Startup Building: Challenges & Opportunities

Ben Horowitz
I think, like in the Boeing case, I think the big problem was the CEO was an accountant. Like, I mean, nothing against accountants. Like, what do they know about building airplanes? I mean, this is like, I think it's important that the person running the organization understands the core thing that the company does. You would never cut it as a Harvard business school professor with an attitude like that.

Yeah, exactly. They're just making widgets. 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 investor or potential investors in any a 16 z fund. Please note that a 16 z and its affiliates may maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a 16 z.com disclosures.

Marc Andreessen
Welcome back, everybody. We are here for round two of building startups and companies in the post COVID world. Last time we got through five out of 25 of the excellent questions that people submitted. So we'll see if we can get through at least five more this time. Let's dive right into the deep end, which is startups.

And so we talked a lot about the sort of general post COVID world last time. Now we're going to start talking specifically about startups. Lars Schmidt asks, how should companies think differently about organizational design and talent deployment as they now have more ways to address core business needs? And then he puts in parentheses, full time employees, freelancers, gig workers, outsourcing, and now the rise of AI. And I also put this again in the post COVID world, where companies also have now a lot more geographic flexibility about where to put different operations.

So I interpret the question is there's a whole bunch of new options or previous options that are becoming more viable now. And how should startups, let's start with startups. How should startups think differently about sort of configuring how they think about their employee business? Yeah, so I'd say it's important to kind of think about first sort of the old reasons why you would have somebody kind of in the company versus outsourced. And it really like, probably the biggest reason is knowledge.

Ben Horowitz
So if you think about a job like data entry, where not much, it doesn't matter the knowledge of the last data you entered to do the next data, then that's a really good outsourced opportunity because you don't have to retain the knowledge so you can, you know, kind of buy the service as you need it as opposed to kind of have it permanently. The opposite of that would be something like a software architect who designed your whole product and then that person leaves, then, you know, that's going to be a very big setback because you're going to have to reconstruct whatever they knew. And this kind of gets into it that historically companies have been pretty bad at kind of retaining knowledge systematically. So, like, there's what's in your people's heads and then there's what's in your systems, and the kind of former is much bigger than the latter in terms of just a total body of knowledge. And look, it's very hard to, you know, have a great company if you can't learn from the past, if you don't kind of move forward.

You know, if you think about it, we had a really interesting conversation with Adam D'Angelo years ago. I don't know if you remember Mark, where he was saying like, one of the keys to Facebook at the times growth team was that they had the exact, not just that they had a growth team, but they had the exact same people on it for years and years and years. The reason was the work they were doing was fairly alchemistic in nature and they try a few things, see if that worked, try a different thing. And if you didn't remember all those experiments, then you were kind of, you just be doomed to repeat failed experiments. And that's kind of like, I think the key to why you'd want somebody kind of as an employee.

The other reason why is just the agency problem. So, oh, should we have an outsourced attorney who has just as much love for the company on the other side of the transaction as they do for us, or they're optimizing for something different than the long term good as a company. That would be the other reason to have somebody inside. So then how does that change now? Well, a big change is AI.

So you go a couple of things. One, can you outsource it to AI? Like, can AI actually do the job, but also can AI just remember what you did and then can you query that knowledge base in a better way than you could say Salesforce? So you have these old systems that we call them often write only databases because information goes in but it never comes out. Or with AI, does that equation change?

And now you have some jobs where, well, previously you would have had to definitely retain the people, but now if you just plug them into the system, even if the human is doing the work, the system remembers what they did and somebody new can come in and quickly kind of figure out whether what they're doing makes sense or not. So that would be probably a big change in terms of international. I think that's with kind of remote work being more effective that, you know, we've seen the huge pickup in kind of people just deploying people internationally for not just cost reasons, but also just, that's where the talent's available. So it used to be, I mean, when we started, you wouldn't even have an engineering group in another city, right? Like, you have to have everybody in the same city.

And, you know, as engineering software tools changed and it was possible to kind of spread them out more. You could have distributed built systems and other kinds of things. But, you know, now, you know, even more so you can have, you know, people in different countries being very, very effective, being kind of plugged into the company in a way they weren't before. And, you know, there may be more talent in Poland that's available to you. Thank, yeah, talent in San Francisco.

Marc Andreessen
Yeah, that's right. How much in the sort of post COVID in office versus hybrid versus remote kind of scenarios, how much difference do you think companies can tolerate in terms of some group of employees is required to be in the office five days a week, some group three days a week, some group once every two weeks? Some groups are people fully remote. The natural human impulse towards fairness, like does, at some point do people start to get like really frustrated and upset if they have colleagues who are more remote or completely remote? And as a CEO, like, how do you, how do you navigate, how do you navigate explaining to people that, you know, people are gonna be treated differently depending on what group they're in or when they were hired?

Ben Horowitz
Yeah, so I think that, I don't think that like internal, you know, maybe it's because we work mostly with startups and we're not that big ourselves. I think the internal equity thing is probably less of an issue than if you're doing new things you really need. It's very hard to do that all remote, I find, and then talking to lots and lots of CEO's. So it's very easy. You have a plan now let's go execute it.

And we're all executing to the plan and we're not really planning to change the plan much. That can be done more effectively, probably remotely, than putting everybody in the same office. Cause you're not, you know, whatever, talking about random stuff, you know, shooting the shit, getting snacks, playing foosball, all that kind of thing, you're actually kind of working against the task. On the other hand, if you're like, okay, we need to build a brand new product to solve, like some customer need that we've identified, and we've got to figure out how to design the thing, what the kind of storyboard flow is, how to architect it, who goes on what task and so forth. That's really hard to do remotely because those are very, very long conversations.

Sometimes you want them to take hours. I think most people can't pay attention on a zoom for more than, you know, 45 minutes or so, tops, whereas in a meeting, you can, and you can have a pretty intense discussion. So I think, I think that's the main thing that drives, whether you want people together or you're fine with them being remote. And some teams can do all their planning in a short spurt, so they can do a three day offsite and do all their planning and all their new ideas, and then they just run with those ideas. So I have a different point of view, different than a lot of people would say in terms of how creative work can happen.

Marc Andreessen
And so I would argue there's a very extreme barbell, which is you either need for truly creative work, like defining a new product, building something new. I would argue it's a full barbell, which is you either need everybody right there in person for the reason that you articulated. But I think there's another side of the barbell, which is doing it entirely 100% online, and in particular over chat over some sort of text based system. And the reason is because open source, most of the big open source projects got built with people working remote. And almost like in our day, it would have been overusednet.

And then more recently, it's over various kinds of messaging systems, and then crypto projects be done that way. Bitcoin, as an example, was done that way. It was all over mailing lists. They didn't even know she was. Yeah, exactly right.

And then we see this with some of the new generation software companies where they run like this and they're just on chat all the time. And of course, obviously, we run a lot of in person also, but we have very active slack, and there's a real feeling that everybody's kind of in the same virtual space all the time because we're all on slack. And so I wonder if there's a polar opposite where you could, you could, where you have to push it all the way to the other side of the barbell, which is you have to build intentionally to be like 100% in an online collaboration space with no expectation of in person. And maybe it's everything in the middle that just gets all fuzzed up. Yeah.

Ben Horowitz
Yeah, I definitely think that could be. And it's interesting. It's like, I think it has to do a little bit also with the team dynamic and like, what is the nature of the team leader and how they work and so forth? And I see that this a lot with CEO's where like some CEO's are like, we don't do anything now that we're remote. Like, I've got to get people back together.

I don't know how to do it. Nothing happens. And then other CEO's are like, no, it's working great. And I think it has more to do with the CEO personality and whether they can't feel the work if they're not there. And so it seems like, yeah, we're getting nothing done.

And then something like, something always goes wrong in the business. And then you go, well, the business is all messed up because we're remote. Or, or like, things are going great because we're remote. That's right. Cargo cult idea, other factors.

