Pod Insights a16z Podcast A True Second Brain About this episode How many people spend more time organizing their “second brains”, instead of leveraging the information within them?
With consumer AI now capable of processing simple language prompts and interfacing with unstructured data, is the landscape of information management on the brink of a transformative evolution?
Founders of Mem, Kevin Moody and Dennis Xu, plus writer Nat Eliason, explore what’s gotten in the way of a true second brain, and how AI may finally unlock what “knowledge management” tools have promised for so long.
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Speaker 00 Our brain isn't a filing cabinet. It's very messy. It's very interconnected.
0:05 Speaker 01 The same search for multiple different people should actually yield different results because you think about the world differently. Imagine if there was just a digital colleague on every team who could sit on every single meeting, read every doc that was ever written, and had all of the context that was shared across the company, and knew exactly what information they could share with who and when they should share that information.
0:30 Speaker 00 You can delete 95 to 99% of everything you have and never feel that pain for the rest of your life.
0:39 Speaker 01 Does it actually adapt to who you are? Does it learn about who you are? And does it then apply that in future interactions with you, like a human would?
0:47 Speaker 01 I think the running joke here is klippy. It was just ahead of its time.
0:51 Steph Smith? Since the rise of the computer, humans have been gripped by the idea of having a second brain. But has reality held up to that promise? How many people spend more time organizing their second brains instead of leveraging the information within them? How many people constantly look for a better-to-do app instead of actually checking the to-dos that they're organizing? And how many people have an endless stream of data that they'd love to one day process, but they just don't have the right tools or time to do so? Due to the structured directive nature of computers to date, computers have been a shell of maybe the second brain that we've long hoped for.
1:33 Steph Smith? Still incredibly powerful, but also requiring discrete instruction from the director. But within the last year, consumer AI has shown up to the party, now capable of processing simple language prompts and interfacing with unstructured data, possibly fundamentally changing this game. Today, together with the founders of Mem, Kevin Moody and Dennis Su, plus writer Nat Eliason, we explore what's gotten in the way of a true second brain and how AI might change the game, finally unlocking what knowledge management tools have promised for so long. So, can our second brains finally work for us? Let's find out.
2:15 Steph Smith? As a reminder, the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com/disclosures.
2:45 Steph Smith? Let's start with Nat. Nat has worked on a ton of projects over the years, but primarily considers himself a writer. At one point, you were writing a lot about productivity, knowledge management, and that's one of the reasons that we have you here. Tell me about why you got into that space.
3:01 Speaker 00 At some point, I just got interested in it. I think at one point, I adopted the belief, as many do, that the more I learned about productivity and the more knowledge I collected, the more financially successful or happy or whatever I would be. And ended up going very hard down that track because a lot of people do, once they get that idea in their head. And something with the tools just clicked very well for me, where something about the way I was using it wasn't as immediately intuitive to people. And then in particular with Rome, for those who remember, that it was really like a mania, I think.
3:42 Steph Smith? It was.
3:43 Speaker 00 I just got very lucky that I was super early on it, was using it before a lot of other people. And I think I was one of the first people with a small to medium sized audience who talked about it. And then I think that helps people with much larger audiences pick it up.
4:00 Steph Smith? Tell me a little bit more about your saying, you seem to have this intuition about how to use them. And then you shared that with other folks. What was it delivering for you and then subsequently others?
4:11 Speaker 00 That's a good question because I'm not sure that I could honestly say. All of the effort has delivered a compensatory amount of value in the sense that even as a kid, I very much enjoyed finding interesting bits of knowledge and then sharing them with people or trying to explain them. And so, almost like this kind of digital magpie, I enjoy just collecting shiny bits of information. To the extent that it's helped with writing, most of it hasn't. The only thing that I could say definitely has really helped is a habit that I've had for 12 or 13, 14 years now of like taking notes from the books that I'm reading.
4:55 Speaker 00 And so I used to do this on Kindle and now I do it on physical books and use Readwise to extract the highlights. But the medium that I organize those highlights in doesn't seem to matter because I had them in text files at one point. I had them in Evernote, I had them in Notion, I had them in Rome. Like I've moved them around, I've lost a lot of the formatting and additional annotations so many times that it almost doesn't matter.
5:18 Steph Smith? You're almost saying it could be on a stone tablet and as long as you have that there? Or is it maybe the searchability? There's only maybe a few functions that you need in order for it to be useful, not all of the feature set.
5:29 Speaker 00 Yes, I think it's really the searching, the indexing and the collecting that are very helpful. But pretty much all of the other features often get in the way or become distractions or give you this sense of productivity, right? Like, oh, I'm doing all of these, I'm organizing my knowledge. You probably should be writing or like playing with your kids or doing something else. But expanding on that, the thing that's really been helpful is the actual reading the books, right? Like, I wish I had spent more time reading more books instead of continuing to hyper-optimize my highlights from the books I had already read.
