YT YouTube2Text

YouTube2Text → Transcripts → Building Muse, Working with Mark Zuckerberg & The Future of AI | Alexandr Wang, Meta

Building Muse, Working with Mark Zuckerberg & The Future of AI | Alexandr Wang, Meta — Transcript

by David Senra · 14,806 words · 2,085 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:02Tell me the history of Muse. The first
0:05thing to start with is probably like
0:06what our vision was when we started Meta
0:09Super Intelligence Labs. And when we
0:11started it, Mark published this memo
0:14that myself and Natt Friedman and a
0:15bunch of others like worked together
0:16with him closely on called Personal
0:18Super Intelligence. And the whole idea
0:20was like how do we make
0:24AI how do we build powerful AI that
0:26actually makes all of our lives better?
0:29like how do we build something that
0:30actually like lifts the human experience
0:33broadly speaking. You know, in that
0:34memo, we we spoke to a bunch of the
0:36ideas and we basically alluded to I
0:39think a lot of what muses ultimately
0:41become, but we were gesturing at it like
0:44very very early like this was in you
0:47know June 2025 um before really agents
0:50had happened in any meaningful way you
0:53know in the summer of 2025 like AI was
0:55just you know in a totally different
0:57spot. But that's where what we wanted to
0:58build towards. That was the mission of
1:00MSL and that was really the where we
1:02wanted to take everything. Then we kind
1:04of just got to work building models
1:07because you know the endto-end process
1:09of developing any of these frontier
1:11models is many many many months. Um like
1:14you have to go through you have to build
1:15your entire stack you have to produce
1:17the data sets you have to you know
1:19pre-train the model then you have to do
1:20pro training the model you have to build
1:22a whole stack for that. like is an it's
1:23a very lengthy end to- end process. Um
1:26and kind of this whole um production
1:28line at the start of this year um you
1:32know Opus 4.5 obviously like really
1:35caused many people including including
1:37us to sort of like see what the future
1:39of agents could be or was and really
1:42brought the sort of like potential of
1:44agents to the foreground and openclaw
1:48was happening at the start of the year.
1:50Natt Freriedman, who you know I work
1:52closely with, was I think the first
1:55person within MSL to uh work with
1:59OpenClaw and try OpenClaw. He had an
2:02experience I would I would describe as
2:05somewhere between terrifying and
2:07euphoric. Like I think he [laughter] he
2:11really like I think went all in. He
2:13trusted entirely. you know, he's told
2:15some of these stories that uh he had
2:17like a a Stripe Sessions interview where
2:19he talked about some of this where, you
2:21know, he told his open claw that he
2:23wanted to drink more water and then, you
2:26know, his openclaw would like watch him
2:28in security footage to make sure that he
2:30was drinking water and tell him good job
2:31and he had like many many crazy stories
2:34like this and it also makes you realize
2:36like, oh, there's something quite quite
2:37magical about what's happening here. And
2:40I remember one of the things that that
2:41Nat told me and this is one of the
2:43things I remember the most is he said I
2:45think there really is something here and
2:48I don't know like early adopter Nat
2:50Freedman is like a pretty good indicator
2:52cuz he was one of the he was one of the
2:54first customers of Stripe back in the
2:56day and then was really ear early on a
2:58lot of open source developer stuff and
3:00he also you know helped create GitHub
3:02copilot and many other things. So that
3:04was like a clear signal. It's like,
3:05okay, there's clearly something pretty
3:07magical about personal agents generally
3:10speaking. But you know, when you
3:12actually kind of trust the personal
3:14agent and and really bring it into your
3:16life, it is the ability to do so many
3:18things. And after hearing all of these
3:20crazy stories, I start using it very
3:22deeply. You know, I think for many weeks
3:24I think both of us were sort of like
3:26every spare moment we would just spend
3:28talking to our open claws and like you
3:30know, you know, really immersing
3:32ourselves in like what personal agents
3:34could be.
3:34>> What was your initial reaction? Euphoria
3:37or terror?
3:38>> Nat at the time he had this um he had
3:41built this prompt to do this like
3:44basically like full psychoanalysis of
3:46yourself. There were like multiple waves
3:49of like deep psychological probing. He
3:51basically wrote this prompt for you to
3:53to use an open claw. And I think that
3:55was like one of the very first things I
3:56did with with my own openclaw. It was
3:58maybe the equivalent of like 3 years of
4:00therapy all, you know, compressed into
4:02this one moment.
4:04>> Wait, I'm curious why when you get your
4:06first shot at open call, this is one of
4:08the first things you do.
4:09>> You know, I downloaded it. I installed
4:11it. And then Nat was like, "Oh, you
4:13should do this." I was like I was like
4:14next to him. He was like, "Okay, I'll
4:16just do the thing you're telling me to
4:17do." And then
4:18>> okay um [laughter]
4:19>> and then you know this crazy experience.
4:21>> Wait say more about the crazy experience
4:23though.
4:23>> Oh well it was just like very intense. I
4:26mean it was like again it was like 3
4:28years of therapy because it was just
4:29like successive rounds of like deep uh
4:32psychological probing cuz it had other
4:35information about me. It had you know
4:36access to my email and access to a bunch
4:38of my photos and like it could like go
4:40and look up other information and sort
4:41of pull it all together. I was very
4:42vulnerable through that process. This
4:44was in February. uh both of us I think
4:48had come to this like joint conclusion
4:50that this is this is clearly like you
4:54know there's clearly something really
4:56special happening here and we both wrote
4:58memos at that time I wrote a memo about
5:01and both of these we you know kind of
5:03like sent up through the meta board um
5:05cuz you know for whatever reason it's
5:06board season I wrote a memo about how I
5:10felt that in many ways this was like the
5:13final consumer product and that the
5:15super agent was something that was like
5:17as far as consumer products went like a
5:20clear end point in many ways. And then
5:23he wrote one called he wrote actually
5:25two uh which had very memorable blames
5:27the clause the law and trust is a must
5:29which ultimately defined um a lot of the
5:32a lot of the core of how we built these
5:34products. So we came to this conclusion.
5:36This was in February. There was
5:37something very important here. Um and
5:40then we built you know our teams we
5:42built a prototype within
5:45like a a week maybe two weeks. Um and
5:49demoed it to the board like it was this
5:50like very fast cycle and many of the
5:53decisions that made its way into the
5:55final Muse product were present in the
5:57first prototype. The Jollybot was there
6:00know the early version of the Muse Charm
6:02was in that first that very first
6:04meeting. Not all of the ideas and not
6:06all the tabs were there but a lot of
6:07them were present. Then from there, you
6:10know, obviously that was February, we
6:12launched the product 7 months later was
6:15like the really really hard process of
6:18taking this, you know, prototype where
6:22it it like sometimes worked and was
6:24magical but most of the time didn't work
6:26and would constantly break and like
6:28going from there to something that would
6:29be highly reliable like you know kind of
6:33like a beautiful crafted consumer
6:35product that we felt confident as a
6:38contribution to the consumer world. And
6:41that that was like this very lengthy and
6:42quite painful process. And
6:44>> before we get there, I want to go back
6:45to the idea that you wrote the
6:46prototype. You guys created the
6:47prototype in like a one to two weeks.
6:50You show to the board. What's Mark's
6:51role in all this?
6:53>> He was like experiencing the sort of
6:57like euphoria/ terror of the product at
7:00the same time as all of us. I think he's
7:01talked about this in various interviews,
7:02but like I think he he used it for a
7:05bunch of things around the home and he
7:07was using it with his kids and he used
7:08it in his um uh in his MMA training.
7:12Like he had it watch videos of his of
7:15himself doing MMA training and like it
7:18would give him tips on like how to
7:19improve. there were like some staff
7:22meetings and by the way like you know
7:23this kind of became a thing for for much
7:25of the the meta team um was like having
7:28these having a staff meeting and saying
7:30like what were the what were the crazy
7:33experiences that you had with [laughter]
7:35um with with your agent again this was
7:37back in February. It's very interesting
7:39to kind of um think back to this whole
7:42cycle because obviously the openclaw
7:45phenomena it peaked and then
7:48disappeared. Yeah, there were, you know,
7:50it didn't it didn't persist and and I
7:52think a lot of people kind of had this
7:53experience which was it was just like
7:56you'd have these magic moments but for
7:57the most part didn't like work that well
7:58and there were all these problems and
8:00then you know it just kind of like died
8:01down and so part of I think the building
8:06the product a lot of the focus of the
8:08product was just like how do you make
8:09this into an actual reliable consumer
8:12experience that you feel comfortable
8:15putting in front of over time billions
8:17of people and you can feel confident
8:19that like most of those people have an
8:21amazing experience.
8:21>> Why were you able to do that and the
8:23developers open call were not?
8:25>> Well, first of all, I think we respect
8:27Peter Seinberger and everyone who's
8:29contributed to OpenClaw immensely. Like
8:31I think Peter is definitely a visionary.
8:32I think whatever happened in his
8:34beautiful brain that created OpenClaw is
8:38magical. And I think maybe the
8:39unglamorous part of what made made Muse
8:42so magical was just really grinding out
8:48and sanding out all of the details to
8:50make the combined model plus product
8:54experience really amazing and we did
8:58many many iterations on the model
9:00itself. We aimed the overall model
9:03roadmap towards building towards
9:06personal super intelligence broadly
9:07speaking but certainly personal agents
9:09as like
9:10>> say more about that and how that
9:11differentiates between your other
9:13competitors.
9:14>> Yeah. Well, this is this is kind of
9:15>> that's actually really interesting.
9:16>> Yeah. It's actually useful to go back to
9:18the moment because if you go back to the
9:22start of the year, February, Opus and
9:24Claude were way ahead in coding, way
9:26ahead of anyone else and they were
9:28exploding. It was the beginning of the
9:30sort of like you know enthropic takeover
9:33in the whole industry and it was a very
9:34scary moment I think generally speaking
9:36which is like oh wow this is you know
9:38they're so ahead on coding and and then
9:40obviously they had mythos and you know
9:41they were they had like this clear lead.