Marc Andreessen
So we'll start talking about big companies, actually. So Andy Aarons asks, how can these build new lessons, which sort of what we've been talking about be applied to mature companies in order to rebuild, refresh them without having to start from scratch? And I want to dig into this a bunch, but let me start with this question, actually, I've been dying to ask you. So when you were an executive at HP, as I recall, tell me this is right for a short time, but I believe. I believe you were there at the time.

I think the many. This was way pre COVID. That was way pre COVID, pre remote. This is in the old days, I think it was the case that many of the senior executives lived in basically random locations and didn't actually live near their office. Is that right?

Ben Horowitz
Yeah, that is right. Yeah. I bring it up because more recently. They got more extreme than that. What do you mean?

Marc Andreessen
How so? Well, during a period when I was there, they wanted to save money. Like, they had this very interesting cost cutting idea. It was actually a good lesson for me in incentives that if you incentivize, if you pay the CEO based on earnings per share and particularly quarterly earnings per share, which I think we did at HPA, or at least annually, if you optimize for that, you can do things that would undermine long term productivity. And one of the things that I remember them doing was they wanted to save money on chairs and desks.

Ben Horowitz
And so we actually had more employees than we had chairs and desks, which I guess either forced remote work or just force people not to go to work, depending on how you looked at it. Yeah. And it was. And I was on the board, you know, I was on the board, so I had a perspective on it also, you know, so, yeah, it was basically, the logic was overall real estate footprint. Yeah.

Marc Andreessen
Right. And so just total number of square footage, which. Which translated into total number of cubicles. Yeah. Which did ultimately translate total number.

It's literally total number of chairs. We did calculate, I think, at one point we were. We had 50,000 fewer office chairs than we had employees. Yeah. I miss of Zoom or remote work or anything.

Ben Horowitz
Right. Like, it was. It was kind of interesting. Exactly. And so there was actually all the engineers.

You'll have higher earnings per share in the short term. Yeah, well, they did in those days. They did a lot of other things. They turned off the. One of the famous things that happened at HP in that era was they turned off the satellite music subscription for the company.

Marc Andreessen
Jim. And so the music when he worked out was 24 hours. Christmas carols. Yeah. Motivating to get your workout over with.

But, yeah, they had this thing specifically, I recall, where the executives lived in various locations, and then they had an off site culture. Right. For the execs, where they would fly into various places. I think you famously at one point had to go to Thailand. Yeah, I went to Thailand.

Ben Horowitz
I went to Jamaica. Many places in my short time there, I was all over the world. Yeah, exactly. But that was kind of where the executives had come together, was at these offsites. And then I also bring it up, because this has been on the press coverage, and I don't know if this is true, but in the press coverage on Boeing, all the scrutiny that Boeing's under, there are very similar stories being told about them.

Marc Andreessen
And maybe it's not true, but the story being told is the executives don't actually all live near headquarters. They're sort of dispersed. And so one of the things people are wondering is, can you have a high functioning airplane company where the executives literally are not together? And I just bring this up because this is a natural consequence. I think more and more companies operate in this way because in theory, if everybody's available on video all the time, does it really matter where people live?

And can you run a big company with having the 20 top executives in 20 locations? Do you think that can be made to work? Or is that just flat out a bad idea? Look, I think these things are more malleable around the nature of the business and the composition of the team. And so I do think that, like, if the team's been around a while and they've been working together and it's very cohesive, and the leader is just, like, very good at getting people together on an as needed basis and so forth, then I could definitely imagine it working.

Ben Horowitz
On the other hand, like, if you have a brand new team with a new leader and so forth, like, setting the culture, the expectation, and so forth, it's just a lot more complicated when people are remote. I think, like, in the Boeing case, I think the big problem was the CEO was an accountant. Like, I mean, nothing against accountants, but, like, what do they know about building airplanes? I mean, this is, like, I think it's important that the person running the organization understands the core thing that the company does. You would never cut it as a Harvard business school professor with an attitude like that.

Yeah, exactly. They're just faking widgets.

Yeah. Building an airplane seemed just, like, awfully complicated to me. And so if you didn't really understand that at a very detailed level, like, it's hard to imagine you kind of seeing when things start to go adrift or sensing, okay, like, yeah, this isn't right. If you're just going, what were the earnings per share? Not the stretching.

It's not actually the business, despite what you may learn in business school. Yes, exactly. So when we talk to. Let me back up. During COVID there were a lot of discussions in public, and we talked about this a lot, which was like, wow, these big companies, every big company CEO had this massive realization kind of all at the same time, which was like, oh, my God, we can just move everybody online, move everybody to zoom, and the company keeps operating.

Marc Andreessen
It was actually quite remarkable in that era that basically every big company and actually every functional system in society that was an operation just kept running. The stock exchanges kept running, and the banks kept running. I would have predicted a much higher level of chaos, but basically just stuff kept running. Maybe companies weren't innovating as much or whatever, or maybe people were less productive, but fundamentally, everything kept running. I remember a lot of conversations with big company CEO's during that period, which was like, wow, now that I know that, I can now think about doing permanent re engineering for my company.

So even after COVID, it's like, okay, if I used to have 100,000 people in Manhattan, maybe now I can have 20,000 people in Manhattan and have the other 80,000 people in North Carolina or North Dakota or the Philippines or somewhere else. So there's sort of many conversations along these lines. How do you think that's played out in practice? Have big companies been doing that, or did they just kind of, did they just kind of go back to the way things, you know, were before as COVID sort of came to an end? You know, I can tell you our companies have moved a lot of jobs, and it's been surprising to me because, and particularly, like, a lot of engineering jobs have gotten moved as a result of COVID And, you know, you know, it's funny.

Ben Horowitz
There's still, like, many, many companies in San Francisco, just as many, but there are definitely fewer kind of engineers in that both a lot of engineers have moved out, and then a lot of the engineers that those companies hired have sprung up all over the country and in other countries. So I definitely think that's happened. And it depends what companies are kind of like people in that if they have to do something, they'll do it. And if they don't have to, they won't. And so the companies, I think a lot of it happened after 2021 when, you know, like, money got tight and people were like, okay, how do I do it?

And I know I can move stuff now, so I will. And it's also like, it's a way to do a reduction in force in a way that's not as or doesn't feel as draconian as you go. Well, like, your job's still here. It's just move to Singapore. So if you'd like to live in Singapore, we'd be happy to have you.

And then, you know, all of a sudden, you have fewer employees as well. So I think all that kind of comes into play. But for sure, I think companies think about how they can change much differently than they did pre COVID yeah, as. Sort of Ben pointed out, there's this interesting phenomenon for most big companies. Not all we could talk about the exceptions, but for most big companies, there's this pattern that sort of goes with the broader economic cycle, which is during an expansionary boom period where stock prices are rising and the economy is doing well, companies basically just grow, and they just add they add projects and they add people and they add resources, and it's because their revenue is going up and they can afford to, and it's great, and everybody's having a good time, and it seems like the whole thing is working.

Marc Andreessen
And then there's this corresponding period where when the stock market takes a big hit or the economy enters recession, or there's just generally something horrible happens like September 11 of the global financial crisis or COVID. Then it's this basically oh shit moment where people are like, oh God, the business is going to contract, or at least the growth rate is going to slow way down. We have to get more efficient and I'm not going to hit my EPs targets. And so I have to cost cut. And so basically it's sort of this weird paradox, which is to your point, companies basically only do the cost cutting rationalization, efficiency work under duress.

And the duress is usually caused by something macro. They often wish that they had used the boom times to do that work because it's a lot, because in theory at least, if during boom times you. Have a lot more cash. Yeah, exactly. So they often get under sort of extreme stress during the down cycles.

So there's a picture you can paint that basically says that basically productivity improvement in the economy, like sort of people doing things more efficiently in the economy is a function of down cycles, not up cycles. Right, right. Okay, yeah, go ahead. Look, there's always a reason you can rationalize why you're not going to get more productive in any kind of boom cycle. You know, you can go, well, like, yeah, we would be a lot more productive.