6:01 Steph Smith? Maybe there's something meta here though, where to your point, it's like, because you feel like you're being productive with the notes that you take within these knowledge management tools, it makes you want to read more books because you feel like they're more useful. But at the same time, it does seem like there is this false understanding of what you're getting from these tools. But one thing that I noticed was when you and a few other creators saw Rome, like you said, there really was this almost like mania. And maybe something behind that mania, at least it felt to me, was as though there was an unlock. As though the prior knowledge management tools were lacking something that this now provided. So maybe you could speak to that experience.
6:41 Speaker 00 Yeah, it's funny because this all happened at an interesting time where I had gone through Thiago Forte's Building a Second Brain a couple of years before. And when I went through it, I was having these conversations with him about how much I hated Evernote. Because it really forced you into this hierarchical way of organizing your information. And so you had a folder for home or whatever, and then you're putting your notes about your house in there. And you have a folder for books, and you're putting all your book notes in there. But pretty much every piece of knowledge should live in multiple places. And you could also do tags and stuff, but the tags felt clunky.
7:20 Speaker 00 And so what was exciting to me about Rome when I saw it was it introduced more of a like knowledge graph. I think they might have invented that or popularized it for note taking tools, right? Which is really, I think, how we think about things, right? Like, that's more of how our brain works. Our brain isn't a filing cabinet. It's very messy, it's very interconnected, and it felt like it fit that much better for me. And that's what made me so excited about it, and I think made so many other people excited about it. We kind of take for granted, if you're getting into knowledge management today, almost every tool has this, right? Right. Notion very quickly adopted this same interconnected backlink structure that Rome popularized. And I think that's great because I use it in Notion all the time now. And that feature in particular was what felt so magical.
8:05 Speaker 00 Because to the point that we talked about earlier, where very few things in these tools are actually useful for improving the creative output, which is, I think, what you're really optimizing for by using them. Search is very useful, and the captions are very useful. And I do think that these bidirectional relationships between bits of information are super useful as well. So that you can log a few connections on something, as long as it's connected to two or three other things. In your second brain, or whatever you want to call it, it will be very easy to resurface it later, much easier than if you have to pick an arbitrary folder to put it into, at least for me.
8:44 Steph Smith? You have to remember the route, right? You're saying with Evernote, you have to remember, oh, there's this thing hidden in four layers of folders. And I have to remember it's there versus you're saying the interconnectivity allows you to surface it more easily, almost like stumble upon it, would you say? Or?
9:00 Speaker 00 Yeah, a little bit more like that. And the thing that I liked about it, too, was any hierarchical file system requires very deliberate placement of each piece of information. You have to think about it a lot where you put it. It might not seem like you do, but you kind of do because you're saying, okay, in 10 years, if I don't know I need this piece of information, how do I accidentally refine it? Whereas with a much more like widespread, hyperlinked, knowledge graph style, you can kind of just like slap 10 different relationships onto it and you'll probably be able to find it again in the future.
9:30 Steph Smith? That's so interesting. I was filing something away because I'm still in ancient times, I still use Evernote. But I was filing something away and I had this thought as I was doing so yesterday of just, I will never see this again. It was in that folder where I'm like, I haven't revisited this in years. And so why would I in the future?
9:48 Speaker 00 I had a few people reach out to me over the last week or two because Evernote's shutting down their US office and a lot of people are migrating out to Evernote. And they said, hey, I need to migrate all of my stuff into a new tool. What do you recommend? And I said, well, like, don't migrate anything. Just like, start using a new tool. And then for the next three months, see which pieces of information you end up needing and then go migrate them. But what you will probably find is that you can delete 95 to 99% of everything you have and never feel that pain for the rest of your life.
10:24 Steph Smith? It's so true. And in a way, it reflects on the fact that so many of us are digital hoarders, where if we did the same thing with physical items, we'd end up in these homes, just, you know, boxes and boxes and boxes. And in a way, because it costs so little and it is kind of out of sight, out of mind, we do it digitally. But that kind of brings me to this question. Do you think that maybe the way that these knowledge management tools haven't quite lived up to their promise, is that just because we misunderstand the way our own brains work and what we really need? Or is it a facet of these tools not having the right technology, maybe the right feature set yet?
11:02 Speaker 00 I think. It might be more of a false promise issue. Because I think if you look at, say, the Notion marketing or the Evernote marketing, they advertise themselves as a wiki, as a digital filing cabinet. And they do those things very well, right? They allow you to collect and organize information. And I think the false promise that many of us believe or have bought into or whatever is that if you organize information, you will write better articles, you will make more money, you will be happier, you will get all of these benefits from it, right?
11:46 Speaker 00 That one, I think, is probably just not true, right? Like Andy, he had this great tweet where he was like, you know, for all of the influencers talking about personal knowledge management and note taking in those things, remarkably few of them have written a great book or won an Emmy or an Oscar or like actually done anything that incredible with their knowledge management systems. The main use case seems to be talking about your knowledge management system.
12:16 Steph Smith? Yeah.