9:43I think the the conventional wisdom in
9:47the industry was that coding agents were
9:50the only form factor that mattered or
9:52maybe the only technology that mattered
9:53or only the only capability that
9:54mattered for advanced AI. you know
9:57there's this incredible force in the
9:58industry to just like go and crowd that
10:01use case and compete with enthropic on
10:03coding agents because of our experiences
10:06and our conviction that I think formed
10:08around this idea of personal agents and
10:10the sort of taking a step back I think
10:13the reason why we did personal super
10:15intelligence in the first place was that
10:17it was also what we believed meta was
10:20uniquely positioned to do and we can
10:22talk more about that but like I think
10:23because of that
10:24>> well let's talk about that now
10:25>> okay we'll just do do the weave. I think
10:27one of the first things I told Mark even
10:29before the whole scale and meta deal
10:32happened and and everything, one of the
10:34first things I told him was I think Meta
10:37is actually a really special company if
10:39we have AGI or ASI or whatever you want
10:42to call it. Because part of the promise
10:45of AI is that for all of us, we have a
10:48mix of things that we want to do and
10:50things that we have to do. And the
10:52promise of AGI is that we're going to
10:55spend almost none of the time doing
10:57things that we have to do because the
10:59AIS will just start doing a lot of that
11:01for us and we can spend all of our time
11:03on things that we want to do. And for
11:05me, Meta's products represent the things
11:09that we all want to do. Like my use of
11:12Instagram or WhatsApp or Facebook like
11:14these are products that represent what
11:16I, you know, I follow my interests. I
11:19follow my hobbies. I follow all people
11:20that I like my friends and the people I
11:22care about like the product span of meta
11:26encapsulates the sort of like what we
11:29want to spend our times doing sort of
11:30realm. I think on a long arc meta is an
11:34incredible winner through all this.
11:36That's like one of the first things that
11:39I believed about meta and super
11:42intelligence. And then ultimately a lot
11:43of those ideas made its way into this
11:46concept of personal super intelligence.
11:47You know, you can kind of look at it a
11:49few ways. One is what's the heritage of
11:51the company? The heritage of meta is
11:54connecting people with their friends and
11:56family. Um, helping them discover their
11:58interests. Like these are things that
11:59are very human. And then there's also
12:02the sort of like you can take the
12:04business lens which is Meta has
12:06distribution to three and a half billion
12:08people using the apps every day. And the
12:12natural
12:14area where Meadow was going to be
12:16successful is consumer AI.
12:18>> It's not work. It's not coding. It's
12:20personal.
12:20>> Yeah. Exactly. So personal agents and
12:24personal AI and and consumer AI, broadly
12:28speaking, were always the zones where we
12:30always felt that Meta was going to do
12:32some be able to do something very
12:33special. [sighs] So that brings us to uh
12:36okay, early 2026. Now you have the
12:39prototype, you have the buy in,
12:41everybody knows Mark's on board and now
12:43you're doing the seven months of the
12:45difficult work of refining this product.
12:48>> Yes. There was kind of the spectre which
12:50was like oh personal agents like they
12:52came and went and we always felt that it
12:55was because you know the the experience
12:57wasn't perfect. It wasn't polished yet
12:58you know all that kind of stuff. But
13:00like we knew that that was something
13:02that we had to like think about as you
13:03know as we were developing this product.
13:05We focused our model development road
13:08map. The top focus uh or one of the top
13:12focuses was towards building on personal
13:14agents and we had like these long
13:16spreadsheets of hundreds or even
13:19thousands of different not thousands but
13:22but more than a hundred like specific
13:24behaviors and capabilities that we knew
13:27the model needed to have to be able to
13:29build the ideal personal agent. Like it
13:31was very very meticulous. And then over
13:33time we built evals for all of those. We
13:36figured out how to train the model to be
13:37really really good at every single one
13:38of those things. And we have these
13:40reviews where we like look at that
13:41spreadsheet and we figure out the things
13:42that we're bad at and make sure that
13:44we're improving on them. Like we did
13:46this like very meticulous process to
13:48slowly just keep improving and sanding
13:50out all the edges of the model so that
13:52it could power the product.
13:54>> Is that really the culture of meta too?
13:56This like constant like sanding of the
13:58products?
13:59>> Yeah. A lot of the culture of meta is
14:00like once you know what you're
14:02measuring, there's very talented people
14:04and you can optimize for those things,
14:06you can send out all the details, you
14:08can make those things amazing. I think
14:09in the sort of like preAI consumer um
14:12product era, like a lot of this was
14:14around how do you optimize your
14:16conversion? How do you optimize the
14:18referral rates? How do you optimize like
14:20every part of the onboarding flow? Like
14:22how do you make all that stuff just
14:23perfect? And then I think the AI era I
14:26think uh a lot of that especially from
14:28Muse has been in the form of sanding out
14:31every detail of the model to be perfect
14:33for the product.
14:34>> I want to tell you about the presenting
14:35sponsor of this podcast ramp. I have
14:38been reading a lot about SpaceX lately.
14:40SpaceX is one of the most valuable
14:41businesses in the world and one of the
14:43main themes in the history of SpaceX is
14:45constantly attacking and questioning
14:47your cost. Ramp helps many of the most
14:49innovative businesses in the world do
14:51exactly that. The median company running
14:53on ramp cuts their expenses by 5%. And
14:56one thing SpaceX has demonstrated is
14:58that a religious dedication to
15:00controlling costs can help actually
15:02increase revenue because you can pursue
15:04opportunities you couldn't otherwise.
15:06And we see that in the RAMP data, too.
15:08The median company running on ramp also
15:10grows their revenue by 16%. So, when
15:14you're running your business on ramp and
15:15your competitors are not, you have a
15:17massive competitive advantage that
15:19compounds over time. Ramp is the only
15:21platform designed to make your finance
15:23team faster and happier. Many of the top
15:25founders and CEOs I know run their
15:27business on RAMP. I run my business on
15:29RAMP and you should too. Go to ramp.com
15:32to learn how it can help your business
15:33save time, save money, and grow revenue.
15:36That is ramp.com.
15:38Deal is how the best founders turn the
15:40world into their talent pool. I've been
15:42studying how history's greatest founders
15:44operate for a decade. And one thing they
15:46all have in common is they understand
15:47that recruiting and hiring the very best
15:49talent is your most important priority.
15:51A players recognize other A players,
15:53which is why top companies like Ramp,
15:55Shopify, 11 Labs, Uber, and Door Dash
15:58all use deal. Many of the top founders I
16:00know have personally invested in Deal
16:02after using their product. And what they
16:04discovered is that Deal is the best
16:06company in the world at building
16:07infrastructure for global hiring. Deal
16:08will help your business hire, pay, and
16:10manage any worker anywhere in the world.
16:13So you can retain the best talent
16:14anywhere and spend the rest of your time
16:16focusing on what you do best, delivering
16:18value to your customers. The founder of
16:2011 Labs has a great description of the
16:22value Deal can give your company. He
16:24said, "We built 11 Labs to break down
16:27language and communication barriers.
16:28With deal enabling us to hire and
16:30support exceptional talent anywhere, we
16:33can accelerate our innovation and bring
16:35more voices, stories, and ideas to every
16:38corner of the world. Deal is trusted by
16:40over 40,000 companies and growing fast.
16:43Learn how they can help your business by
16:45going to deal.comra.
16:47That is deal.com/enra.
16:51>> Did you see this profile that's written
16:53on Zuck in Colossus written by Jeremy
16:56Stern?
16:57>> I saw it. If I'm being honest, it was
16:59like too long. Didn't read. But
17:00>> it's 15,000 words. I just had Jeremy on
17:03the podcast. He's he's my favorite.
17:04>> I saw I saw I saw that. He's he's just I
17:07think he's writing the best profiles on
17:08on entrepreneurs and investors in the
17:10world right now. I think he's literally
17:11the best in the world at what he does.
17:12But in that Zach says he was talking
17:15about the advantage that he has relative
17:16to I think Elon and like Sam Alman and
17:18he's just like listen my skill set is
17:20very unsexy but I'm really good at
17:21building teams and then just improving a
17:23product slowly over a long period of
17:25time and relative to Elon and and Sam
17:28Alman and others. like I'm not going to
17:30lose control of my company and I don't
17:31need to raise any more money and so I
17:32can just do this for an excessively long
17:34period of time. That section of the
17:36profile is popping in my mind when
17:38you're talking about this 7-month uh
17:41stretch of trying to refine and build
17:43muse.
17:44>> I think that's exactly right. Like I
17:46think that it took immense patience. I
17:49mean 7 months like you know maybe it
17:51doesn't sound like that long but AI is
17:53like fierce. seven months in AI time.
17:55>> Seven months in AI time, which maybe
17:56like a decade and like nonAI time. And I
17:58think it it for all of us. Um but but I
18:01really give Mark a lot of credit for
18:02this. Like it took a lot of patience and
18:05restraint and a lot of trust in the
18:08process. And like you'll you can go back
18:10and look at the you know what Mark said
18:13in earnings calls for this entire period
18:16from February through through um
18:18September and he will allude to the fact
18:22that we were building personal agents
18:24and we weren't really trying to hide it
18:25like we were we were building personal
18:27agents and we were talking about it but
18:29I think you know there's a lot of doubt
18:31in the company for that entire period
18:33like people obviously the one of the
18:35dominant Wall Street narratives is like
18:38oh my god Gosh, Meta is just burning all
18:39this money. Like, are they even going to
18:41pull anything off in AI?