Ben Horowitz
And, yeah, like, you know, probably 20% of the people that we have aren't doing anything, but like, it would mess up the culture to do it now. And we have plenty of money and everybody's making money and that's all good. And so, I mean, I think it's a lot, the nature of human organizations. I remember I had a conversation with Andy Grove years ago where I asked him some question about, like, well, you know, why don't people change, like, the way they do things like that? And he's like, or he had an insight, like your insight.

And I said, well, you should write a book on that. He said, why would I do that? And I said, well, what do you mean? Because it's a, you know, you're right. And he said, yeah, but like, it's human nature.

People can read a book that goes against human nature. They're just not going to do anything I like. He's probably right. Yeah, that's right. Yeah.

Marc Andreessen
And so what I observed, you know, by the way, I should also highlight, not all companies are like this. And what I learned so at HP when I was on the board, you know, we worked with this activist named Ralph Whitworth, who has unfortunately since passed away, but was sort of a legend in the activist community and actually joined our board and had a big had a big stake in the company. And what he taught me during that period is that it goes, Ben, I bring it up because it goes to your earnings per share thing. What he taught me, basically is earnings per share. Actually, the earnings per share is a metric, or just more generally, a lot of managers are sort of incented by size of empire, and it's sort of like just aggregate revenue growth or even aggregate spend, just like the number of heads, the empire building, the number of heads that people have under them.

But even earnings per share, he would say, is a bad way to motivate for the reason that you described, which is it's sort of just like, okay, that's some percentage of revenue, but that takes you back to basically having revenue be your target, empire building expansion being your thing. And he said, basically, the way to get companies to run in a mode where they're continuously trying to get more efficient, if that's what you want, is to shift compensation instead to the financial metric of return on invested capital. And so it's to basically measure every dollar coming into the business as sort of equity dollars and then measure the actual return on those dollars. And if there's something that you can do. And then basically what he said basically is, and he did this at HP, actually, and it was very successful, was that you basically, if you tie executive compensation to ROIc, you have this thing where all of a sudden people are like, wow, even though it's an upcycle, if I can go make it more efficient, I can boost my Roic, and I can get this, like, disproportionate bump in the stock price.

I can get the, and then I get, like, directly paid on that. And so he viewed that as the lever to be able to basically get people into this mode that normally they'll only get to during down cycles. But he said, you have to, you have to fundamentally change, you have to fundamentally change the incentive away from expansion. And then the question is, what's wrong with expansion? You said, what's wrong with expansion is as companies grow, they basically reach further and further into basically worse and worse lower margin businesses in order to continue to grow.

They're trying to empire build. They're trying to grow revenue. They're trying to grow earnings. One example, an example is they go into more and more countries that are just less profitable to operate in, or they have product line extensions that are just less profitable, or they're doing acquisitions that are just less profitable. And so even though earnings per share might be growing through that, from an economic standpoint, those companies are getting less efficient, return on investor capital is dropping, and then you basically, over time, become a giant inefficient mess, which is what a lot of big companies turn into anyway.

There is a way to run big companies. It goes to your Andy Grove thing. If Ralph had written a book, it would have been this is how all companies should run all the time. But that book would be equally unlikely to be read as Andy's book because it's the kind of thing that managers do not voluntarily take on themselves. It generally has to be enforced by an activist at some point.

Ben Horowitz
Right, right. It's really interesting. Well, not only does it need to be enforced by an activist, but there's no growth investor who looks at it that way. Right. So from when you're a little company, kind of before you run into that particular issue through when you go public, you're kind of very, very, very incentivized to grow.

Growth has a tremendous amount of value. And even if it is ultimately going to be bad, inefficient, low margin growth, you get rewarded for it in the short term. So that's a very interesting thing. Could you actually, earlier in the life of a company. Yeah.

Marc Andreessen
So what Ralph said on this, he said, look, every company started as a growth company. So there was a time when, like, all these companies were like fancy high tech startups of their era. Yep. And that's built into their DNA now. Exactly.

And then, you know, ten or 30 or 80 or 150 years later, they're still trying to get growth. Right. Because it's in, you know, human nature. Like, you know, as you said, you want to expand ge, right? They did that crazy, you know, probably fraudulent, insane ge finance thing that kept their earnings per share going, but ultimately, without a bailout from the US government, company would be gone.

Ben Horowitz
Right. Yeah, exactly. Exactly. So, I mean, I don't know how 100 years after Thomas Edison. Yeah, that's right.

Marc Andreessen
That's right. Yeah, that's right. And so, yeah, he said, he said sort of the cardinals, he said, look, he said the cardinal sin. Yeah. If you're a true growth company, you don't want to do this.

If you're a true growth company, if you can actually get growth like actual, like high velocity, high margin growth in a new market, that's fantastic, you should go do that. But there comes a time in the lifecycle of every company, he would say, where that has stopped. That's just over. You've done that. Your market is now mature.

You've done that. He said at that point, basically, executives will spend arbitrary amounts of time and effort and then company resources in order to try to chase, to re spark the growth, relight the growth. And then when they do that, they financially, economically take the company off the cliff. And then they wonder why investors are all mad and the stock price isn't going up. And he's like, basically, basically it's like you well intentioned idiots, you need to think more like an investor.

And as an investor, you think in terms of like, okay, I put in a dollar, what do I get out? And that's sort of the core internal economic efficiency, which is a much less fun thing to think about than just top line growth, but is actually the way to then get, actually have shareholder value in a sort of rationally run company where you're actually pursuing the right opportunities in the out years. Well, and it also causes you to re examine money you're putting into old things and whether actually generating any kind of return. Yeah, yeah. And like, you know, there's a lot of old, I mean, we see this.

I mean, this, we see this a lot in tech is just, you know, there are should name, I mean, you know, there are legacy tech products that are very widely used today that have very big businesses and they'll have, you know, five or ten or 15 or 20,000 engineers working on them. And the last new feature anybody cared about was a decade ago. And.

Ben Horowitz
Well, it's funny, right there, there was that Charles Wines companies was like the great geniuses buying those companies and then taking out all the expense. Yeah, exactly. And so anyway, the reason I went, the reason I went through all that is because it is kind of been a down cycle. I mean, it's been a weird economy for the last few years because the COVID kind of economic takeoff was super weird and then the COVID sort of, then everybody getting bummed after COVID ended was kind of weird. And then the inflation thing has been weird.

Marc Andreessen
And then you're in this mode right now where in theory the economy is doing well. But a lot of people, as we talked about last time, don't feel it because of inflation. So it is this weird thing, but it feels like I bring it up because it feels to me like there's going to be another, at some point there's going to be a severe recession, no idea when, but at some point there will be. And at some point there are big companies. If you're a big company CEO, there's going to be an opportunity at some point to really, I think, do radical restructuring with all of these new technologies and methods, and maybe at that point it will be half really trying to get to the bottom of exactly where the workforce should be located, and then maybe the other half will be how to incorporate AI.

But it feels like a lot of big companies have a fundamental restructuring ahead of them that hasn't happened yet. They do, but I'm still skeptical just because there are the new ways to do it, that they'll actually do it to that degree. It was interesting, and there's a big set of big company layoffs, and it was really interesting to see Elon reduce x by 80%, and then everybody else had numbers like eight or ten or 15%, and companies of the same generation with the same kind of growth with the same basically profile. And it's hard to imagine that the gap, the optimal gap of how many kind of people you could let go was really that large, maybe, but doesn't seem that way. A lot of companies, I had a lot of conversations, a lot of CEO's over the last five years, and there's all this kind of lore around layoffs.

I find, see what you think of this. There's all this lore and stories get told, and so it's like, okay, we're going to do a layoff. We need to cut costs, and we know that we over hired and we got sloppy shit all over the place. And so we're going to do a layoff. And, like, we, we don't want to do a 5% layoff because, like, it, layoffs are traumatic and so you don't want to, like, 5% is like, too, too low, but, like, boy, if it's double digits, you know, that's really going to damage morale.