12:16 Speaker 00 And I think that he was kind of like brutally honest or brutally correct about it. Like I looked at that tweet and I was like, wow, I feel very seen right now.
12:27 Steph Smith? I mean, I've seen other tweets similar where you're like, yeah, if you want to learn from the best founders, they're busy building companies, right? Not writing tweet threads.
12:35 Speaker 00 Yeah, and there are obviously exceptions and things, but I make this analogy all the time that 99.9% of the best books ever written were written on a typewriter or by hand. Like, if you think that you need a good writing app to write your book, you're just wrong. Like, it's just not true.
12:50 Steph Smith? History has shown us this is not true.
12:52 Speaker 00 Exactly. So I do think that there is this big element of the other thing they're delivering is fun. Right? And we don't like to admit that the reason we read productivity books or watch motivational Youtubes or stuff is because it's fun and makes us feel good. And that's part of what they're delivering too. Like, it's fun to organize your stuff, right? And I had this really funny experience. Because our older daughter's almost two now. And one of her favorite hobbies is to just organize random things, like opening the pantry and like pulling stuff out and like moving it from one container to the other, or taking all the mixing bowls and like sacking them in the correct sizes. And I'm watching her do this and I go, oh. Like, this is just hardwired in me. Like, there's this part of my brain that likes moving things around and putting them in a satisfying order.
13:37 Steph Smith? That is so good.
13:38 Speaker 00 And like, that's what I'm getting out of these tools.
13:40 Steph Smith? That is so good, because we've all been there, right? Like, oh, yeah. I mean, the most meta thing, right? You have in your Evernote or in your Notion, organize Evernote. And it's just like, this is like so circular.
13:50 Speaker 00 Or everybody's had the experience of, maybe not everybody's had this experience. Weird people like me have had this experience of getting a bunch of things done and then looking in your task manager and being annoyed that you didn't put those things in there first.
14:00 Steph Smith? Oh, yeah. And then you add them.
14:01 Speaker 00 Then you add them and then check them off so that you can feel good about yourself for the day.
14:06 Steph Smith? I know, and you somehow convince yourself that you did the right thing. Instead of following the things you had outlined were important, you did unimportant other things, but they still must be checked off. No, totally. So I think you're right. I think there is this level of just humans wanting to feel productive and these tools can sometimes validate that. And sometimes I do think that they can help us. I mean, if I truly had to remember off the top of my head, all of the, quote, important things that I had to do, I just couldn't. So there is a level of utility here.
14:35 Speaker 00 Totally. I mean, the number of times I had this experience of writing an article and then I think like, I feel like I've seen this or I've felt this idea somewhere before. And then I look over at my bookshelf and one or two books pop out that remind me of what I'm talking about. And then I go into my note-taking tool and surface my highlights from those books. And then I like immediately have a quotation that I might have saved 10 years ago from a book that I can use in the piece that I can reference. Like, that's so useful.
15:02 Steph Smith? Right, right.
15:03 Speaker 00 But the useful things were reading the book and capturing the highlights and then being able to search for it. And it kind of like stops there.
15:08 Steph Smith? Right, and it's being able to get the information you want at the right time and not having to do as much digging. And so with that in mind, if we think about things along the spectrum of like actual utility that these tools are indeed bringing, and then this false promise. Tell me a little bit more about how you've maybe moved closer to that utility side of the spectrum.
15:28 Speaker 00 I really think the next big utility is going to be a note-taking tool that offers custom GPT or whatever embeds as a service for your knowledge graph.
15:41 Speaker 00 Because I've seen people do it. Dan Schiffer has that great project he did where he put all of Andrew Huberman's podcast transcripts into a GPT-embed model and then you can just query it with anything you want to know and it feeds you back Huberman's answers from podcasts.
15:56 Steph Smith? Right.
15:57 Speaker 00 Theoretically, any note-taking app should be able to offer that. I'd pay like a thousand dollars or more for this, very happily. Where I could just say, okay, here are my notes from like four or 500 books now, thousands of articles. Every little like idea I've written down over the last 10, 15 years, here's all of it. I just want you to like index it and then as I start writing, just give me little like pop-ups along the side or when I ask for them, give me little pop-ups that say, oh, you had this idea or oh, this idea from this book or this idea from this article or whatever. Like, these are all relevant to what you're doing right now. Because I'm pretty sure we can actually already do that with the tools that are available on the market. Yeah, nobody's just created a consumer side version of it. That's like literally a second brain, right? That's literally
16:42 Speaker 00 your previous thoughts and work and effort being processed by this external cognitive entity and then fed back to your like current working mind or whatever to enhance what you're trying to do without you having to go look for it. And that would be incredible. Like, that would be so useful. I would use that non-stop every day.
17:02 Steph Smith? And if we just break down what the difference is there, it feels like maybe something you're pointing at is the productivity. You are having, as you're saying, a second brain support you in a way where you don't have to go and say, hey, I need this. Let me go search for this. It's popping up and saying, hey, I've actually gone through all of this unstructured data that you've compiled over years and years.