18:43>> It's more than that. It's like Zuck
18:45cannot win AI. It's said over and over
18:47again for a long period of time. And
18:48again, going back to Jeremy Stern's
18:50profile on him. He then off the record
18:53starts talking to his competitors,
18:54including one former researcher that is
18:57now at a competitor that used to work
18:58for with Zuck and is now at another lab.
19:01And he goes, "I don't want to have to
19:03compete with them cuz Mark's like the
19:04I think the line is Mark is the
19:06Terminator." [laughter] And it
19:07was like he just does will not stop. And
19:10he's like uh then he talks about um one
19:12of his superpowers that he's never happy
19:14with the state of the product. So he
19:15always wants to constant prove but he's
19:16not an Like he's not you know
19:19dict he's he is definitely a dictator
19:20but like he's not you know rude and like
19:23kind of will destroy the chemistry of a
19:25team. He actually talks about in the in
19:26the piece that team cohesion is
19:28excessively important to him.
19:30>> I think it's all right. I think one of
19:32the magical things about Muse was that
19:35we
19:36really like took quite a bit of time to
19:38sweat all the details before ultimately
19:41coming out with a product because I
19:43think that that personal agents was one
19:45of these product areas that kind of
19:48rested on a knife's edge, which is if it
19:50can reliably deliver these magical
19:52experiences, it has the potential to be
19:53one of the greatest consumer products
19:55ever. But you're dealing with models,
19:57you're dealing with something
19:57nondeterministic, you're dealing with
19:58agents, they can break, they can be uh
20:01unreliable. Like if people have these
20:03unreliable experiences, they think it's
20:04trash. It was one of these things where
20:06like getting to the product to a point
20:09of like quality and value such that it
20:12resonated with so many of the people who
20:14downloaded and used it. I want to talk
20:16more about this process. I'm curious how
20:18much resources, how big the team, how
20:20many people working on it. Do you guys
20:21have any data on like okay I'm using
20:23Muse nine great experiences two like
20:26inconsistent or you know shitty
20:28experiences and then I turn out like do
20:30you have any information on this? You
20:32can see this from like a lot of the
20:33other agent products at the start of the
20:35year like almost everyone turned out of
20:37those and it's because they were
20:38unreliable. We did a lot of user testing
20:42of Muse. Like we did many many rounds of
20:46new people trying to like download it
20:49and and use it from scratch and just
20:51seeing what that experience was like. A
20:55huge part of the premise of Muse and I
20:57think what's really landed is that it is
21:00like a different thing from what most
21:03people have experienced with consumer
21:05AI. Like it is not a chatbot. It is an
21:08agent and it can do agent things for
21:11you. And like I think to a developer
21:14this is kind of like old news because
21:15they've had to like you know download
21:17agents and have been using them for a
21:18while. But for most people like just
21:20even that graduation process like you
21:22know it needs to be explained and needs
21:25to be onboarded on on a very thoughtful
21:28way. And then for sure we definitely
21:29have seen like when because we've done
21:32tests with like different versions of
21:34the model and various model versions and
21:36continuously do those AB tests. You know
21:39the product experience is very very
21:40sensitive to model quality. It's very
21:42sensitive to how reliably the models can
21:46undergo these tasks, how reliably they
21:48communicate with the user about them,
21:49how reliably they like, you know, check
21:51back with the user if there's things
21:52that they require guidance on.
21:54>> But how'd you know it was ready to be
21:56released? Part of it was was pretty
22:00numerical which is like we had set you
22:03know you go back to that spreadsheet of
22:05100 plus things.
22:07>> We had set for each of those rows what
22:09threshold is launch blocking and above
22:11what threshold is good. We had many
22:13checkpoints that like were like green on
22:1580 of the rows but red on 20 of the
22:17rows. We knew that wasn't good enough.
22:18And we had our first model checkpoint
22:21which was green across the board which
22:23was a specially trained version of
22:25Musepark 1.3. So there was like the very
22:28numerical part of it and then there was
22:30just the experiential part which is we
22:33tried the model. We could tell that it
22:35was a lot better than than anything we
22:36had iterated with before. So going back
22:38to that seven-month period you were
22:39talking about, how many people were
22:40working on the product
22:41>> between the model work, model work and
22:44the product work? Like under 200 people
22:47total, which I guess sounds like a lot,
22:49but in big company land is
22:51>> it's actually a lot smaller.
22:52>> It's not very many. That was kind of
22:54cool looking back on this and thinking
22:56about it is like we knew this was going
22:58to be the most important product that
23:00the whole executive team, everyone knew
23:02that this was probably going to be one
23:06of the most important products that Meta
23:08would ship this year, but we had the
23:11restraint organizationally I think to
23:13make sure that like it was a really
23:15small focused team working on it and
23:17that it didn't balloon to this like oh
23:20you have like you know huge swasts of
23:22the company working on because we knew
23:23it was a single work of art. Like it
23:25required strong point of view and strong
23:29focus to like pull that through into one
23:31beautiful experience. Say more about
23:33this. Why did you just call it like a
23:35single piece of art and you need a
23:36single perspective or point of view? I
23:38think one
23:40um
23:43criticism of products that come from
23:45larger organizations and I think this
23:48like is even now starting to plague
23:52a lot of the AI labs is that they become
23:54kind of this Frankenstein
23:57almost like Croninberg amalgamation of a
24:00bunch of different people's points of
24:02view and visions and and beliefs and
24:06it's sort of like it's sort of PM hell
24:08in some sense where like every PM has
24:11like a thing that they're trying to get
24:13into the product and jam into the
24:14product and then all that kind of like
24:15blends together into this like like
24:18smoothie almost. People feel it as users
24:22and consumers like you feel you can feel
24:23when a product is just like it feels
24:25like lots of different people worked on
24:27it and there were like clearly different
24:28people who were gold differently so they
24:30just kind of like jammed it all
24:32together. Muse had to be the exact
24:34opposite, which is like there's one
24:36clear point of view. There is one
24:39cohesive experience that we're trying to
24:41deliver um to the world. I love that you
24:44said that and it just happens to be pure
24:45coincidence. Like we're I don't know
24:46when this is going to come out, but
24:47we're recording this on the 15th
24:48anniversary of Steve Jobs death. And
24:50when you think of somebody like a pro
24:52consumer products that had a single
24:54point of view, you obviously think of
24:56Steve first. I've read every single book
24:58on Steve and the history of Apple and
24:59I've done like I don't know 15 or 20
25:00episodes on them uh for my other podcast
25:03founders, but one of my favorite lines
25:04to describe Steve's impact was it says
25:07that um Apple is just Steve Jobs with
25:0910,000 lives. The products are to his
25:12taste, his perspective, his point of
25:14view. What he wanted to see in the world
25:16is exactly what the the the end user
25:18got. In Muse's case, like Nat Friedman
25:21deserves a lot of credit. like his taste
25:23and sensibilities ultimately dictated
25:25and shaped what that product became.
25:27>> Do you talk about how many people are
25:29using Muse? Are you guys talking about
25:30this publicly or no?
25:31>> We have said millions and then I have
25:36retweeted other people who have shown
25:40graphs where we would imply that Muse is
25:43the fastest growing consumer app of all
25:45time. Consumer AI app of all time.
25:48>> So you won't say a number though.
25:49>> We've said millions. Yeah. Well,
25:51millions could be hundreds of millions.
25:53That's also millions. So, you're just
25:55not adding the the the beginning to
25:56that. I want to like go into like you've
25:58been kind of going like uh and I mean
26:00this in a loving way like unhinged
26:03on X. So, you talk about like your is
26:05this like marketing strategy? Like what
26:07is going on? Like I feel like you're
26:08there's a distinct difference in your
26:10public tweeting since Muse launch. What
26:13are you doing?
26:14>> There's a few pieces of what's happening
26:15here. This has got to be more fun than
26:17scale AI when you were building scale.
26:18It has to [laughter] be. We'll talk
26:20about that later, but it has to be the
26:22past few weeks have been some of my most
26:23fun uh building like yeah certainly my
26:27career and I think there's a few things
26:28happening. So, one is it felt important
26:31that we figured out a way to make Muse
26:34break through. It's a pretty like
26:36torrential going back to this like this
26:38torrential stream of updates. Like for
26:40most people, there's just so they're
26:43just like constantly being like like
26:46kind of bombarded with new AI products
26:48and startups and releases and things and
26:50models and here's another model and
26:52here's another product and here's
26:53another feature and it's just like this
26:55torrential stream. We had actually
26:57meeting large meetings where we talked
26:59about this as a team. It was like we
27:00need to break through and obviously like
27:03meta we have lots of distribution like
27:04we can make sure everyone knows about it
27:06but that's not enough to like make it
27:07break through in an interesting cultural
27:11or fascinating way for the world to kind
27:14of like you know pay attention to. I
27:17separately in my life have happened to
27:21have met and um become friends with like
27:25memers and and uh I have these uh I
27:30affectionately call them my skitso
27:31friends but but people who like
27:33literally they spend all their day like
27:35thinking about memes in the internet and
27:37and understanding that I have people
27:39that are obsessed with Muse and they're
27:41also obsessed with your X feed and
27:43they're convinced that you have like a
27:44hundred muses coming up with memes. to
27:46to be used.
27:49Um, turns out you just have a collection
27:51of schizophrenic friends that are
27:52terminally online.
27:53>> The X feed, just for what it's worth, is
27:55like literally every one of those tweets
27:57I write. There were like a few a quite a
28:00period where I was like, it was like
28:02really hard for me to pay attention in
28:03meetings because I would just like think
28:04of a meme and I'd be like, "Oh god, I
28:06have to get this out." And I would like
28:07[laughter]
28:08it was like moments of inspiration.