That's going to feel really bad. And so we'll do 8% or they'll have that same conversation. They need to do more. They'll say, well, you know, we need to do more than that. But, like, you know, boy, at 15%, it's really going to hurt.

So we're going to do 13%. And so you get, you get this, you get this like, nibble around the edges thing where they're literally backing into almost this intangible idea of like some sort of minimax number where it's going to be enough to get credit but not enough where people get mad. Yeah. Do you find, do you find that to be the case? I have heard it.

I've experienced that a lot. Have you, have you seen that? You know, people definitely do that. I think so. There's, there's a couple of reality.

Ben Horowitz
Like thresholds and in a cut. So the first, like if you do any cut, then you've broken trust with the employees. So you've kind of created a cultural problem in the sense that, okay, I gave you a job, I told you this was going to be great, and then like I let you go. So I broke a promise. I broke trust.

Trust is very important. See, if you do any laugh, you have to rebuild trust. So that's problem one, the first kind of threshold of badness. There's a second threshold of badness where, okay, if you're letting go of kind of new mediocre people, that has one kind of effect. If you start letting go of either genius people or like people that are the most popular people in the company and so forth in large numbers, and that has a greater effect than just breaking trust.

Then I think at the highest level, when the company, which nobody ever cuts to this level, but the company just can't even function, is kind of the third level. And I think that what happens is companies don't know where that second level is and they often think it's way, way, way lower percentage than it really is. So like what's a 50% cut might get you into, they think happens at 12%. And I have seen that a lot. Do you find, by the way, a related thing on that is, I forget, is it price's law?

Marc Andreessen
There's one of these laws named after a person, but I think the law is the square root of the number of employees generate 50% of the output.

Ben Horowitz
That seems right. I mean, look, some of it gets back to knowledge, right? So particularly in a startup, what often happens is the original people, particularly in the product areas, have amazing knowledge because they've been there the whole time. They know every product decision that was made, they know every customer that got pissed off, they know every deal that was lost and so forth. And so they're very, very productive because they can make really sharp decisions very fast and then they know how the company works.

And then people who join that, we had this problem at Netscape in a huge way. People join out, 800 people, 1000 people don't know how the company works, don't have good judgment and so forth. So even if they're good people, they have a hard time being productive. And so you kind of get to this feeling where the same people, no matter how many people we hire, the same people are doing all the work like that. That is a real thing.

And so this is, I would say you can do better than that if you've got really great onboarding, really great training, and then high performance standards. If you do all those things, like, I like Ali's company, databricks. They're really, really good at that. I think Zuck did an amazing job in the early days, for sure, at meta on that, where their onboarding program was so good that they were able to just keep turning out massive numbers of very high quality products for a really long time. More how it is now, I won't ask you to comment on it whether it's good or bad, but there are companies who I would say are better than the square root of.

What was the exact opposite of the square root of the number of problems? Yes, I pulled it up. So it's a guy named Derek Desola Price, who was actually a physicist who worked on these issues. Specific law is the square root of the number of contributors generates approximately 50% of the output. And if I just spend a moment on that, square root is an interesting thing because the square root of 100 is ten.

Marc Andreessen
And so an organization, 110 people are producing half the output. But in an organization of 10,000, the square root is 100. And so what's so great about the square root here is it really gets to, at least if you take the law at face value, it really gets to this thing, which is like, wow. Even when you're at massive scale, it's still a very small number of people producing half the output. And by the way, we talk here for a moment about half the output.

I think. Tell me if you'd agree with this. I think you can see this in engineering, where you have some small number of engineers who are doing most of the great work. Much of the great work. You see it in product management, where it's the same group of people, same small group of people doing all the breakthrough work.

You see it also in sales, where you generally like enterprise sales, you generally have a few reps that have these. Sales in enterprise, enterprise distribution on kind of performance against quota. The bigger the deals, the bigger, the wider the distribution. So when it actually distribution, when it takes a lot of skill. Right, right.

Ben Horowitz
If you're just. Yeah, so you'll have it. So for people who don't know this, you'll have a small, if you have enterprise sales, if you're selling stuff to businesses, you'll have some small. You'll have a large number of people who just, like, sell stuff to businesses. And you'll then have a small number of, like these hyper sort of competent people, sales reps.

Marc Andreessen
And you may pay them a million dollars a year or more. And they have these giant accounts that are extremely complicated and extremely large. And it's just, they're just like Jedi masters at, like, dealing with that. And they're, you just, you, you know, you always wish you had more of them, but you have a few of them. And then when you tally up all the sales at the end of the year, it's like, oh, yeah.

You know that. You know that that group of, whatever, ten people generated half the revenue and the other thousand people generated the other. Like, that's a fairly routine thing. And then, by the way, Ben, we can also say the same thing is true. The same thing is true, actually, in two other areas of organizations.

One is by country. Yeah. Just the nature of the distribution of countries is that a small number of countries are going to generate most of your revenue. And then you also see it of product lines, which is most companies have one or two products that generate most of the revenue. I mean, Google is an example.

Google search is still some giant percentage of the revenue. And if you basically like. And by the way, and then Google search plus YouTube is like, most of the revenue. Right. And then they've got 100 other products that are a tiny.

So anyway, so it's this power law curve thing that you just see over and over again. And the reason that my price is law is so provocative is because. Well, and connect it back to the ROIC thing that we're talking about earlier is it's just like, okay, what are you running the company for? And at the end of the day, there is this question of like, okay, why do we have the other? Why do we have the other?

Ben Horowitz
It's really interesting. So I have two related comments. So on the one hand, I would say, I do think with really good management, it might go from the square root of the number of employees to the square root times like three or four of the number of employees or something like that, where you can get kind of quite a bit better. But we had a really interesting kind of meeting with Jensen from Nvidia, where he was just pointing out that the reason Nvidia is still so productive is because it is really small. If you look at their headcount against their market cap, it's a remarkably small company.

Marc Andreessen
And just. You mean just in terms of raw headcount? Yeah, raw headcount for like, you know, somebody who makes complicated chips and stuff. So it is interesting. I think that's.

Ben Horowitz
I think that's mostly correct. I do think. I mean, there's a lot of things that you can do, like while Helix Packard, the original helicopter, not like pre computers, Hewlett cockroach, I think, did a remarkable job of scaling the company through design without hitting that law as hard, for example, you know, by, you know, and they basically organized into, you know, very independently run businesses that weren't tied together at the hip in the way. Although other things are. Yeah, although.

Marc Andreessen
Although, you know why that ended. You know why they. The specific reason that ended was priceless law again, which was the computer industry. Ended up business by like a factor of like 50. Exactly right.

And every other business all of a sudden, and other businesses that have been perfectly good businesses at smaller scale all of a sudden were just running errors. Yep. Yep. Now that. That is.

Ben Horowitz
Yep. The parallel wins in the end. Although, like a brand, what, 40 years or 30 years that way. Yeah, it's a long time. Yeah, for a long time.

Marc Andreessen
Yeah, for a long time. Okay, good. That actually takes us to another really provocative question. Ashraf asks, do you think it's finally possible for a one person, billion dollar startup? And I would say, and again, this is in this sort of post, COVID, you could see, you could put a whole bunch of things in this bucket.

It's the inherent scalability of software and the Internet. But you could also say AI. Then you could also just say contracting, outsourcing. Did you have one person with a billion dollar startup? And I'll just volunteer and I'll say, you already do.

It already exists. It's bitcoin. And so there's, there's experiments. Yeah. Bitcoin says a trillion dollar outcome with, you know, one guy and one guy with a bunch of people kind of.

Ben Horowitz
Weighing in and helping, but King Skims, Kim. Okay, go ahead, Anthony. Right, so explain or to explain that. Yes. Yeah, go ahead.