17:22 Speaker 00 Yeah, I would say it turns your note-taking tool from a means to surface known unknowns. So, you know, right now I can say, I know that. I know this thing, but I don't remember it exactly. I know that I have notes from what your food ate, but I don't remember what this note was. I will go find it. It does that very well, but with this, it could provide you with your unknown unknowns.
17:48 Steph Smith? It's making connections that you haven't thought of yet.
17:50 Speaker 00 Exactly, exactly. It can look at every single sentence of what you're doing and say, oh, you actually wrote an article about this six years ago. Like, I forget about articles I've written all the time, right? Like, somebody will say, oh, I really love that article, or they'll send me links, something like, oh, my God, I wrote that?
18:05 Steph Smith? Have you ever had that experience where you go reread it and you're like, did I write this?
18:09 Speaker 00 Yeah, yeah, it's super weird. Totally. Yeah, or impressed, or maybe like annoyed sometimes, where there's this feeling of, oh, I already wrote this article three years ago, and I kind of like forgot, and now I'm rewriting the same article. And that would just be so powerful, because you don't have to try to remember the surface level of everything you've stored. Like, it's handling that for you and just feeding you back things that are yours. Because there's definitely going to be this phase, I think at least for a while, of AI helping to enhance high-performing creators, right? Like, if you look at GPT-4's writing, it's better than 90% of people, probably 95 or 98% of people, but it's like way worse than the good writers.
18:48 Steph Smith? Than the top 1%.
18:50 Speaker 00 Yeah. And it's actually pretty hard to get a ton of value out of it for the parts of writing that you're good at. But if it can actually surface stuff for you to use, right? Like, if it can handle some of that search for you and whatnot, so that you can produce a lot more, produce faster, right? Like, that starts to get pretty interesting.
19:09 Steph Smith? Yeah, you can imagine how the unstructured nature of AI can actually maybe surface some of these more interesting connections, where it's, again, you're talking about one concept, but it's like, hey, this isn't directly related, but there's a similar arc, there's a similar learning.
19:23 Speaker 00 Yeah, my dream tool. Would just be a blank page where as you start writing, words get underlined or sentences get underlined and like very subtly. So it's not going to interrupt your flow, but it's there as a little note. And then whenever you want to take a break, you can go back and hover over it and it'll pop up and show you, you know, the stuff that you've written about or done or said in the past might be related to what you're talking about right there. And then you can surf through it at your leisure and all of this just being done on the fly as you write without you having to interrupt yourself or without being interrupted.
19:57 Steph Smith? Right. And what's really interesting about that is we've obviously seen how different AI tools have been used for code. And when I think of many of the UIs that have been designed for code, they originally started with this idea of like, oh, it's just going to write the code for you. And similarly, they came to the understanding that actually developers still want a code, but they just want suggestions, especially when those suggestions are likely to help them. Right? And so you can kind of design the tool to say, oh, I'm pretty sure if it's like a 95 plus percent chance this is related, I'm going to showcase it. People even have their own dials to say, I want it to be as broad or as precise, depending on what kind of article I'm writing.
20:37 Speaker 00 And what's really magical about that is the more you write or have written, or the more you do, the more useful the tool becomes and the more powerful you become. ChatGPT is a very, or really any of these is a very democratic tool right now in the sense that everyone has access to the same quality of recommendations. But if you have your own private custom model that's been trained on everything you've ever done and your style of writing and really you writing at your best. Because I think everybody who writes has had this experience of they sit down and they're just on fire and they bang out an article in 30 or 60 minutes. And at the end of it, they're like, whoa, like that was sick, right? Like who wrote that almost? And you're sad when you don't get that feeling because you know it can be better.
21:21 Speaker 00 You can be really on it, but you can't have it all the time, right? You could theoretically train it on your best work. And then have it help you configure anything you're writing to emulate you at your best. And that's a pretty magical thought, too. But I love that idea of it helping you become the TEDx better version of yourself. As opposed to what it does now, where it basically just like raises the floor for everyone.
21:44 Steph Smith? Right. Yeah. I mean, as we discussed, like maybe a true second brain. So on that note, final thing is just if you imagine a tool like this where an AI can parse your personalized unstructured data. It can be proactive and it can surface things as you need them. What do you think this unlocks?
22:02 Speaker 00 There's this interesting element or world that we're getting into of the valuable thing is the ideas. But everybody receives ideas in different ways. For example, like one thing I realized about myself at some point is that I have a very hard time processing auditory information. Like I don't remember it. It just doesn't really imprint for me. But when I read things, they stick with me really well. And so if I read a book, I can remember a ton from it. But if I listen to an audiobook, I don't remember anything. And some people prefer the like spunkier punchy Twitter thread style. Some people prefer the long form essay style. Like some people would prefer a short form video of the idea, right?
