around. Spatial reframing where you can actually tilt the camera a little bit because it has a lot of spatial knowledge. It can sort of fill in any gaps even after you took the photo. I've used it to like— I left a sign sticking out of my daughter's head and was able to, you know, move it around so it didn't look as funny because it was otherwise a great picture of her. And so there's so, there's so many other things that, that you want, but it's not AI for the sake of AI. It's AI for the sake of features you always wanted to build that makes life easier. And that thankfully now that there are these LLM tools, we can actually power them.

So you've left the device company in the world. Clearly OpenAI is working on devices. I think there are startups who are thinking about it. Are you optimistic? I guess if I say optimistic, you're going to say yes. No, but I don't know what device category. Do you think we will have new sort of AI devices, or do you think existing devices will become imbued with AI powers? Do we yearn for a pen or AirPods, whatever we're calling them, that are AI native? Or people have talked about sort of a portable disk.

I don't know.

Do you have a view on what types of devices lend themselves to AI?

I think we are at the dawn of what's hopefully an incredible explosion of new AI capabilities that are going to make a lot of these devices a lot more useful. What is really powerful with AI? when it senses data and it can get more information in, and then when it can use that to help compute things and deliver them back to you in a useful way and take the action for you. So a lot of what we're seeing are things that look like more sensors, whether there's always-on recording, even the new Apple Watch.

I know, I was shocked by that, that they're gonna let you sort of just record ambiently.

You know, I think it's very much in a privacy-safe, very moderated way, but I think that's an incredible feature. But there'll be other devices that will help us do that. There'll be certainly other things that can be on the lookout for us, you know, whether that's eventually coming into smart glasses or things in our ears or things we'll wear on our shirts. You know, I'm not sure which things we'll catch. I'm less the, the perfect creator, but I think the people who are designing these things and building them that have visions, like, we can very much, you know, and Andreessen Horowitz help them amplify those. And this is a moment for new hardware just as much as new software. And, you know, again, the inputs, you know, it's still work to like pull your phone out of your pocket to say something and have it do something. You know, when you're wearing your AirPods and you have Siri there, you can just quickly ask questions or get things out that you don't have to necessarily pull something out of your pocket. That's the dream of the glasses, and you can capture things.

Right. Given we wear glasses, I'd be happy to have smart glasses. It's like, why not?

I mean, I focus on buying the most like Unbreakable and lightest glasses possible. These— I break glasses at the hinge a lot, so these actually have a hinge that I can reset. So like, on one hand, like, but once it's as light and easy as my glasses, I would love it if they were AI-powered.

Oh, interesting. My daughter is 11 months and she rips off my glasses all the time. So that's, that's the new problem. Are we about to enter a world where everything is recorded? We have a transcript of everything. In some way, I'm not so opposed, but There are a lot of problems with that, certainly in a business culture. Like, are you optimistic that we're just gonna like ambiently record everything so that all these companies have the data they want to give us, you know, all this value?

You know, I think we are definitely in a world of building new norms. And I think, you know, on video calls, it's very easy for people to have both a recorder that's visible to other people and even one that is not. In the world, we still haven't figured out the norms yet of when this is there. If something were to have a camera that could take pictures, being able to trust what the person is actually doing, where they're using it, and what our new norms are. Like, this is not something that's going to like be solved overnight, but I do think a world where we have these additive things to help us remember things and keep track of things and hold ourselves accountable to what we said we were going to do and recall cool things we heard about, like, is a really, really useful pattern. And the right to forget and the right to To not remember and not record and have safe spaces is really valuable too. But there's so much trust that goes all of it. I mean, the moment you tell someone a secret, the secret is no longer yours, but you have trust in that person. And so we're gonna have to figure out with whatever tools and with whatever people that same level of trust and safety continues to be there.

And we're gonna have much more meetings with like no devices because there's so much ambient Recording going on.

No device meetings in the sound room. Or if any one of us is recording this, share it with everybody and trust that it's still Chatham House rules or whatever set of rules, you know. And, and, you know, I have chat groups with people that, sure, everything's recorded in the chat group, but everybody trusts each other. And there is always some risk that one person could burn the chat group, and you just have deep trust and deep vetting You know, and that's why is that, you know, on chat that different than in person? But these are all norms and trust that have to get built out. And a lot of it's not software, it's humans.

I'm curious, the decision to go to Andreessen Horowitz and I mean, both to get— there's one decision, get back into venture. Yeah, I'd like to hear about that. And then how did you pick the venture firm you did?