Well, I mean, I, you know, skims kind of a person, a personality endorsement, and then, you know, nothing else in that company, I don't think has to be permanent. I mean, I don't know much about skims, but I imagine, like, she could just hire some top designers and outsource the manufacturer, put it online, and, you know, there, there she is. And I think it's worthwhile. A billion dollars or well over a billion dollars. Like, there you go.

Marc Andreessen
Yeah, that's right. Maybe Dolly can design the next set of clothes. Bitcoin is a better example, though, because trillion dollars, that's good. Ethereum is pretty close. The Ethereum core team.

It wasn't one person, but it was a pretty small core team. Yeah. And I do think the invention, or the original invention was largely Vitalik, as I understand it. And then he had certainly great people around him. And then other nominations here, obviously, is Instagram was a billion dollar sale with eleven people, and then WhatsApp was a 15 plus billion dollar sale with 50 people.

So they're again, and then probably the most provocative one is telegram, which is a shoestring operation. And by the way, for people who haven't followed telegram, it's this very interesting story. And actually, Pavel Durov, who's the two brothers, the Durov brothers, who did Vikantakhte, which was the social networking company in Russia that the russian government seized actually at one point, and then they telegram after that. They're very, they don't talk, but they don't show up in public much. And Pablo Duroff, who's the CEO, just did a long form interview with Tucker Carlson, where he goes through in detail kind of how they run the thing.

And I think it's a billion users. And I don't know, I remember exactly the head count, but it's a tiny, tiny, tiny number of people.

So, Ben, would we expect, especially in the sort of new world of AI, like, how many of these, like, does this thing just, like, are these things one of flukes for the next 20 years? Or do you think we'll start to see a tempo of sort of super geniuses who will figure out how to harness AI and all of a sudden we'll see more of this pattern? Yeah, I think we'll see more of it. I mean, look, I think the Internet is actually probably in some ways a bigger breakthrough than AI on it. And that being able to distribute over the Internet versus the way companies were built prior, like, so much of the headcount goes into distribution.

Ben Horowitz
And so I think the distribution kind of cost for something like telegram was like, no AI, just the Internet. And they're able to do that. And then you add AI in. Although I'm getting a little more skeptical about the distance between, like, what we can do now with copilot to complete human elimination, just in terms of the technologies that we've seen, like that last, that last mile is a long mile, bro. It is not.

We're going to need a breakthrough to get to there. So we're not quite at the point where you go, okay, like, all my coders are AI's, or at least like, that doesn't exist yet, and we'll see. But, like, I think we're probably, we're further away than I thought we were. I'll just put it that way. So that would be your theory there is sort of similar to the observation people make to self driving cars.

Marc Andreessen
Right. Which is like. Yeah, there's like a tremendous. If it's, if it's remote work, fine, but the minute there's an edge case, you have a problem and the world is filled with edge cases. Yes.

And so that last ten or 20% actually takes, you know, a long time. Yeah. I had a really interesting conversation with Jan Stoicha the other day. Who? The professor at Stanford, and he was saying, you know, like, you kind of think, well, like, this isn't that hard a problem.

Ben Horowitz
You've got the, you know, you've got the software generation that comes out of the model, and then you can have the model. Right test cases and the model tests it. If it doesn't work, then. And he says, but it turns out that that's not actually true because, like, the test case will be wrong and the, you know, the results of the test will be wrong and then the fix will be wrong. And, and so, like, it just turns out to be harder than it looks, amazingly.

So we'll see. I mean, we'll see, we'll see. But I will not like you would have thought that, you know, like, in a way, like, why isn't it done now? And we're not, like, we haven't seen anything. That's.

And then the other thing is that a lot of the kind of examples are always on very, very well defined problems. Right. Most almost no interesting engineering problems are like, really well defined. So, like, build telegram. Okay, that is well defined now, but it wasn't well defined then and anyhow.

Yes. So very interesting. Yeah, that's right. I'm probably a little bit more optimistic just in the sense of, I think progress and reasoning is actually going to move pretty. The combination of just, it just turns out you can keep training these things and they get smarter.

Marc Andreessen
That was a key part of the llama news this week. Watch the interview with Zucker. He read through that. It was very, very interesting. They don't even know when it stops getting smart.

Yeah. For people who haven't been paying attention to this, there's been this theory for a long time that you basically take, you take input into training AI and you get the chips all set up and everything, and then you train it up to a point. You let the training process run for a certain amount of time, and then you hit a point of diminishing returns, and at some point you stop training and release it. At some point. You just can't make the model smarter by training it more and it's just turning out.

There's more and more cases now where people are realizing, no, actually, if you keep training it, and actually, if you keep training it to what people used to think were kind of absurd extents, like, if you train it to be like, train it for 100 times longer than the sort of predictions would have suggested. And once data. Right, which is the other, like, whoa. Yeah. And then there's also this thing increasingly where you train.

Yeah. There's a specific form of synthetic data, which is you run it, you sort of train it up, you train it to do this sort of thing called chain of thought reasoning. So you train it to solve problems across multiple steps, but then you actually retrain the model based on the result of that and then all of a sudden it can solve those problems in a single step. And so this is the sort of self improvement loop that people have been looking for for a long time. And apparently that method is really starting to work.

And so it may be that there's an iterative process here you can get where even if it can't do something out of the box, you basically hand hold it for the first time through and then you retrain it and all of a sudden it can do it. Right, right. That's interesting. So you're actually pushing the know how down into the model. So it's not going to be, maybe it won't be a layer above the model that kind of works with the model to get you all the way to the right answer, but it'll be part of the training set that makes the model smart enough that it can generalize.

Yeah, or both. Or literally both of those. You do that first part, first you help it through the multi step process, but then you retrain on the basis of that result. And then next time it doesn't need the handholding, which, by the way. By the way, it's, by the way, it's also kind of how people work.

Right? Like, you know, it's like, okay, I teach you to do a new task. You know, we're going to sit and do it together, you know, do this with my kid all the time. Right. Step by step by step.

And then, you know, he's like, oh, okay, now I get it. Now I go soft and he's able to do it himself. Like, that's not a, that's not a crazy concept. Yeah, yeah. That'll be interesting to see how far that goes.

Exactly. So we will see. Anyway, so, Ashraf, very good question. I think this is going to be a key. Oh, and then, yeah, the other question that Ben and I debate a lot.

It's probably premature, but just like, yeah, like how many jobs, you know, just like how many jobs in a startup? Sort of the same question, but how many jobs in a startup could you think about over time, automating and like, you know, my favorite example of this is everybody assumes, here's my provocative thing I bring up with people, everybody assumes that white collar knowledge work, call center work, respond to customer service requests. We're going to convert that to AI. But of course, you're still going to have a human manager overseeing the AI call center bots. The question I always ask is, well, are we sure it's not the other way around?

Ben Horowitz
Is it AI bot training the humans? Exactly. As a human call center rep, talking to people on the phone and having a human connection, maybe you actually work for the bottom. In other words, maybe it's actually easier to automate the manager job than it is the worker job. And look, it goes to this very interesting question of what do workers expect from their managers?

Marc Andreessen
And I think everybody has an idealized model of what a great manager is. Most managers and most businesses are not great. Most people have had experiences with mediocre or bad managers. And if you had an AI bot that was super sympathetic and super supportive and always happy to mentor you and teach you and always had, we actually. Have investment in a company, Cresta AI, that kind of takes that approach, which is what the AI does, is it trains the call center people, basically.

Ben Horowitz
It coaches them. Like literally step by step. How do you answer the call? What do you say when they say this, this and that and the other in real time, which is probably a better manager than any call center manager because they're like, they never get tired. They're very precise in the answers that they make and so forth.

So could be, yeah, yeah. They never get frustrated, right? They never get frustrated. You know, they're 24/7 you know, if you're, if you're trying to figure out something at 02:00 in the morning on the night shift, you know, they're happy to talk to you about it. They never get, you know, they never get drunk, they never get high, they never have issues at home.