22:44 Speaker 00 And so if you have an idea and you can turn it into one of those forms. Something that's trained on you should be able to help you produce those other forms of your work to help that idea reach the most people possible. And we're seeing something else in tandem with the potential for that, which is like platforms. Aren't really letting you use them as funnels anymore. Early internet marketing with social media 10 years ago or whatever. Like your Facebook and your Twitter and whatever were all to drive traffic back to your site. Right. Where you could get email science or sales or whatnot. And it's way harder to do that now. Like, it's really
23:24 Steph Smith? Well, because they actively punish.
23:26 Speaker 00 Exactly. Yeah, yeah, yeah. They punish it. They don't want you to do it. And so you actually want to create native stuff for each of those platforms for the people who want to spend time on those platforms because that's what's going to get rewarded. But now you're like not creating one article. You're trying to create like 10 assets off of each idea, which is a huge time suck. And so it's interesting to think too, like, what would it look like if you had this kind of like content assistant? Who you train not only on all of your writing and stuff, but on your preferred format for short and long form videos? And train on your voice so that you can immediately create like a podcast from your articles and all of these things. So that you can be this like hyper distributed idea person. That world is not very far away. It's probably like less than two or three years out.
24:10 Steph Smith? I think what you said that's interesting there is in your unique way. Cause you can hire agencies, you can bring on other folks and many people do that already. But I think maybe the nugget there is if you were able to take one nugget and repurpose it in your unique style over and over. That does seem special and that does seem like you can't quite do that today.
24:29 Speaker 00 Yeah, you made a really good point there about agencies, right? Because I've been reached out to by a number of like short form video agencies where they say, oh, we looked at your TikToks and we think they could be better. Here's an example of how we improved it. And I looked at it and I'm like, I've seen 10 other people on TikTok with this exact same style of editing. You have templates that you can just copy and paste for everyone. And it probably works to some extent, but then people start to notice, right? They start to see the edges of it. And so if you can actually have something like this where it learns your preferences, right? Like the fact that mid journey doesn't remember the images that I like and their settings to recommend those kinds of settings for the future is very annoying. Again, that doesn't seem like a complicated thing for them to do. But obviously they have a hundred other product decisions.
25:12 Speaker 00 But I think that era of personalization using your past work like is coming and I'm very excited for it.
25:19 Steph Smith? Yeah, it's hard to believe that won't be on the horizon. Speaking of what might be on the horizon, next up is our chat with the founders of Mem, a company trying to utilize AI to deploy a version of the second brain that Nat and I were just exploring, but not just for individuals. In this conversation, we dive into the why now, but also the cost of making this a reality. But first, here's a stat about just how much productivity is lost within organizations.
25:50 Steph Smith? Knowledge workers apparently spend 2.5 hours a day, which is around 30% of their time in a workday, searching for information. And so I'd love to just hear, like, what are your reactions to that?
26:04 Speaker 03 Yeah, I almost think that's probably an understatement, the 2.5 hours number. It's definitely still relevant, if not increased since 2016. I think in part because it's so much easier to produce unstructured information, share information. The barrier for creating content and sharing it within your team and being productive as a part of a team has gone down. But I think something interesting there, you mentioned 2.5 hours per day just searching is this probably hidden time sync too, which is you don't even know that you are looking for some sort of shared piece of information. You just begin doing your work for the day or for the week. And at the end of that week, you realize you've just reinvented the wheel, that somebody else on your team or some other team in your company
26:49 Speaker 03 did that exact same work. And it already exists in some spreadsheet somewhere or some document somewhere.
26:55 Speaker 01 And then I think the question I have in my head is how much time are people actually spending then taking those things and then performing really mundane transformations or tasks on top of what they found? And I would imagine that's kind of at least another few hours, right? So I think what's really interesting is we're now entering this world where a lot of that work and not just search, right? Search is kind of just the beginning of really where people are going to get value out of these new technologies. But I think it's going to be really interesting to see how the entire workflow of a knowledge worker changes over the next few years.
27:31 Steph Smith? Exactly. And I feel like if we kind of go back and look at the arc of knowledge management, like, let's truly go all the way back. You get the ability to write and people were doing this on stone at first. At some point, the printing press arrives. Then you have individualized typewriters. The computer shows up at some point. The internet gives you access to almost everyone around the world these days. And so you have this promise of technology, which has rung true. It has given people the ability to codify and share information and knowledge in.
28:05 Steph Smith? You could say exponentially better ways. Yet, at the same time, if we bring ourselves back to that stat, it's kind of surprising that there still are just incredible inefficiencies in collaboration. What do you think is preventing us from having this really streamlined back and forth between folks within an organization or even as an individual?
28:25 Speaker 01 What's really interesting is all of the things that you've mentioned, all of these devices in the past, they've actually just primarily been broadcasting devices. If you really think about what is core to collaboration, sharing and broadcasting is only one aspect of that. It's only one component. The other component is how do you actually digest? And how do you actually consume that piece of information? Right? We've never truly had really hyper personalized consumption devices. Search is very much a personal problem, right? It's very much about, hey, I'm looking for this piece of information. And the same search query, right? The same search for multiple different people should actually yield different results
29:08 Speaker 01 because you think about the world differently. And the tools that we've used up to this point really just don't model that reality.