28:10There was like um I was uh I was with
28:13one of my like I was trying to have like
28:15dinner with one of my co-workers and
28:17then during the dinner I was like wait I
28:19figured it out and then I like posted a
28:20different me. It was like it was almost
28:23I almost felt like I was like an
28:24engineer solving bugs again cuz it was
28:26like you know you could tell when
28:28something was going to click. But yeah
28:29so one things we wanted to break
28:31through. Another thing that like I just
28:33thought was important is like how do we
28:34make meta cool again? cuz like for
28:38whatever reason despite having like
28:41these incredible pieces of cultural
28:42software um like Instagram and and uh
28:46Facebook historically meta was not cool
28:49for a while and so wanted to make meta
28:53cool and interesting and kind of like
28:54funny and then the unhingedness was
28:57something that happened kind of like
29:00naturally over time when Muse first came
29:03out I was just like mostly focused on
29:05promoting all the interesting use cases
29:06that I saw I was like retweeting like
29:08tens of interesting use cases that I saw
29:11of people using Muse. And then at some
29:14point it was like because I was just
29:16like tweeting so much. There were like a
29:18few tweets where I was just like didn't
29:20think that much about it and just said
29:21it post. And then some of those
29:23tweets like did really well. Have you
29:25seen this like time spent thinking about
29:26it and then banger you seen that? Great.
29:29>> I haven't se but I agree with that.
29:30That's like I just some of these I
29:32didn't spend that much time on. I like
29:34tweeted them. I didn't even check my
29:35phone for a few hours and I was like,
29:36"Oh, wow. That one went really well."
29:38And then I kind of like remembered
29:39something about the internet which I it
29:41wasn't super intentional, but I
29:42remembered like the internet rewards
29:45risk and it rewards like surprise and
29:48like things that like people don't
29:49expect. And so then I just kind of like
29:52went off. A lot of my co-workers uh I
29:54think were somewhat surprised at uh
29:57something that um Nat said for for the
29:59whole week when I was posting all the
30:01memes. He was like, "You know, I thought
30:02we would learn." He said this like in a
30:05bunch of meetings. He was like, "I
30:06thought we were going to learn a lot
30:07about um you know, Muse users and like
30:10how people were using Muse, but actually
30:12like we're just learning a lot about
30:14Alex's mind."
30:16[laughter]
30:17It was a lot of fun. I felt like I was
30:19drawing on all of my internet knowledge
30:21for years and years and years to like
30:24pull into those moments. And I think it
30:26actually, you know, I don't know, we
30:27don't have the tracking and it's like
30:28hard to tell, but I actually think it
30:30actually genuinely made an impact on
30:32grow growth, which is kind of hilarious.
30:34>> I definitely think it did. I I can't
30:36stop hearing about it. So, and all the
30:38people I'm hearing about it from also
30:39downloaded the app and are now using it.
30:41Go back to what you were saying though.
30:43It's like, hey, we have distribution,
30:44but that doesn't make mean that we can
30:45make something that people are going to
30:46come back to and actually sticky. So,
30:48like, how did you do this with Muse?
30:49>> I think it was like, how do you make
30:51this into how do we how do we make this
30:53breakthrough as a product? And I think
30:54there were like a few parts of that, but
30:55I think one is how do you get people to
30:57a match the experience? Like how do you
30:59get someone to to a wow moment with the
31:03product as early as possible? And this
31:05product I think naturally lends itself
31:07to that because I think you know there's
31:09it's like it's AI. It can do a lot. It
31:11can there's like if if someone has never
31:14used an AI agent, there's a lot that it
31:16can do that will surprise you. So that I
31:18think was a big part of it. I think
31:20another part of it was like how do you
31:23harness all of those stories of people
31:26having these wow moments and like use
31:28those to help make the product sing and
31:32make the product fly.
31:33>> How are you using the stories though?
31:34Cuz you you guys aren't running ads for
31:36me, are you?
31:36>> We ran a few ads. I mean honestly like I
31:39think at first it was just the fact that
31:41I would retweet so many of them. Like I
31:43was just
31:44>> anytime I saw someone do something
31:45interesting about news I was like quote
31:47tweet. I couldn't understand why both
31:49Enthropic and OpenAI's um ads are so bad
31:52because remember Anthropic was doing
31:54this huge like outdoor campaign and it's
31:56all about them and same thing with um
31:58Chet. It's like they'd have like the the
32:00icon which I don't even think people
32:02recognize that as a logo and it just say
32:03CEX and just be two people sitting at a
32:05desk. I'm like what the does this
32:06mean? Where it's like you have all these
32:08use cases. All you have to do your ad is
32:10to show the benefit of your product.
32:12[laughter] That's it.
32:13>> You have millions of them. Run those
32:15ads. It's it's so like elementary that
32:17it's interesting that people make this
32:19mistake over and over again.
32:20>> Yeah. I think one thing that we came
32:21back to a lot in in both building news
32:24as well as just like building stuff in
32:26general at me was just like why would
32:29someone care? They don't care about you
32:30or your company. They care about
32:32everybody's self-interested. This is
32:33like I'm there's this very old book.
32:36It's called Scientific Advertising. It
32:37was written in the early 1900s. is now
32:39it was so valuable that the guy that
32:41made the most money as an advertising
32:42agency founder in the like 1920s and 30s
32:46he it was written by this guy named
32:47Hopkins who was this best copyriter he's
32:49like these are all the ideas how I'm
32:50making this money I'm putting this
32:51manuscript in a safe for 20 years right
32:54but essentially it's not about
32:55advertising it's about human nature
32:56which doesn't change and his whole point
32:58was just like your ads are about you
32:59your ads about your product your ads are
33:01like hey I have this thing buy people
33:03don't care about that they're
33:04self-interested and if you just all you
33:05have to do is just do the work of what
33:07this does if for you and you just lead
33:10with that and there's people that took
33:12that ideas those ideas were written in
33:14the early 1900s and went on to build
33:15massive companies advertising agency
33:17foundations in the ' 50s ' 60s7s and
33:19even to this day
33:20>> yeah 100%. I think there are a few
33:23insights. Let's say first is most people
33:25in the world, they don't give a
33:28about like OpenAI or the research or you
33:31know these other companies or the
33:32research or even us or our research or
33:34whatever. They care about like what it
33:36can do for them. We had this gift of the
33:38mascot which was the little jolly guy. I
33:42remember the tweet where I where I
33:44realized I was like, "Oh, wow. This is
33:46mileage." because I was basically I was
33:48um I was driving into work uh and while
33:52I was on the way I like generated a a
33:56version of the guy holding a briefcase
33:58and I tweeted MFW I go to work on
34:01Monday. I remember thinking like this is
34:03like one of my best tweets in a while
34:04and then I kind of realized like you can
34:06put him into a lot of situations. One
34:08thing I like kind of realized as we were
34:11going through that process was like how
34:13do you create surprising moments that
34:16are very different from any other
34:18marketing that you've seen of like a
34:19product. I don't like really remember
34:21the last time someone like you know
34:23leaned into racy memes to like market a
34:27product. Um I don't actually know if
34:28it's been done before but I kind of like
34:30realize like oh wow this is like a thing
34:32that we can do with this little guy. And
34:34that's something that like is
34:34interesting. That's something that like
34:36drives intrigue to like the average
34:38person.
34:38>> Okay. But that might get them raise
34:40awareness, might even get them to
34:42download the app. But then how do you
34:43get them to stay engaged once the app is
34:46they the app is on their phone? There's
34:48a cycle that people who who really love
34:52the app and and we even like we back in
34:55February had which is you try to give it
34:58a little thing and it does it. You're
35:00like, "Oh, wow. It did that." And then
35:02you try to give it a little bit
35:03something like a little bit more and
35:05then it does it and you're like, "Oh,
35:07wow. It can do that. It could do that
35:09first thing. It could do the second
35:10thing." And then you give it like a
35:12little bit bigger of a of a problem. And
35:14it's kind of this like trust fall almost
35:17that you have with an AI agent, which is
35:20at first you like don't really trust it,
35:22but you're like kind of intrigued. So
35:24you're going to give it a little bit
35:25nibble and then you give it like
35:27something a little bigger and something
35:28a little bigger and something a little
35:29bigger. Certainly like when I talk to
35:30like college students or whatever like
35:32some of the ways that people use AI it's
35:34like a full trust fault. They just like
35:36are recording voice notes and brain
35:38dumping to the AI and they send it and
35:40then they just trust the AI to like
35:41organize all the thoughts and give them
35:43like clear plans and think clear things
35:45to do etc. I think for the successful
35:48users of Muse it really is like getting
35:50them onto this trust fall process of
35:53using it for slightly bigger and bigger
35:55and bigger things and then having it
35:57work every one of those times.
35:59I found one of my all-time favorite
36:01quotes when I was reading the book 0ero
36:02to1. The quote says, "The single most
36:04powerful pattern I have noticed is that
36:06successful people find value in
36:08unexpected places." And they do this by
36:11thinking about business from first
36:13principles instead of formulas. That is
36:15exactly what Apploven has done with
36:17their advertising platform. Apploven
36:19connects you with over a billion
36:20potential new customers in mobile games.
36:22Apploven allows you to capture undivided
36:24attention. Apploving ads are full screen
36:27videos that are watched for an average
36:29of 35 seconds. That is retention that
36:31blows other ad platforms out of the
36:34water. And you can launch on Apploven in
36:36minutes. You set the goal and Apploven
36:38achieves it. No complex setup, no
36:40expertise needed. And Apploven scales
36:43quickly. They can put your ads in front
36:45of over a billion potential customers.
36:47Other businesses have seen immediate
36:49results scale to hundreds of thousands
36:51of dollars of spend per day and increase
36:53their revenue by millions. So, you want
36:55to get started quickly before all of
36:57your competitors are on AppLoven. And
36:59you can do that by going to
37:00apploven.com.
37:02That's apploven.com.
37:04I wonder how much of this is actually
37:06driven by word of mouth. So, obviously
37:09talking to you, I was going to have to
37:11download the app no matter what. But
37:12what really piqued my interest is a
37:14friend of mine was sending me
37:15screenshots of how he was using Muse.