And interestingly, they need the worker in the loop, the human worker in the loop, because of the edge cases, right? Like, who knows what the personality other than the phone is going to say. Could be something completely unanticipated. And it turns out that humans are very good adapting to that. Machines can go a little haywire.

The Microsoft one tay. Off the handle. From time to time.

Marc Andreessen
Well, yeah, tey became a Nazi. So that was the, you know, the. Other AI's, the modern AI solved that by just refusing to answer any questions. Exactly. Or turning the Nazis black.

So the more prosaic version of that recently was, I think this is actually, I think this actually was technology that predates. No, it must have been an LLM because of how it behaved. But I think it was one of the, one of the airlines. I won't, I won't name names. One of the airlines had a case where they had a customer service bottom provided by a third party.

And apparently there's this thing, or some airlines have this so called bereavement policy where you can get a ticket refund or whatever for flying to a, you know, flying to a funeral of your parents or something. And so somebody, there was an airline apparently didn't have that policy, but they had a customer service bot. And, you know, the person was like, you know, okay, I kept to go see my, you know, you know, this, that. And apparently the bot literally, like, made up a bereavement policy on the fly. And it's just like, oh, we, you know, we, you know, we're, you know, we're all about customer service and like, we really support you in your hour of need.

And, you know, of course we'd be happy to per our supportive bereavement policy, we'd be happy to do this, that, the other thing. And of course, why did it do that? It did that for two reasons. One is because it knew, presumably, about bereavement policies because it knew about all the information off the Internet about this topic. And then the bots, it's trying to be friendly.

It's trying to give you answers that you're gonna like. And then there was another case where a car dealership, it was a car dealers, I think it was a Chevy dealership that put up a customer service bot. And within a few days, it discovered the bot was trying to convince potential customers to buy Teslas. To be helpful. To be helpful, right.

To be helpful. It's like, well, you know, do you care about the environment? Oh, yeah, I care about the, oh, and then you should consider buying an electric car like a Tesla. You know, let me tell you about all the new great Tesla cars, right? And so anyway, like, to your point at edge cases, like, yeah, there's, there's, there's going to be, there's a fair amount of.

A fair amount of work to be done here and may. Right. And maybe the actual job of manager is actually the bounded job and then that you fan out to the people anyway. Right, right. The manager can train on all the known cases.

Yes. And then you're into deal with the edge case. This is going to be the great exploration of the next decade. Try to figure this stuff out. Okay.

Ben Horowitz
All right. Slightly different topic actually. Very interesting. So Nate asks to what extent the California's COVID policy. And here I think what he means by that is basically long lockdowns, lots of controls.

Marc Andreessen
For a long time, California was the earliest state to go into lockdown and have sort of very, let's say strong slash harsh, you know, lockdown policies. And then it was sort of first state to lock down and then it was one of the last states to open up. So to what extent did California's COVID policy contribute to the spread of startups to other states? Yeah, so, and I have a. Yeah, go ahead.

Ben Horowitz
Yeah, so I'd be interested in your view. So what I've observed is it's not been as much the spread of, there has been some spread of startups, other states. I think the more dramatic effect was the spread of engineers and employees to other states, which has been profound. And you can see it in the numbers in California where I think more people have left California than any other state in the last couple of years. But the blossoming of many startup hubs, we're not quite seeing it the way that they think people thought it would happen.

So New York has definitely a pretty vibrant startup environment. You know, the ones that were before, you know, like Seattle and LA are kind of where they were. People thought Austin and Miami would emerge. You know, Austin has kind of gotten, I'd say a little more robust and Miami not very robust so far, at least, you know, from our observation. So I don't think what people thought might happen, which is like it really spreads out, hasn't happened as much, but the people have really spread out.

I would say the workers are all over the place now, which by the way is I think are going to be an interesting challenge for California from a fiscal standpoint because that was a huge source of tax revenue and we've kind of not seen it for that reason, but for the reason that there haven't been a lot of ipos lately. California's got a small minor deficit problem. Yes, giant surplus to a giant deficit. Yeah, that was amazing. So by the way, New York is an even more dramatic case of what you're talking about like a very large percentage of the top taxpayers in New York, which is a lot of tax base, have moved to Florida because it finds finance and finance is a post attack.

Marc Andreessen
But in finance it's such an easy, go ahead. Yeah, yeah. The hedge funds all moved to Florida, right? Well, on the principle moved in the hedge fund case, which is different than in the tech case, I think. And there's several things that have catalyzed that.

That's been a dramatic story in New York. Several things catalyze that. One is that it's just easy to get to Florida. It turns out it's just, it's a short flight from New York to Florida, a couple hours. Number two, you know, tax that, you know, New York is a very high tax state.

Florida is very low tech state for individuals. Third is the weather. And a lot of New York finance people already had second homes in Florida anyway. We're going to retire there anyway, so. Might as well get a jump on it.

And then the other is just as we talked about last time, but the crime dysfunction stuff. And then by the way, the weather also sort of this big push and then finance, a lot of the like hedge fund jobs are relatively fungible and you have as much access to your Bloomberg terminal in Miami as you do in New York. So that's been a big shift and that has people in New York very concerned, both in New York City and New York state very concerned about the tax base, which I think is a legitimate concern. Yeah, I think somebody said 5000 families in New York are half the tax base. Or maybe it's more, it's like a crazy number.

Ben Horowitz
When I heard it, I was like, wow, it's kind of the dark side of the progressive taxes. If they move, it's not a lot of movement that causes the entire budget to go bananas. Yeah, that's right. So the extreme, I think that's right. In New York, the extreme, if I recall correctly.

Marc Andreessen
But the extreme version was California in the year 2000. In the year 2000, the top is to go back to the prices. A lot of thing in the year 2050% of tax revenues in California were from the top thousand taxpayers. California is a very big state. Like that's a dramatic statement.

Now if you're on the left at this point, you're howling because you're like, well, that's just because of income inequality and like how unfair society is. But if you're trying to like pay the bills of a state, those guys, yeah, if they're not in your state, like, you have a big problem. And, yeah, to your point, it is. This is the natural consequence of a progressive tax policy, which is to say a tax policy, the tax is higher earners higher than it does lower earners, you know, which generally is viewed in our society as a moral thing to do. It just has this, it just has this side effect, which could be very.

Ben Horowitz
Very pernicious if you take it to that extreme where if everybody, and I think it's like half of the country pays taxes or something like that, like, it's a smaller percentage than one might think. And that's great in a way, in that you don't want to tax people when they're kind of struggling to make ends meet, but it isn't robust. And that, you know, people move is the problem. Yeah, that's right. And then different states have different ways of gathering tax revenue.

Marc Andreessen
And so there are states that don't need. Don't have state income tax or, you know, have it very low because they have other ways of getting income. So. Yeah. Or they have other population distribution such that they just don't need as much money or whatever it is.

So. So, yeah. And then, by the way, there's another side of it. I don't know how common this is, but I've seen instances of it. Now, I'm curious, Ben, if you've run into this yet.

Have you seen a case yet? We go to a retail store and there's a customer service person on a screen instead of in person. So it's a live human being on a screen that you can talk to. Have you seen this? Yeah, no, I haven't seen it yet.

So apparently this is just starting. And so, like, for example, you could do, you know, you have all these, you know, we talked last time about all the legal changes that are leading to fast food restaurants using more and more robots and kiosks. Yeah. And so there are, I've seen cases now where you walk in and it's like it's an automated, you know, whatever. Everything's automated.

But there is a human being you can talk to. It's just the human being is not actually in the store. Oh, yeah. Are they like on a beam, a mobile thing, or are they. Yeah, no, they're just like on a big screen, just on a big zoom.

You know, it's just like basically their eight hour shift on a zoom screen. And they talk to whoever walks up to it, and they're super happy and, you know, they're sitting at home and they're having a great time and doing customer service for a restaurant or food outlet or whatever it is. Yay. So it's, it's. Yeah, yeah.