29:15 Speaker 03 If we look at it from the perspective of one piece of information and its lifecycle, when information in an organization especially is captured, it's oftentimes really hard to know all of the people and all of the contexts in which that one piece of information could be useful. And so if you think about a book being written, it's useful when somebody chooses to read it. And how people come to discover that book is usually it's recommended or it's marketed. But when you think about an organization, so much information is getting created on a daily basis. And oftentimes, the times where that knowledge or institutional knowledge would be best used is weeks, months, even years down the road. And so
29:59 Speaker 03 The best effort attempt historically to organize that information is to use a traditional knowledge base. And the person that's capturing that information is putting in a lot of time and effort into trying to organize it into a certain folder or a certain file structure. And despite their best efforts, at the end of the day, that structure makes a lot of sense to them and varying levels of comprehensibility to everybody else on their team. And usually by the time they move on to another team or another company, that organization makes zero sense. And so I think kind of one fundamental unlock and trend we're seeing and that we're pursuing is that historically, after information was created, the onus was on the creator of that
30:40 Speaker 03 information to not only distribute, but also organize that information within a company, for example. And now we think that there's actually this opportunity for information to not need to be organized and instead for some automated system or some AI to be able to come in and actually match every piece of information to the context and the situation in which it would be best used. That way you can kind of get past all of those inefficiencies that happen.
31:07 Steph Smith? Totally. It seems like there's maybe two aspects here. One of them is this idea that humans want to document, but in order to document, you really just, like you said, you have to spend so much time. Like, some people spend more time organizing their second brain than actually using it, right? So that's one part, but then there's also another part, which is just there's already this vast amount of unstructured data that other people have created, right? And so, given the technology that's evolved, like, what is the opening here? Are there new approaches that exist now that we have access to these LLMs? Or how are you guys thinking about kind of turning the page?
31:44 Speaker 01 The organizational structure that we rely on is still very much so the folder. And the folder is kind of this thing that was this like skeuomorphic representation of this filing cabinet, right, that we used to have. And then when computers were first developed, like in the 1950s and 60s, we just said, hey, well, people understand how filing cabinets work. So why don't we just put filing cabinets on this computer? So that it's a little easier for people to actually adapt to this new technology. And then that kind of begs the question, well, it's been 60 years since that time, right, where computers have been around. And what makes this moment really special? Where we can actually say, hey, can we actually eliminate the folder?
32:27 Speaker 01 And we think the answer is yes. And part of that is, I think the missing ingredient has always just been a lack of contextual understanding by machines of human language and the knowledge that humans produce. And this new generation, right, of L L Ms, what they actually demonstrate is reasoning capabilities. And that's where I think the magic really starts to happen. When you can marry the reasoning capabilities of an L L M with kind of like the vast storage capabilities of just computers from the past.
33:09 Speaker 03 One way to think about this in terms of people wanted faster horses. And instead, the answer was cars is instead of having manually organized folders, you have automatically generated folders. And at one point, I think that's the direction we thought made most sense. But I think that what has become apparent and what large language models have allowed is to flip that model on its head and recognize the fact that you actually don't want to organize a piece of information into one or even five folders. You want to organize it into almost an arbitrary number of folders, which are the different ways in which you're going to need to recall or retrieve that information. And the thing that large language models now allow us to do
33:54 Speaker 03 is in a very generalized fashion, take any piece of text from any domain. It could be in the medical realm. It could have to do with real estate. It could have to do with technology and be able to comprehend that in such a way that then. When it comes time for somebody to need to make use of that knowledge, they can ask very simple natural language questions, communicate in their own voice and find the right answer without any intermediate categorization step even required.
34:22 Steph Smith? Something that's really drawing my attention is this idea of proactivity, where not only can it reorganize your information in different ways, but it can actually step up and say, hey, this is going to be useful at this time. I know you're working on this kind of project. Have you considered this? Which, again, someone else in the organization maybe has already done. And so how are you guys thinking about that dynamic, which also feels fundamentally new?
34:47 Speaker 01 I think the running joke here is Clippy was just ahead of its time. Right? Fundamentally, the idea of Clippy and this proactive assistant, obviously, I think people have been attracted to this for a long time. It's so embedded in the sci fi fantasies that society has to the world. But in order for proactiveness to not be annoying and to kind of cross that boundary and to truly useful and delightful, you actually have to have a deep understanding and the ability to reason about someone's life and what they're thinking about at that moment and all of those things. Right? And I think if you think historically, honestly, even today, right? With tools like ChatGPT, the quality of our interactions with computers are still very much limited by the instructions
35:32 Speaker 01 that we explicitly provide them. Right? And we used to have to provide instructions and code and obviously the cyberbully. But now, a lot of the time, we can just provide instructions in English. But we're still providing instructions. And this is where things, I think, will get really interesting. If you actually have an understanding of truly of who this person is, of what they care about, of what they know, then what you can actually start doing is kind of flip that model on its head and say, instead of me constantly having to instruct this machine, what if the machine will just do things for me and then confirm afterwards?