37:18So, like when you download the app, it's
37:20it doesn't really tell you like too much
37:21of like what you can do. It was kind of
37:22like open-ended like a you know like a
37:24open text box but a friend of mine who's
37:27running like a hundred billion dollar
37:28company he left his ID on a flexjet and
37:32then he went through and he's just like
37:33well I want to he tell me I don't want
37:35to be involved in this process at all
37:37you have to figure out how to get the ID
37:39that like go pick up the ID this is
37:40where it's at and then this is where I'm
37:41at and then figure out all the steps
37:43along the way and then like arranged the
37:44courier went there got to his security
37:46in his office somehow and got upstairs
37:48to his office like and it was like holy
37:51this is the best marketing. It's
37:52just three or four screenshots of Muse
37:54going back and forth and him saying, "I
37:56don't want to be involved. You have to
37:57do everything." And it figuring out on
37:59the way. Yeah. Know, this is actually
38:00one of the things about um uh I don't
38:04know how it was like somewhat
38:05intentional, but definitely like modern
38:08product marketing is like very
38:10screenshot driven. Like I think to have
38:12a modern consumer product like it has to
38:14work on the screenshot. like the
38:16screenshot has to has to fly so to speak
38:20either through like group chats or word
38:22of mouth or like online. And I think
38:24that's been a big part of it. This
38:25wasn't intentional or engineered, but I
38:27think the fact that we had the little
38:29guy in the screenshot is like was a big
38:33deal because I think having Jolly or
38:36whoever your muse is like in the
38:38screenshot immediately communicates this
38:40is a Muse screenshot. It's just like the
38:42best marketing you could possibly have.
38:43So, are you on your Muse like all day
38:45long? like what is your own personal
38:46usage?
38:47>> I mostly like set up a lot of workflows
38:49using it and then I um I guess my
38:54meetings a lot of the time but um so in
38:57practice like I think in my job
38:59quoteunquote I'm supposed to like pay
39:00attention and and be pretty present and
39:02interact with people. I set up a bunch
39:04of workflows and like kind of treated as
39:06like a second brain.
39:08>> Yeah. You said that in the post that you
39:10wrote and it was like really short and I
39:12think I saw last time it was like 5 and
39:13a half million views so far but you said
39:15uh I love how you describe Muse like
39:16well if everyone had a second mind
39:18beyond their own. I think this is
39:21philosophically speaking one of the most
39:24interesting questions about like what
39:26does it mean to be human when you have
39:28like powerful AI and what is that
39:31relationship like and and how exactly
39:33does that look and different people have
39:35different takes on what exactly that
39:37looks like. I think the sort of like
39:38fear cases obviously that you know AI is
39:41above us and that we're just sort of
39:42like doing what they do and there's like
39:45other worlds where that relationship is
39:46different. And before you go on, what is
39:48your own personal view though?
39:49>> I really believe in muse. I believe in
39:50the personal agent as like a long-term
39:53form factor of just it is human nature
39:55to have wants and desires and dreams.
39:58And I think like AI and personal agents
40:02like muse will be the mechanism and the
40:05bridge that enables us to accomplish
40:07those things and continue accomplishing
40:08them. And then like humans will dream
40:10bigger and bigger and bigger and bigger.
40:12Kids will dream of like having their own
40:13star systems or whatever. And like Muse
40:15will help them do that. Like there's
40:16this thing that like um Bezos uh wrote
40:21in like well I think it was one of his
40:23shareholder letters which is like the
40:25beautiful thing about uh consumers is
40:27that they're like always beautifully
40:29unhappy or something like that or like
40:31beautifully unsatisfied with the options
40:33they have. That is like one modality of
40:35our relationship with AI. I think that
40:37there's other modalities like I think
40:39that AI will start doing more and more
40:41of the scientific discovery process. I
40:44think AIs will start inserting more and
40:46more into various parts of the economy,
40:48but I do believe in this mod in this
40:51like form factor long term of everyone
40:54having an AI that supports them and what
40:57they want. So you see AI as kind of like
40:58removing all like the cludge of life,
41:01right? The stuff that we don't want to
41:02do. I think in the the piece you said
41:04something like the world is full of like
41:05gatekeepers and obstacles and like Muse
41:07can get around this for you without
41:09spending any more of your time so you
41:10can actually focus on the stuff you want
41:11to do. I think Muse can
41:15give everyone the adulthood they dreamed
41:19of in childhood. Everyone when they're
41:21kids, they have like big ideas and big
41:24dreams and when they like imagine their
41:27life and they sort of like, you know,
41:29play it out. They they imagine this life
41:31where they can like be an astronaut or
41:33they can save the planet or they can
41:34change the world or, you know, whatever
41:36it might be. It's like a big they they
41:38think big and then for, you know, a
41:41variety of reasons by the time people
41:44finish school, they enter the workforce,
41:46by the time they get to their, you know,
41:48have been in a job for a while, it's
41:51that like all that hope and ambition and
41:54and ability to dream has been sucked out
41:56of them.
41:56>> No, it's been beaten out of them.
41:58[laughter]
41:59The life is beaten beaten out of them. I
42:01think as adults, I don't think we like
42:03just like appreciate the degree to which
42:05we're all zombies. We lost our agency
42:08and we lost our I think ambition is
42:10really a good word for it. One of the
42:12promises of AI broadly speaking and I
42:16think through products like Muse is to
42:19um keep it going from when you're a kid.
42:20Like you're a kid, you have big dreams,
42:22use AI, help make those things happen.
42:24Then like dream bigger, make those
42:26things happen, dream bigger, make those
42:27things happen. and kind of like
42:29experience this like escalator of
42:31agency, an increase in agency throughout
42:34your life versus a sort of like you have
42:37lots of agency and then it gets like
42:38crushed. You know what I was thinking of
42:40when you were speaking just now? You you
42:42mentioned earlier this this experience
42:43that you and Nat Freeman went through
42:45where you had this like disturbing uh
42:47psychological audit and like coming in
42:50waves from this AI. It's like yeah, what
42:52if we had that but it's like the
42:53opposite of what you're saying. It's
42:53like actually gave you more
42:54self-confidence like gave you
42:56understanding like the world is
42:57malleable and if you push on it hard
42:59enough you go after it with enough
43:00energy and drive you can actually change
43:01it around you.
43:02>> Yeah. I think this is something that
43:04like certainly the most impressive
43:06entrepreneurs exhibit which is like I
43:08think they like dream big they
43:10accomplish that then they dream bigger
43:12and then maybe they work on that for
43:14like a decade but then like you know if
43:16they accomplish that then they dream
43:17bigger. Like Elon is obviously a great
43:19example of this. Um I think Mark is a
43:21great example of this. I think but I
43:22think a lot of like the iconic
43:24entrepreneurs this is like like kind of
43:26like their lived experience and I think
43:30there's a version of that that should be
43:31true for every person and I think what's
43:34kind of like tragic is I think for most
43:38people you have like bigger dreams big
43:40ideas you enter the corporate workforce
43:44you're you kind of become a zombie maybe
43:46at some point you want to go tackle your
43:49dreams but then it's like really really
43:51hard cuz you all these commitments. You
43:52maybe you have a family, maybe you have
43:54whatever it is. It's just like it's
43:56almost you become trapped in the sort of
43:58like zombiehood. There's one more thing
44:01on miuse before I want to get to scale
44:03AI and like the the partnership that you
44:06do that you did with Meta and making
44:08that decision. But in this essay that
44:10you wrote or this like short postage you
44:11wrote, I love what you said. said, "The
44:12world until now has been shaped by the
44:14small number of fanatics who have
44:15somehow found a way to make their wants
44:17real, but we've never seen humanity with
44:19every single person's agency fully
44:21switched on." I think the promise of
44:24this world that I'm describing where
44:27every single person has this like agency
44:30increase throughout their lifetime and
44:31they have the ability to accomplish
44:34their wants and their dreams. I think
44:36that world looks crazy in a very good
44:39way. Like I think it will be very
44:41interesting. It'll be artistic and cool
44:44and you'll like go into different
44:46pockets of the world and it will be very
44:47like diverse. It'll be kind of insane to
44:50think about what exactly that looks like
44:52where
44:54literally billions of people have, you
44:56know, all of a sudden because of
44:57abundant intelligence like the resources
44:59to make incredible things happen. I
45:02think that's part of the promise. Like I
45:04think in my head it's almost like the
45:06Rick and Morty interdimensional cable
45:08but but just like somehow manifested
45:11into like reality for humans. Like it
45:13could just be really awesome. So I keep
45:16hearing uh this line from that show
45:19where it's like the universe eats smart
45:20people. [laughter]
45:22>> Um okay so I want to go back. What I'm
45:24personally curious uh about is you're
45:27you founded Scale AI, you're running
45:29Scale AI, and then one day Mark
45:31Zuckerberg reaches out and says, "Hey,"
45:32I assume he says, "Hey, I want to talk.
45:34Can you tell me about this?"
45:36>> Yeah, I think the exact message was
45:38something like, "Hey, do you have time
45:39for a call?"