And so it's like, well, this is the sort of, you know, the full version of this is, would be sort of the idea that, you know, with sort of sufficiently advanced, ubiquitous video conferencing, all of a sudden immigration law becomes less relevant, right. Because, like, if, you know, it's perfectly legal to stream bits across a border, even if it's not possible for people to move across the border, right. And so, you know, if you have sufficiently ubiquitous screens everywhere, cameras everywhere, remote, you know, telepresence robots, all this stuff, you know, does it really, you know, at some point for more and more, like regular jobs? Does it matter? Well, another example is, you know, the working.

Ben Horowitz
Yeah, right, right. For, I mean, there's of course, some jobs like picking fruit that still, at least today we don't have good robots or anything for that. Well, that's what's interesting. Again, it's just like with the knowledge work jobs, or in this case, like customer service jobs are easier to either automate or they're easier to replace with a person on a screen, to your point, than something that is manual labor. And so the assumption in our field forever has been that we automate the low value tasks first, you know, just like cleaning toilets.

Marc Andreessen
And then later we automate graphic design or something. And it turns out it's like the other way around. Software engineering first. Exactly. Or management or whatever.

And so it turns out there's these intermediate kind of customer service jobs. Well, and another example of this is, I don't know this to be, you know, this to be a fact, but I think for, like, waymo, for like the way most self driving cars, I think there's a human supervisor who can take over. Right. And they, I think they, I think each supervisor is responsible for some set of the cars, you know, if they run into an edge case like a police, you know, stop point or something like that. But those people don't need to be in California.

They can be anywhere. Yeah, right. As long as they don't have to be in the car, they could be anywhere at all. Right? They can be anywhere.

And then it turns out if you just. Did you see Amazon? There was news, I think, this week, Amazon shut. They had these stores. They had, they had these retail stores.

They were doing for, like, groceries where it was, it was the just walk out. And so you, like, load up. You load up all your food and then you just, like, walk out. You don't even do the self checkout. You just walk out.

And it. Pavement is magic. And it turned out the key to the whole thing was a lot of people in India watching on cameras, which is just like, you know, it's just like the obvious way to do it. And, you know, there's somebody sitting in Bangalore or something who's just keeping a tally of, like, you know, Ben just put two gallons of milk in his cart, right?

And maybe that was the prototype for what, one day could have been a fully automated, you know, kind of thing or whatever. But anyway, the point being is, like, you know, maybe you have a lot more economic activity spreading out, not even just because of AI or even remote work, but also actual job redefinition. Yeah.

Exactly. Okay, so. Yep. Okay, good. And then this is a kind of a fun question.

This is a meme question, but I'll explain it. Bobby Lancaster asks, should startups implement a 90 minutes morning coffee ban to increase productivity?

This is the online fight I'm always having with my friend Andrew Huberman, which is. Andrew Heberman is always citing. You know, he's a Stanford professor and has studied all this health stuff, and he's always citing this evidence that basically says people are overly addicted to caffeine. And so if you. If you.

If you don't drink caffeine, if you don't drink caffeine in the first 90 minutes after you wake up, you don't get the fall offs later in the day when you go into, like, basically you have, you know, caffeine becomes less effective. So it's so sort of a way to make caffeine effective across the entire day. My response to that is, no, under no circumstances am I going to not have caffeine in the first 90 minutes. I don't care what the science says. Those 1st 90 minutes are important to me.

I'm trying to wake up. I'm trying to be in a good mood. I'm in caffeine withdrawal. Give me my cup of coffee and stop talking to me about science. And so that's the joke.

But the serious version of the question would be, this is actually a serious question, less a company question, but more of a general life question of what you're seeing is nicotine. So it's, of course, very socially acceptable in our culture for everybody to be drinking caffeine all day long. Nicotine went very badly out of style, obviously, with the kind of war on smoking. And then there's at least parts of society in which is becoming very much back in style through Zinn is like the or then there's this company, Lucy, that our friends formerly at Soyland are doing. And so there's these new non smoke caffeine or nicotine delivery systems.

And I said, I know two people in particular who are on nicotine basically all day, every day. And they both say it's just like, the best. They're just, like, completely in love with it. It may even make you smarter. So there's some, like, neuroplasticity nicotine pathway.

Ben Horowitz
So it may not only, like, not be bad for you and keep you completely alert, but it may make you, like, learn stuff better. And then there's historical. There's a fair amount of historical. I don't know if it's historical evidence, but there are historians that will basically say that, you know, there was a time when people generally were, like, drinking alcohol that, you know, back, you know, hundreds of years ago, it was actually fairly common for people to drink alcohol day, in part because. In part because life sucked and they wanted to be drunk, but.

Marc Andreessen
But in part because it was hard to get clean drinking water. And so if you drink alcohol, my. Favorite wc fields joke, he said, you know, I don't drink water because I don't drink anything that fish fucking.

Exactly, exactly, exactly. And so, you know, and basically the argument, historians will say a culture that runs on alcohol or other forms of, like, barbiturous depressants, it's just going to be a less productive culture. And then there was this kind of change. One of the explanations for kind of the rise of the European Enlightenment and the scientific revolution was basically the shift from alcohol all day to basically tea and then coffee on the one hand, and then smoking when smoking, when tobacco really took off. And so basically everybody during the European Enlightenment was basically drinking caffeine and smoking tobacco, like, all day long.

Ben Horowitz
We got the best stuff. That was a very good period. And then also there's, you know, some evidence, and it's hard to figure out, but, you know, some people point to the rise of the rise of obesity in western societies is basically an exact inverse match to the graphs of the decline of smoking. So should companies have nicotine? Should companies have zin stations next to the coke machine?

Well, yeah, I mean, my, you know, based on what we know now, probably the only thing with these kinds of trends is, you know, the longitudinal studies are usually completed after it's too late to save all the people. So I, you know, I never. I don't think I'll do it, but. But it probably makes sense. Maybe we should do some self experimenting first before we actually implement such policy.

Marc Andreessen
The other big news this week is apparently, apparently the weight loss, the semaglutide weight loss drugs like ozempic, apparently they are causing pregnant. They're apparently causing pregnancies. So they're. They're apparently causing possibly a hormonal birth control work less effectively, or they're just ramping up fertility. You know, fertility would be kind of positive in society because we have, like, a fertility crisis.

Exactly. And then there's a question. This is, like, brand new research, and so I don't think they haven't figured out yet, but. And there's this question of, like, okay, to the extent it is ramping up fertility, is it happening because it's actually doing something directly, or is it because actually more body fat actually, actually lowers fertility? And so, in other words, maybe it's a direct effect caused by the drug, or maybe it's a side effect of weight loss.

Right. Fertility crisis is actually just the fat crisis. And then if you fix the fat crisis, you fix our other problems. That would be nice. Yes.

Yes. Because there's not. Whoa. My God. The amount of hand wringing about everything from plastics to this to that.

Ben Horowitz
And maybe it's just people are too fat. Yeah. Yeah. Well, and, you know, you might hypothetically say, boy, if, you know, people are thinner, they might feel like having sex more often. And people might feel like having.

Marc Andreessen
And, you know, maybe. Maybe, you know, I don't know, maybe you can kind of squint. You can kind of paint a picture of, like, okay, how about a nicotine and ozempic fueled. That's a bright future. Exactly.

Ben Horowitz
Than weed and percocets. Exactly. Exactly. Or as I like to say, meth and cheetos. Yeah.

Marc Andreessen
So. All right, well, we can leave that topic there. We're gonna get in trouble for something. You know, we're more than an hour in, so nobody's watching anymore. Anyway, let's close on one.

Actually, the question is. I don't even. The question is a funny question, and I'm actually not sure if he intended it seriously, but it actually is a very interesting underlying question either way. So the question is, John Bind asks how much of the backlash against remote work is investors concerned about the commercial real estate market versus actual productivity concerns? I kind of like the idea.