You know, as we talked about this moment last fall of seeing OpenClaw, reminded me for the first time that you can build consumer startups again, where the models are a piece of the puzzle, but are not the actual value. And the value is really in the software that's created around it. Before that, while I was working at Apple, I really wanted to make sure that we got Siri back and shipped. I also had the belief that like to build a great startup right now, you actually need model innovation. And if you don't have researchers on staff, you aren't building a startup that is truly differentiated. But once I saw OpenCLAW and then all the things we've seen built since, and even what we were building with Siri, you know, Siri, Apple did a great partnership with Google to use some of Google's frontier models, but it still is so much software built into Siri that is way more complex than the model itself reminded me you can build consumer startups again. So that got me like, you know, as I spend, you know, my 50s, you know, I'd really rather be helping companies and I really want to help the startup ecosystem again. And I believe it's possible. And then when I thought about that, you know, I had an incredible experience at Greylock. What a great set of partners. You know, they've decided they're not as focused right now on consumer, and, you know, maybe they'll get back into that again, but as I talked to them, that was not their focus. And so I thought, of all the firms that I've gotten to know over the years and seen, which one did I see really stand by founders always and was really built with a whole different way of supporting founders? And, you know, of all the bad things I've seen in venture, there's a lot of stuff that ends up becoming anti-founder that never gets talked about in the world. What's the firm that I felt really had, had never crossed those lines? And that was Andreessen Horowitz. And so I knew a couple of the folks there and was fortunate enough to kind of pitch them something that I wanted to, to do and be part of the venture ecosystem and ended up becoming a great match.

So you're like investing or there's another layer to this?

You know, I'm doing a combination of supporting investments and helping companies And, you know, I'll be helping us do new deals, but it's unclear. I'm not necessarily going to lead a whole bunch of deals myself.

Interesting. Like, you want to sort of like give product advice to startups?

Yeah, really give pro— I mean, give product strategy, growth advice, and really be part of the team that's helping to support companies. Because that's really what Andreessen does, is it's not just you get one person who you can call sometimes, but you get actually a team of support. And so I'm meeting a ton of founders, new founders. I like to say, future founders and existing founders in the portfolio and helping them be great. And as I find new founders, I'll partner with them on getting, you know, new, you know, investments done with Andreessen Horowitz, and we'll get to go build great companies.

What is sort of the growth playbook at the moment? Or, I mean, how do you— I mean, I care about this in my own business. It's like, oh, for a while email was a core one. Uh, it's so hard to get customers and attention. I mean, Andreessen Horowitz has succeeded in sort of the launch.

Yeah.

Like, I feel like you guys deliver like launch in a box. I think there's some, you know, pushback that, all right, that playbook has been done. And like, it only sustains if you keep coming up with like, what's the new one that's differentiated? And like, all right, you've got the launch video, that's now table stakes. What's the next thing? So there's always an arms race no matter like how, how good you get. But I don't know if you were talking to Yeah, like a consumer seed startup that's like, I need to get 1,000 people on this app. Is there a platform you're bullish on or what's your advice?

It's a great question. And if you ask this, and it's one of the things that it's not yet clear that there's some new growth distribution vector in AI. There really are the channels we all use to communicate and spend attention on. But I always tell people the first most important thing is have a great product. Have a product that people love, use, come back to, find a place in their life and use regularly. Now, for somebody like you who's doing a newsletter and a set of events, it's have a community that is like, I want to read what Eric is going to say next. Every time I read what Eric has to say, it's smart, it enriches my life so that I want to keep coming back. For a product that you use, it has to have that same experience. I am glad I used the product for this purpose and I want to keep doing it. I talk a lot about what is the action someone's doing in your product, What is the purpose of why they're coming to your product? What is the cycle at which you expect them to use your product? You need to deeply understand that. And without that, growth doesn't matter at all. Okay. But once you have that, then it's finding ways to tell your story. And there's different channels that depend on the product and who you're trying to reach. If you're trying to reach people in the tech community and the tinkerers and everything else, Twitter is an amazing platform to have conversation. And it's very obvious that as you have good conversations there, they can spread and they can grow. If you're trying to target teens, you need to be using the platforms that they're on and speak to them in a way that is meaningful to them, whether that's, you know, more on Instagram or on TikTok. And we've seen incredible growth that comes from engaging with creators who tell stories themselves about the products that they use. And the creators are getting really, really savvy. Now, sometimes they are paid, but even then, they only advocate for things that they believe in. And, you know, some of the best strategies I've heard are You know, if you want to go talk about something that's health and fitness, don't necessarily go to the health and fitness gurus. Go to the people who are telling stories about their daily lives, the moms, the, you know, young girls or whatever, and introduce the health product to them, and then they'll talk about how it enhances their life. And so now the people who follow them and are interested in them hear their regular stories of their day and the new products they're using too, as opposed to the health and fitness guru that just keeps pitching another health and fitness product every single week. And so there are so many different ways to take your story and the product and find ways to get it in the right hands of your users. But you fundamentally have to know what the— it has to deliver on that or the loop won't work at all.

Josh Elman, thank you so much for coming on the podcast.

Eric, thanks so much for having me.

That's our episode. This is the Newcomer Podcast. Thanks for listening. Please like, comment, subscribe. Go find us on our Substack at newcomerpodcast.com. If you're hungry for more podcasting, I also have a talk show called the Cerebral Valley Show. We love your comments. Thanks for all your support, and see you next week.