45:40>> Did you have a relationship with Mark
45:41previously? Did you spend any time with
45:42him? Like what was what's the background
45:44there? I had a friend who uh this guy
45:47Alex Schultz uh who I've known actually
45:49for many many years um from when scale
45:52was maybe a year old. I met him in San
45:55Francisco is currently the chief data
45:56officer at Meta but was longtime Meta
46:00executive lieutenant and he introduced
46:02me to Mark
46:04I want to say in about like I think it
46:06was 2021 or 2022 to talk about AI
46:10because I think at that time Mark was
46:11getting a lot more interested in AI
46:13scale. We were doing a lot of stuff in
46:14AI. My memory is that I think it took
46:16like 1 year to schedule that first
46:19meeting. Like from from the intro to the
46:22meeting being scheduled, I think it was
46:23like a full year. Why? Mark obviously
46:25has like an insane calendar and like has
46:27like a bajillion different things to
46:28deal with. And he's also very good at
46:31prioritization. Like he will spend a lot
46:34of time on the things that are really
46:37really important in some ways like too
46:39much time on the things that are like
46:40really really important. the only way
46:42he's able to do that and like be a good
46:44dad and be a good partner and all this
46:45other stuff is to just like ruthlessly
46:47prioritize. I had spoken to him a few
46:50times kind of after that and had gotten
46:52his advice as a founder to get advice
46:54from you know Mark Zuckerberg obviously
46:56like a really big deal. Um and I
46:58actually remember at that time like he
46:59he wrote these like like I would ask him
47:03a question on WhatsApp and then he would
47:05write these like really long responses
47:07and I was really confused. I was like,
47:09how does he have time to write these
47:11like really long responses? But now I
47:12realize he like is just like really fast
47:16and rigorous at writing to like a lot of
47:18WhatsApp messages. He'll write like
47:20write a very rigorous like long response
47:22very quickly. It's like a very
47:24impressive skill of his.
47:24>> Gives you a sneak peek in what's going
47:25on his mind.
47:26>> Yeah. Like yeah, there's racing. Meta
47:29was a important customer of scale. Um,
47:32and I would say like prior to this
47:34instance, maybe we spoke once every like
47:37six months or so. Maybe that was like
47:39the cadence of our interactions. And
47:41then yeah, I think he he had called and
47:44I think it was it was actually like not
47:46obvious at first what the like end point
47:50was going to be because the first
47:53conversation was just like, you know,
47:55this was after Llama 4 and it was kind
47:57of just like asking what do you think we
47:59should be doing? Why is that an
48:01important point that this is this
48:02conversation is happening after Llama 4?
48:04>> Llama 4 was not on the trajectory that
48:07Meta wanted as a company. It definitely
48:09was like I think internally speaking a
48:11disappointment and it was also at a time
48:12where AI was clearly becoming just like
48:16increasingly increasingly important. It
48:17was clear that AI was going to be like
48:19really critical to the future of meta.
48:20So yeah the first few conversations we
48:22spoke on the phone and he was sort of as
48:24saying you know asking for advice on
48:26like what do you think we should be
48:28doing? What do you think we should be
48:29focused on? This is what I've heard
48:30about him privately that he has this
48:32insane it's not a board of directors
48:34like a meta board although there's some
48:36people on the board that he also does
48:37this with but he's got this group of
48:39world-class entrepreneurs around him and
48:41I've spoken to some of them like you
48:42don't understand like he asked for
48:43advice constantly he's like hitting us
48:45up he's like here's what's going on in
48:46my life or the work would you do and
48:49he'll do that over and over and over
48:50again I think this is like quite an
48:52impressive trait yes because it takes
48:54humility obviously like continue doing
48:56that even you know after all the
48:58incredible things that, you know, he's
49:00been able to accomplish. I mean, the
49:01whole, you know, timeline from when he
49:04first reached out to when we announced
49:08the deal was like five, six weeks. Like,
49:10it was like pretty quick. Another thing
49:12that world-class entrepreneurs have in
49:14common. We had Jonathan Ross, the
49:15founder of Grock, on this podcast. The
49:17first time we ever spoke uh publicly
49:19about the $20 billion deal he did with
49:21Nvidia was on the show. And he's like,
49:23"From the time first call from Jensen to
49:25money's in my bank account was 3 weeks."
49:27>> Yeah. That's amazing. [laughter]
49:29Um, I just had lots of ideas on like
49:31what I'd be doing if I choose and like
49:34how I'd be thinking about Meta's AI
49:35strategy. One of the first things I said
49:37was what I said earlier, which is like,
49:38hey, I actually think because Meta is
49:41like heritage wise focus on the things
49:43that people want to do is actually one
49:47of the most specially placed companies
49:49with this incredible shift AI. I like
49:52had lots and lots of ideas that I sort
49:54of sent his way and then that was kind
49:57of this like conversation happening in
49:59parallel with this like oh yeah maybe we
50:01should potentially consider if there's a
50:03way to work together.
50:04>> So when he says that what do you think
50:06cuz this kid you this was not
50:08predictable to you.
50:09>> Not predictable. No. The whole thing was
50:11like it was like a really crazy um
50:13sequence at the time. I was just like,
50:15"Oh, that's nice, but like there's no
50:17way that
50:18>> Well, you know what that means when they
50:19say that, right? They want to buy you.
50:22It's always the same thing, but they say
50:23it's like, "Oh, we got to find a way to
50:24work together."
50:25>> Yeah. I don't think I'd had that that
50:27kind of coaching at the [laughter]
50:29at the time. One thing that I explicitly
50:31felt at the time was like, "Yeah, like
50:34it probably doesn't make that much sense
50:36even." Like I think I personally was
50:37like, "Yeah, like you know, maybe he's
50:40like, you know, teasing it or or like
50:43throwing it out there, but like does he
50:44even like make that much sense?"
50:45>> Why wouldn't you think it makes sense?
50:47>> I mean, obviously like a deal ended up
50:49happening, so you know, what do I know?
50:50But um
50:51>> No, no, but back then, not we know what
50:52happened, but like I'm very curious your
50:54thinking as you're going experiencing
50:56this like the fact that you said, "Oh,
50:57yeah, you said this was like that
50:59doesn't make sense." Like why wouldn't
51:00it make sense?
51:01>> Taking a step back. Obviously the the
51:03like iconic meta acquisitions have been
51:05Instagram and WhatsApp and these were
51:07like very
51:11clear you know product logic driven
51:14decisions. Scale is like you know a deep
51:19enterprise government sales like you
51:22know a entirely different kind of
51:23business from from Meta. And so the the
51:27logic that I had thought at the time
51:29would have made sense is if meta wanted
51:32to get into those things that then I
51:35think you know like you know maybe
51:36that's that's the industrial logic that
51:38would have made sense but this wasn't
51:39that cuz I think Mark clearly was like
51:42predominantly interested in getting the
51:45llama program and over his overall AI
51:47program on
51:48>> he wanted talent.
51:48>> Yeah he wanted talent. It was kind of
51:50shrouded and unclear the whole time, you
51:52know, a lot of the way.
51:53>> When did it become clear? After the
51:55money hit the account [laughter]
51:57when, you know, when we started talking
51:59numbers, I was like, "Oh, wow. Okay. All
52:01right." Um, [laughter]
52:04that's when things became like more
52:06clear.
52:06>> At first, you're like, "Oh, this this
52:08will make sense." Then you guys keep
52:09talking. And how fast did he convince
52:11you like this is a good path for you?
52:13>> There was a part of it was just like,
52:14"Wow, this is like a really fascinating
52:18deal construct." We ultimately landed on
52:20this deal where you know Meta invested
52:25uh owns 49% of scale scale continues.
52:28Myself and a few people join join Meta.
52:31What I was kind of incredulous about at
52:33the time was like it's such a weird kind
52:35of deal but is one that I think
52:39genuinely kind of was like this very
52:43interesting winwin win so to speak.
52:46There's a win for the shareholders of
52:47scale because all shareholders of scale
52:50got a great deal, benefited a lot. Scale
52:52continues and I truly believe the best
52:55days of scale are ahead of it. My
52:57question to you is like or what I'm
52:58personally interested in is like okay is
53:00this he can explain this to you and you
53:04come around to that perspective in one
53:06phone call. Is this just like wait Mark
53:08I got to think this is so crazy.
53:10I got to think about this for a few
53:11days. Like explain this process as much
53:13as you can. many weeks of being like,
53:16"This is insane. Is this even real? Does
53:18this make any sense? If it does make
53:21sense, like how do I feel about it?" The
53:24predominant emotion was more like
53:27this is kind of insane. And then as I
53:28thought more about like, oh, I actually
53:30have to make a decision of whether or
53:31not I do this. That was it in and of
53:33itself. There were like a lot of like
53:35conversations there with the people at
53:38scale, with our investors. Like there
53:40were a lot of interesting conversations
53:41in in that uh on that side. I mean it's
53:44obviously like really hard to to like
53:47let go of or give up your baby
53:49>> cuz you worked on scale for how long?
53:50How many years?
53:51>> Yeah. 9 years from founding until the
53:53>> And you're still young. You're not even
53:5430 yet. So this is like a third of your
53:56life.
53:57>> There was something that Paul Graham
53:58said for a long time which is like you
54:00know if you think of your company as
54:02your life's work like you will operate
54:04differently. I did really genuinely
54:06think of of scales my life's work for
54:09the whole time that I was I was working
54:11on it. That was a tough emotional
54:13process. And then I think what
54:15ultimately got me was like a combination
54:16of wow this is like a win-win winwin
54:21like kind of all um it's it's a good
54:24deal for all parties and also just this
54:26like I saw the potential of what could
54:29happen at Meta. And you know, this is a
54:32point at which like I think meta on AI
54:36certainly
54:38looked like damaged goods in many ways
54:40and like was maybe not the most
54:42appealing place to work on AI, but I
54:46think that kind of like triggered my
54:49entrepreneur side where I was like
54:51>> you can essentially rebuild it. It's
54:52like a refounding of the lab.
54:54>> Yeah. Yeah. Like I think this is this is
54:57one of the things that ended up being
54:59kind of fascinating which is like
55:01because
55:03it was you know it was clear that there
55:05was so much work to do. It really like
55:07cleared the way for me to be able to to
55:11kind of build a lot of stuff up from
55:12scratch, set up the right principles,
55:14build the right culture. Like it it
55:16created a the conditions for us to build
55:18something amazing.
55:19>> And is that how you and Mark discuss it?