36:08 Steph Smith? Yeah, you're evolving with it, right? You can like prompt it initially, but then it can also run on its own. You can give it feedback and it can iterate. Instead of the really like strict rules, hey, deliver me this specific email reminder every morning at 9:00 a.m. But something that's coming to mind is just the cost of all this. Right? We're talking about so many different applications that you're tying into constantly working with data, maybe organizing it to some. Degree, parsing it, saving it somewhere, it has to be retained. So where are we in terms of that? I guess in a way, it's so much more powerful that maybe a new business model can be built around this idea, like you imagine how much people pay EAs and this is kind of paralleling that.
36:48 Speaker 03 Cost is a really interesting question and we think a lot about this because at the most fundamental level, we really want to democratize access to this sort of intelligence and give every person the power essentially of their own personal AI that can help them think better and do more. And so there's really two ways that we're currently focused on monetizing. And the first way is via the individual plan. And I think you can actually similarly equate it to some of the other products on the market in terms of a free to use plan as well as a paid premium all access plan. And that's one that we want to continue to figure out how to drive the price down. But one of the really promising trends is that the cost of compute and the cost of actually
37:33 Speaker 03 running these models is coming down by an order of magnitude every several months. And so given that trend is likely to continue, we think we're going to be able to actually serve all of these things at just an incremental cost above storage. And so storage actually then does become really the limiting factor, interestingly enough.
37:51 Speaker 01 And then I think separately, one of the things you really touched on that I completely agree with is this idea of what if we could actually give each person kind of their personal EA? And that really changes because if you think about the value traditionally of whether it's not taking apps or knowledge bases, it has been almost exclusively focused on information retrieval, right? Kind of the value ends at search after you receive a list of documents. That's where you know, you're done, but we can actually go far beyond just.
38:23 Speaker 01 Retrieving a list of documents now, we can actually do things with them, right? We can help you think through problems, right? That is a much more valuable problem to solve than just pure information retrieval.
38:36 Steph Smith? Yeah, I mean, it's reminding me of the future that many have painted where basically like our AIs interface with other people's AIs and like the idea of meeting time and how expensive it is. Like, there's all these like tools or things that trend on Reddit where you can calculate what a meeting costs a company, depending on who's attending and how much they're paid and C-levels, extremely expensive. This is painting a really lovely picture, but I assume that it's not as simple as just plugging into GPT-4 and calling it a day. And so what are some of the things that you're running into that you're willing to share in terms of getting past those humps to really implement this future?
39:15 Speaker 03 There's a lot that goes on from the shiny demo moment to a really useful product that you or your team can use day in and day out. A lot of it comes down to how much knowledge or information have we indexed and made useful on your behalf? And so a lot of it is actually a data processing and pipeline engineering challenge. So we spend a lot of time focusing on making that really robust, really reliable, really secure, really performant, and that's quite challenging. And then when it comes to large language models and actually applications of those along with search and retrieval, all of those things are evolving by the day. So I'd say it's a combination of figuring out how to apply and innovate on language
39:59 Speaker 03 model technology while simultaneously building this very robust data process and data pipeline. And then on the product side, I think that there's, to your point around, it sounds really wonderful. There's this very rosy picture that can get painted in people's minds. And when an assistive product can't do all of those things out of the box, it can sometimes be disorienting for a new user to a product like that. How do I know what it can and can't do? How do I trust that it can do the things that it says it can? How do I come to learn about its new capabilities as they evolve over time? And so that's the other challenge that we face in terms of as the product evolves, as we unlock these capabilities that we see ourselves being able to unlock over months, not years.
40:42 Steph Smith? Maybe it would be helpful for the audience to understand what some of these capabilities are. What are some of the less obvious ways that this comes into play? Like, how does this really change the way people are working?
40:53 Speaker 03 Yeah, I'll speak to a few different use cases. So one feature that we have in Memex is called Similar to This Mem. And so as you're working on a new document, you're working on getting something done inside of Mem, in real time, we do Always On Search across everything that you've either captured or folks across your team have captured and shared across the organization. And what that ends up doing is, as you're working on something, let's say you're working on some monetization strategy doc. If somebody else across your organization had come up with some ideas months earlier, you would see that surfaced in real time. That way, you would be able to connect the dots between your new thinking and something that had been done before, and save yourself that, as we talked about earlier, that one week of going down this rabbit hole,
41:35 Speaker 03 when instead, you could just reach across and have a conversation with that person to get up to speed. Another thing, though, that we actually are more recently rolling out to our Memex users is what we call Chat with Mem. And really, it's this ended experience where you can simply type in either a question and ask about, "What were the key insights from my user interviews last week?" And get an answer that's synthesized across all of the knowledge that Mem has access to. Or you can ask it to actually collaborate and do thought partnership-style work with you. So for example, generate a sample podcast script for this podcast that I'm going to have with the founders of Mem. And based on just the meeting notes that you had taken from one or two interactions,
42:20 Speaker 03 Mem would be able to actually proactively suggest what that script could look like.