55:20It's like hey we have to hit a reset
55:22here. Like obviously like we're not if
55:25we keep on this path like we're
55:26essentially going to throw out what we
55:28have and redo it with a completely
55:30different uh like level of talent. Are
55:33these the conversations that are that
55:34occurring between you and him? It didn't
55:36seem like he wanted like a minor
55:37adjustment. You understand what I'm
55:39saying? He's just like, "Oh, this is not
55:41working at all. It's not going to work.
55:42Then let's, you know, rip it down to the
55:44foundation and rebuild." Correct. It was
55:46an evolving conversation where I think
55:48there were like like one of the things
55:49that I believe really strongly I think
55:52Elon is like has has demonstrated this
55:54in in various times in his career where
55:57you just have like a small incredibly
56:00cracked highly technical team that's
56:01like very flat and like if you do that
56:03you can accomplish a lot very very
56:05quickly. So that was something that I
56:08had a lot of conviction in and like I
56:10believed in AI like that is the right
56:12approach especially if you had to you
56:14know do what we had to do which is like
56:16move very very quickly and I think early
56:19on that was one of the things I talked
56:21to Mark a lot about and I think he was
56:24excited about as well that ended up
56:26becoming one of the tent poles of our
56:28overall strategies like small very flat
56:31highly technical team very high talent
56:33density how quickly can you move if you
56:35have those ingredients
56:36From an outside perspective, it just
56:38seems like Mark like empties the clip is
56:40the way I think about this, right? Where
56:42um have you ever read this book called
56:42The Mind of Napoleon?
56:44>> No.
56:44>> Okay. So, mind and Napoleon published in
56:451957. Um I found it because I saw this
56:48interview. I think Tyler Cowan
56:49interviewed Sam Alman. And uh I think
56:51this is 2018 and he asked him like
56:53what's the most important book that you
56:54read this year? And he said the mind of
56:55Napoleon. So it's very hard to find. I
56:57buy the book. It's 300 pages of just
56:58Napoleon his own words organized by
57:00topic. And in that he talks about over
57:02and over again that hesitation is fatal.
57:03But he says before you engage in a
57:05course of action, there's a lot of
57:07deliberation. He'll like study from
57:09every angle, make sure he's making the
57:11right decision. I feel there's like a
57:12echo with the way Mark operates. Does a
57:15lot of deliberation, make sure he's on
57:17the right path. But once he makes that
57:18decision, he just goes all in. He won't
57:20like he just empties the clip. And
57:21that's the way I felt about like what
57:23was was occurring right at this point
57:25where he comes and gets you and then you
57:27guys start rebuilding this entire
57:28organization. I think he really has
57:30internalized that failure doesn't
57:33matter. It matters like when you win,
57:35how big you win. And I think this
57:38overall concept I think enables him to
57:40operate I think truly without fear in a
57:42lot of circumstances because I think he
57:44is just very comfortable taking lots of
57:46risk.
57:46>> He's got that great line where it's like
57:48in a world that's changing all the time,
57:50the biggest risk is not taking any.
57:51>> Yeah, exactly. I think he's internalized
57:53this quite deeply. Like there's
57:55certainly like versions of the world
57:56where Meta looks a lot more like Google
58:00in some sense like it's just like you
58:01know more slowm moving and more
58:03bureaucratic and just like looks very
58:06different as a company but I think he
58:08because he's internalized some of these
58:09lessons so deeply keeps the thing
58:11dynamic and moving and live. talk me
58:14through the difference between how like
58:17how you were as like you're you're the
58:18founder, you're running scale, now
58:20you're refounding this lab, working with
58:24Mark, like the different ways that you
58:26approach your work in those two uh
58:28environments at scale. One of the things
58:30like I really
58:33internalized and I don't even remember
58:34where this advice came from, but it was
58:36kind of one of these ideas that like if
58:38you want to be a good founder, you have
58:40to be able to do every job. had a very
58:42strong point of view about how like
58:43every little thing at the company should
58:45happen and should work. And that
58:47resulted in a company that was like very
58:49much so a product of my effort and like
58:53my point of view and it like expressed
58:55itself in all sorts of ways throughout
58:56the company but also made me the
58:58bottleneck in very big ways. There's
59:01like pros and cons of that for a
59:03company. I think definitely I think
59:05that's like you know if you were to
59:06start a company and it's your first time
59:07starting a company you should like
59:08definitely take that approach because
59:09it's like much more likely to be
59:10successful than if you don't take that.
59:12>> Scale your first company.
59:13>> Scale is my first company.
59:15>> Yeah. But at Meta like it was a very
59:17different assignment because we had to
59:20accomplish so much in so little time.
59:23Like I didn't have the luxury of being
59:25able to quote unquote do every job
59:27within the lab. I
59:30uh certainly am not an AI researcher.
59:32I'm also not a like product designer.
59:35I'm not you know I'm not all these
59:37things that are like totally pivotal to
59:40um the organization being successful. So
59:43at Meta very much so I adopted this like
59:48philosophy around just how do we create
59:52an environment where we're going to hire
59:54incredible people like it was like job
59:57number one hire brilliant and incredible
1:00:01people like who are at the top of their
1:00:02field across every discipline.
1:00:04>> Why do you think you're able to do that
1:00:06at Meta? Is it just because you have
1:00:07more resources? Like why can you do
1:00:09that? I think one of the things that was
1:00:13charis very charismatic and attractive
1:00:15about the founding moment of MSL was
1:00:20there was an opportunity to build a lab
1:00:23from scratch with a really really small
1:00:26team and we had a strong point of view
1:00:29of where we're going personal super
1:00:30intelligence that I think resonated with
1:00:33people like I think everyone who works
1:00:34on AI wants to help people like wants it
1:00:37to ultimately like mean something to
1:00:38their mom or to their grandma or to
1:00:40their grandpa wants it to like be
1:00:42something that's like meaningful to
1:00:44every single person in the world because
1:00:46of the fact that we're doing this with
1:00:48the resources of meta. There was an
1:00:50opportunity to we weren't going to be
1:00:52bottlenecked by comput. We weren't going
1:00:53to be bottlenecked by like
1:00:55infrastructure. We weren't going to be
1:00:56bottlenecked by by distribution. like we
1:00:57weren't going to be bottlenecked by the
1:00:58things that like startups often are
1:01:00bottlenecked, but it was an opportunity
1:01:01to like really build something that you
1:01:03could like put your stamp on and have
1:01:06like a lot of influence over and really
1:01:08shape. I think that was like quite
1:01:10exciting for a lot of people.
1:01:11>> And so that's why he chooses to spend an
1:01:13insane amount of money, what looks like
1:01:14an insane amount of money, getting a
1:01:16handful of top talent because then he
1:01:18knows that that those top talent will
1:01:21recruit in turn other top talent. But
1:01:23you need that as a starting point.
1:01:25Correct.
1:01:26>> Yeah. Yeah. And I think one thing that
1:01:27like is maybe somewhat underappreciated
1:01:29about that specific part is that like I
1:01:33mean the stock prices of OpenAI and
1:01:36Anthropic like ran up a lot. So it just
1:01:40happened to be the case that like you
1:01:42know if you looked at how much these
1:01:44people were making if they stayed at
1:01:47Open Eye or Enthropic like it also was
1:01:50like it was a lot of money. So I think
1:01:52for a lot of the talent it ended up
1:01:53being like kind of net neutral in many
1:01:55ways. um compwise, but obviously I think
1:01:58it's kind of like the same thing like
1:01:59oftentimes we don't we don't a lot of us
1:02:01don't think about how much like the
1:02:02early employees or the early researchers
1:02:05at these like labs that have had immense
1:02:07[snorts] stock uh stock value
1:02:09appreciation you know what their comp
1:02:11looks like we started this like fully on
1:02:14the belief of having and hiring like
1:02:17most brilliant people that's one Slar
1:02:20personal super intelligence um those two
1:02:22how do we create the environment where
1:02:24people do their best work at Meta and in
1:02:28MSL. I'm doing much more of like quote
1:02:32unquote traditional management, so to
1:02:33speak. Cuz these are the things I'm like
1:02:35thinking a lot about is like how do I
1:02:36create the environment where all these
1:02:38brilliant people who are exceptional and
1:02:41very very special can all operate and
1:02:45achieve the greatest expression of their
1:02:47talent. I'm thinking a lot more like a
1:02:49almost like a coach in some in some ways
1:02:52versus at scale it was like very
1:02:54different like I was the aur I was like
1:02:56the person whose like point of view had
1:02:58to be expressed in everything that we
1:02:59did.
1:02:59>> Say more about that you're you feel like
1:03:01a coach. I think that like my job is to
1:03:05a like help set the like north star and
1:03:09the like directionality of where we want
1:03:11to go, which again personal super
1:03:14intelligence, winning on consumer AI,
1:03:16like building AI that means something to
1:03:19every person and then spotting in the
1:03:22team special talent and then giving
1:03:26people the environment and resources to
1:03:28be able to express that maximally. Like
1:03:30I think one thing that big companies
1:03:32often mess up is that there's like
1:03:35exceptional people at most big companies
1:03:37but they just like are not empowered to
1:03:39do their best work and that's why they
1:03:41often leave to go to smaller companies
1:03:42or to startups or whatever it might be.
1:03:44You mentioned Jeff Bezos's shareholder
1:03:46letters earlier. That's something he
1:03:47would repeat as well and all the books I
1:03:49read about him but also shareholder
1:03:50letters like if you have great people
1:03:51that can't build they're going to leave.
1:03:52>> Yeah, exactly. And AI is such like a
1:03:55it's like such a special thing. Like I
1:03:58think you know this term research now it
1:04:00it feels like it almost doesn't mean
1:04:02anything but like by definition humanity
1:04:05doesn't know the limits of what these
1:04:09like models are capable of and what are
1:04:11the things you can do with them. Like
1:04:12they really are these like unknown
1:04:15artifacts that we don't understand super
1:04:17world. So like it really is science and
1:04:20research. Like we are exploring the
1:04:22limits of what you can do with this
1:04:24technology and what are the things that
1:04:25can be built and it is a process that
1:04:28requires lots of trial and error. It
1:04:31requires like brilliant people to have
1:04:33incredible insight. Like it's one of
1:04:34these things that you can't think about
1:04:37this process like a very opishly intense
1:04:39process. Like people need space and time
1:04:43and resources to be able to like you
1:04:45know do their best work. And I think
1:04:47about in many dimensions, right? Like I
1:04:49think Nat Freiedman is absolutely
1:04:53brilliant and as I mentioned like so
1:04:56much of Muse if there were a single
1:04:58person whose like single point of view
1:05:00came through on Muse it is now it's
1:05:02important for him to be in a position
1:05:04where where he can like fully express
1:05:06all of that. And then we have some
1:05:07brilliant researchers. They have like
1:05:10very very exciting ideas. How do we
1:05:11create an environment where they are
1:05:12able to explore those? And like most of
1:05:15them won't work out. Some of them will
1:05:16work out incredibly well. The cases
1:05:18where that works out justify all the
1:05:20investment. So there's two kinds of
1:05:23problems so to speak and that analogy is
1:05:26like diamond mining or building
1:05:28skyscrapers and it depends on what the
1:05:30distribution of the payoff of the things
1:05:32that you're working on are. So there's
1:05:34like some areas where the distribution
1:05:37is super power. Like most of the things
1:05:38you do will be kind of useless, but then
1:05:40some of the things you do can be like
1:05:42literally a million times more valuable
1:05:44than everything else. VC I think is a
1:05:46great like diamond mining kind of
1:05:47industry to give a sense. It's just all
1:05:49about like being able to identify the
1:05:52special ideas and like see those through
1:05:54>> and that's easy to do. Just look at just
1:05:56go to your 19th birthday party.
1:05:58>> Yeah. I mean that's like unbelievable.
1:06:01Um
1:06:01>> do people listening that don't know what
1:06:02the hell I just referenced? Can you talk
1:06:04about real quick what who was at your
1:06:0619th birthday party and why that was
1:06:07would have been a good bet if they just
1:06:08invested everybody around the table?
1:06:11>> This is like a really crazy thing to
1:06:12think about. my 19th birthday party. I
1:06:14was a freshman at MIT. It was in
1:06:16January. I was doing a win internship, a
1:06:18winter internship, win internship at a
1:06:21trading firm called HRT. There were like
1:06:2410 of us, 12 of us, um, something like
1:06:26that. Almost all of whom were students
1:06:28at either Harvard or MIT. Most of us
1:06:30already knew each other from various
1:06:32math and science or computer science
1:06:35competitions. Some of the other people
1:06:36were Jeffrey Yan who went on to start
1:06:40Hyperlquid very successful. Um Scott Woo
1:06:44went on to start Cognition very
1:06:46successful. Another person there was uh
1:06:49uh Jesse Zang went to start Decagon also
1:06:51quite successful. Um another person was
1:06:54Vicky Yei who is I think a brilliant
1:06:57researcher works at Anthropic. Everyone
1:06:59was like incredibly talented. I actually
1:07:01talked to the uh the founder of HRT
1:07:04recently and HRT itself by the way now
1:07:07is like crushing it. Unbelievable.
1:07:09>> What does HRT do?
1:07:11>> HRT it's a proprietary trading firm. So
1:07:13they they trade their own money in Q2 of
1:07:172020
1:07:18um 6 had 11 billion of revenue and like
1:07:23I think 8 billion of profit is the the
1:07:25record number. This is a I think it's
1:07:28under a thousand people work at HRT. I
1:07:30mean, it's like unbelievable. But I I
1:07:32recently uh met up with the founder of
1:07:34HRT and they were talking about like,
1:07:35"Oh yeah, we're starting to do some
1:07:36private investments." I was like, "Just
1:07:38invest into the interns. All [laughter]
1:07:41the interns get seed checks."
1:07:45That's great.
1:07:46>> I grew up doing uh math and uh science
1:07:50and computer science competitions and
1:07:52Olympiads and got to know like all sorts
1:07:55of people from all around America doing
1:07:57this kind of stuff. It's like quite
1:07:59surreal to see
1:08:01the people that I did competitions with
1:08:03become so successful.
1:08:04>> It's going to be fascinating what all of
1:08:05you do the next few decades. Before I
1:08:07interrupted you, did you have more to
1:08:08say about diamond mining and skyscraper
1:08:10stuff? Basically, it's like does it have
1:08:12a power law payoff where some things
1:08:14just like pay for like literally like a
1:08:17million times the investment or is it
1:08:19something more like building skyscrapers
1:08:21which I would I would describe data
1:08:23annotation or or or data businesses as
1:08:26more like this which is like Amazon
1:08:28Prime deliveries is another example
1:08:30where the payoff is pretty linear and
1:08:32you have to just get develop a process
1:08:33where you're like really really amazing
1:08:35at doing it. If you're building
1:08:37skyscrapers, then it's like a very
1:08:38operational business and it's all about
1:08:40like squeezing every ounce of efficiency
1:08:43out of that process and you just look at
1:08:45it and you figure out like I can make it
1:08:47like 2% more efficient this way and 1%
1:08:49more efficient that way and half a
1:08:51percent more efficient that way and you
1:08:53just try to squeeze every ounce of
1:08:55optimization out of it. And then there's
1:08:58diamond mining where you're kind of
1:08:59like, "Hey everyone, just like you know,
1:09:03go forth, try all your crazy ideas and
1:09:06like you're just nurturing the ideas
1:09:08that seem promising."
1:09:09>> That's what you feel you're doing at
1:09:10Meta right now.
1:09:10>> That's what I feel I'm doing at Meta.
1:09:12And and and the building server stuff
1:09:14generally speaking is what I felt I was
1:09:15doing at at scale. And that's like this
1:09:18big paradigm um paradigm difference.
1:09:21Going back to the coaching and the way
1:09:22the difference of how you're essentially
1:09:24managing now at meta compared to way
1:09:26scale. Why do you have so many direct
1:09:26reports? You have like what 100 direct
1:09:28reports? How many direct reports?
1:09:29>> Yeah, I think more than 200. Yeah. H
1:09:31>> how and why?
1:09:33>> Going back to it like one of the one of
1:09:35the founding thesis of MSL and and and
1:09:39what we want to do and like rebuilding
1:09:41this this lab was how do you have like
1:09:44really really high talent density? How
1:09:46do you make it like the best place for a
1:09:49lot of these brilliant people to do
1:09:51their best work and how do you build an
1:09:54organization that's like technically
1:09:56focused where the technical people are
1:09:58doing their best work, you know, have
1:09:59have the opportunity to really do their
1:10:01life's work. And one of the design
1:10:04principles around this was making it
1:10:05very flat. you know, effectively all the
1:10:07researchers that we hired into this
1:10:09group called TBD, they report directly
1:10:11to me. And it's more of a statement of
1:10:13the fact that like this is like
1:10:15anti-bureaucracy. Like TBD is like
1:10:17literally anti-bureaucracy. The point is
1:10:20we're hiring brilliant people and you
1:10:23all have like demonstrated clearly that
1:10:25you can do brilliant, brilliant work.
1:10:28You have like done it before in your
1:10:29careers. So I don't need to be
1:10:31micromanaging you to do brilliant work.
1:10:34you're going to be able to do that on
1:10:36your own. We need technical leadership.
1:10:38So, some of you are going to be
1:10:39responsible for like setting clear
1:10:41technical directions for your groups and
1:10:43we have this kind of like pod structure
1:10:44where we have various pods and there's
1:10:47various technical leads of the pods who
1:10:48who tie break on the technical
1:10:50decisions. So, there's like a technical
1:10:52leadership structure in place, but from
1:10:54a quote unquote people management
1:10:56perspective or from a bureaucracy
1:10:58perspective, we're just
1:10:59anti-bureaucracy. And then the other
1:11:02part of this is like we do make a lot of
1:11:04group decisions. Like we want to debate
1:11:07because it is like a very talent dense
1:11:09group of like brilliant people. We want
1:11:11to have those conversations where we're
1:11:13all discussing what do we think the
1:11:15right path is. Everyone [snorts] in the
1:11:17lab has done brilliant things. Like we
1:11:19want to hear all these opinions. We want
1:11:20to arrive at things that we all believe
1:11:22are the best path forward. So it's
1:11:24certainly like an unconventional way to
1:11:26run the team. And like I would be lying
1:11:28if I said I was like the best manager to
1:11:30more than 200 um people. But this part
1:11:33of the point is that like they don't
1:11:36need managers, they need a great
1:11:38environment. They are all brilliant.
1:11:40They're all extremely capable. They just
1:11:42need the room to cook.
1:11:44>> Yeah. And so maybe not a coach. You're
1:11:45kind of like the steward of a great
1:11:46environment so they can do great work.
1:11:48>> Yeah. Exactly.
1:11:50>> Alex, this was awesome, man. Thanks for
1:11:51making taking the time.
1:11:52>> Yeah.
1:11:54>> I hope you enjoyed this episode. Please
1:11:55remember to subscribe wherever you're
1:11:57listening and leave a review. And make
1:11:58sure you listen to my other podcast,
1:12:00Founders. For almost a decade, I've
1:12:02obsessively read over 400 biographies of
1:12:04history's greatest entrepreneurs,
1:12:06searching for ideas that you can use in
1:12:08your work. Most of the guests you hear
1:12:10on this show first found me through
1:12:12founders.

About this transcript

This page contains the full transcript of Building Muse, Working with Mark Zuckerberg & The Future of AI | Alexandr Wang, Meta by David Senra, generated from the public captions YouTube serves with the video. The transcript has 14,806 words across 2,085 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.