42:24 Speaker 01 Going back to what we were talking about earlier, the real magic happens when, instead of you still having to say, "Hey, give me a podcast transcript for Mem", you don't have to say anything, right? And it'll proactively anticipate that you need that. Provide that for you. It'll look at previous podcasts that you've done with similar people. It'll look at the information that you and your team has about them and then combine that to actually give you that automatically, right?
42:54 Steph Smith? Totally. And something that it's reminding me of is just this idea of like narrow information and then broader information. And there's like a time dimension to it as well. Like your company might set quarterly goals, yearly goals, but at the same time, every day you're focused on your to do list and really connecting those in a way that makes sense to you. And I tweeted this a while ago, but I said, I want an AI that calls me out on my bullshit. It's hard to hear that from a human, right? It's like, Steph, you're moving in the wrong direction. You're not focused on the right things. You said, and even in my personal life, like you said, you wanted to exercise this amount. You said that you wanted to like go to sleep at these times. You're not doing it. And I think there's an opportunity here also. Personalization is so inherent in some of the developments
43:36 Steph Smith? in AI, but to personalize the type of support that you're getting. Maybe some people want something a little more harsh. Other people want something really kind and welcoming, reassuring. But there's opportunity to personalize, not just in terms of like the format of data coming back, but also the type of interaction that people want.
43:56 Speaker 03 I love this example you're giving. We've been thinking a lot about personalization with the chat with MEM and conversational experience, MEM as the assistant. And I think we're going to have to add that as a potential knob is call me out on my bullshit factor.
44:08 Steph Smith? Exactly. Yeah.
44:10 Speaker 01 Yeah. And I think a key fundamental unlock in how we really think about personalization is not just, does it have access to my personal data and my personal knowledge, but rather does it actually adapt? Right? So internally, we call this adaptive AI. Does it actually adapt to who you are? Does it learn about who you are? And does it then apply that in future interactions with you like a human would?
44:32 Steph Smith? Maybe before we close out, one thing that I ask basically any company that's touching AI, which feels like every company these days, is just how do you differentiate in this world? Is it about building your own models? Is it about this like personalization flywheel that you can get going? Is it about the fact that you guys have cleverly downloaded so much data from these companies for them, of course, but just really utilizing that? How are you thinking about that moat, per se?
45:01 Speaker 03 To some extent, it's all of the above of what you just said. But I think the whole is greater than the sum of its parts. And for us, we talked about some of those pieces leading up to this. And to us, it's the personalized AI, the proactive AI, and the adaptive AI. Those are really the pillars that make up what we see as the unique experience.
45:22 Speaker 01 We're seeing all of these insanely cool demos all over Twitter, right? All the time. And these are things that people are spinning out in like two to three days, right? And then they go and they try the product and nine times out of 10 or eight or more than that, they realize it actually just doesn't work. And so I think there's actually this like growing sense of fatigue as well of a lot of people are really starting to kind of not believe that some of these things work. And part of that is because it is extremely hard, actually, still today, particularly when you're working with a non-deterministic system like a language model, to actually get reliable results and to get what you need basically every single time that you need it.
46:03 Speaker 01 And so we think there is this huge opportunity to just differentiate on quality.
46:08 Steph Smith? And on that note of broken promises, I'd love to just close out with your, just your raw reactions around what the future of knowledge management will look like. And I don't know about you, for me at least, up until 2023, it's felt like it can certainly be helpful, this idea of knowledge management or the tools that exist. But it's just, it's not quite a second brain. It takes so much work, so much organization, so much effort to really produce something that's usable.
46:35 Speaker 03 The biggest drawback of the promise of the second brain is how much active energy it takes to maintain. And I think that the future and the unlock of what really the second brain should be is all of the active work will get replaced with passive, automatic work being done on your behalf. And the actual prize of that second brain will be essentially having true augmented thought, augmented memory, and increased abilities to actually work through all of the information you've had before and apply it to whatever challenge you have in front of you.
47:13 Speaker 01 Yeah, I mean, just imagine this. Imagine if there was just a digital colleague on every team who could sit down in every single meeting, read every doc that was ever written, and had all of the context that was shared across the company. And knew exactly what information they could share with who and when they should share that information. That's really the future that we're heading towards. That's the future I'm really excited about.
47:40 Steph Smith? I'm very excited about it as well, and I'm going to hold you guys to this bullshit meter.
47:48 Steph Smith? If you liked this episode, if you made it this far, help us grow the show. Share with a friend, or if you're feeling really ambitious, you can leave us a review at ratethispodcast.com/a16z. You know, candidly, producing a podcast can sometimes feel like you're just talking into a void. And so if you did like this episode, if you liked any of our episodes, please let us know. We'll see you next time.
Previous Chasing Silicon: The Race for GPUs Next Salary Transparency: Clarity or Chaos? Page updated September 30, 2026 · transcript and summary are machine-generated