I would say I kind of like the reaction you're getting from enemies. I like the idea that investors are so machiavellian that were collectively trying to somehow paper over the catastrophe in commercial real estate by forcing people to still work in the office. I guess the good news is that's not happening. And the reason that's not happening is investors. Where'd that come up?

Ben Horowitz
Actually? Yeah, well, the investors invest in commercial real estate are different than the investors that invest in companies, and so it's different people. So that's not literally happening. However, having said that, it does surface this very interesting question, which is, yeah, the commercial real estate market is in this state of sort of, I don't know, it's like, it's in the sort of Schrodinger situation. Wiley Coyote, over the cliff.

Marc Andreessen
Yeah. Yeah. So you have all of these, you have all of these buildings in particular. You have a lot of buildings that, where. You have a lot of buildings where the lease payments are not much, or the interest payments on the debt on the buildings is much higher than it was because of inflation.

And so the landlords are in some level of financial distress, and then companies that don't need as much commercial real estate because they're doing more remote work, they're going to start to let leases expire. And then you combine that with the issues we talked about last time, like crime in the city centers and so forth. This was a catalyst for some of the banking crises a while back. And there's a lot of banks that, in theory, are in some trouble on this. And then there are cases where buildings are now changing hands for pennies on the dollar.

Because the way commercial real estate works basically is like, if I buy a commercial real estate building, I buy it in part for equity and part for debt. If the price of the asset drops enough or the interest payments get too high, at some point, I just basically, they say, turn the keys over to the bank. Just be like, fine, that building is bankrupt. You hand the building to the lender, the bank is then holding the building. The bank has no ability to run the building, doesn't want to own it, and so then they liquidate it, and the tenants have fallen out in the meantime, which really impairs the value of the building.

And then the next buyer for the building says, great, I'll buy it for $0.20 on the dollar. And so you've got these assets where you've got this basically massive potential catastrophic crash in prices. By the way, this has all happened before. There have been real estate goes through cycles, and so this does happen. So there's one scenario in which there's a very big wipeout across the entire commercial real estate sector, and it takes down a whole bunch of banks.

And it's a big thing. There's another version of this where, no, actually, people are still going to be mostly in the office three days a week. If you're in three days versus five days, you still need the space and it's going to be fine, and the city still draw people. And then there's another version of this where there's some more complete reinvention of all the other topics we've been talking about, some complete reinvention of work. And companies are configured very differently.

Anyway, maybe just touch on this as the closing thing that we talk about. But, yeah, what's our prognosis? There's another crisis factor, which is real estate loans. Commercial real estate loans are typically made by regional banks in America. And what we learned during the Silicon Valley bank run and the subsequent things is the federal government may or may not bail out regional banks.

Ben Horowitz
And so if you had, say, 500 of them all in crisis simultaneously, what would happen? And hopefully we don't have to answer that question, but it certainly plausible. Now, I do think, you know, one of the things we are seeing in at least office space is the office space. Rents haven't dropped that much, at least in kind of San Francisco, New York, things like that. The mall space could be cataclysmic.

And interestingly, I think that's driven more by, it's also driven by COVID, but some combination of, you know, the lockdowns causing online buying behavior and then the lack of, or the, this new, the new movement to enable theft up to a certain level and things like that, you know, plus whatever lacks or law enforcement, et cetera, et cetera, lacks prosecution has just causing these things to no longer be viable businesses. And so there's nothing to replace them with. And then you go, well, who's going to use that real estate? And that's a very big open question, I think. Yeah, that's right.

Marc Andreessen
Yeah. And actually, I might even, I'd even paint a grimmer picture on the. We learned during SAB, we learned not only, not only if a regional bank kind of hits this kind of crisis point like SUV did and some other regional banks did, like whatever that was a year, year and a half ago, government may or may not bail them out, but the other thing we learned is the government may actually prevent them from being bailed out in the private market. Yeah, that was kind of diabolical what the FDIC did. Yeah.

So there were so people who didn't track the SVB crisis carefully. There were other big banks that wanted to buy SVB. And there were private equity firms that wanted to buy SVB and basically do a private sector bailout requiring no taxpayer involvement. And the FDIC actually blocked those also. Almost as if it was determined to have a catastrophe.

Ben Horowitz
Yeah, yeah. The reasons were really weird. Like, one of the blocks was there was a bank that was going to buy Silicon Valley bank, and by buying Silicon Valley bank, it would move up into the GSIB category, which is globally systemic, important banks. And they're like, no more gsibs. It's like, really?

So you just let the bank die because you've got some principle that you don't want more g sibs? They're all going to be g sibs if you keep up the. This policy. Obviously, it's like the most craziest idea ever. And then.

Yeah, there were other ones where, like, it was a foreign bank, or is it this or that weird behavior? Yeah, yeah. So the perverse outcome from this is that after the 2008 financial crisis, there was this huge focus by reformers on what were called the too big to fail banks. Right? Remember, the whole book on the global financial crisis that Andrew Ross Hirkin wrote was called too big to fail.

Marc Andreessen
And the whole thing was, we want to not have the companies. Because if a bank is not systemically important, if the bank goes down the way a normal company goes down, and it's fine, and other competitors just pick up the slack, then there's no need for a bailout. It's only when these banks get so large they become, quote, systemic. And then if they go down, they're going to take down the rest of the economy with them. And so the big conclusion everybody drew after 2008 was bailouts are bad.

So to not have to do bailouts, you just can't have these banks be that big. So what they did was they put in place a legal regime called Dodd Frank and a bunch of other regulations that had the result know, sitting here 15 years later of the too big to fail banks are now much larger than they were before. So the exact opposite of, yes, gigantically, titanically large. And then on the small end, new bank charters in the US have dropped basically to zero. And the reason is because Dodd Frank, all these regulatory requirements is basically impossible to economically justify.

Ben Horowitz
There was a bank that proposed a new charter where they were going to. They were not going to lend, they were going to use zero leverage, so no risk. Basically, they were going to only take deposits and lend out money that they actually had, and at a ratio such that depositors wouldn't lose their money and, you know, no hedge fund or that kind of activity. And it was denied and the longest denial ever written. So it does seem like at least the current administration, administrative state is determined not to have new banks at, like, a very extreme level.

Marc Andreessen
Yep. And then to Ben's point, you have these. The other banks are these regional banks, excuse me, the mid sized banks, which are kind of local to cities. And the whole point of the regional banks is for sectors like real estate, where you want to actually have, like, a hands on. You know, you want your bank to actually be local.

So they might be hands on, like eyes on a real estate development project where they actually know what's going on and they're aware of everything. And so the specific danger is that a broad based wipeout in commercial real estate, catalyzed by all the other things that we've been talking about, would take down the regional banks. And if that happens, then basically, to Ben's point, the only banks that are left ultimately would be, like the big ten or maybe even the big three. And at that point, you've actually centralized the banking system to a far greater degree in the hands of a small number of players than was the case in 2008, which, again, is the exact opposite of what everybody said they wanted. And so it's a.

For people who are into this kind of thing, the rabbit hole as to why that's the case and why we seem to be in this, like, mad rush to get to that resolve is a very interesting question. Yeah. Yep. I'm not going to even comment on that because it would be a very long conspiracy theory answer. Exactly.

And so we will leave it there. So, Benjamin, it is a pleasure, as always, and thank you again, everybody on Twitter X, for all of the great questions, and we will be back soon. All right, thank you.

Ben Horowitz
All right, thank you.

Startup Building: Challenges & Opportunities

Primary Topic

Episode Summary

Main Takeaways

Episode Chapters

1: Introduction to Post-COVID Startup Landscape

2: The Role of AI in Startups

3: Managing Remote Work Dynamics

4: Long-Term Strategic Planning

Actionable Advice

About This Episode

People

Companies

Books

Guest Name(s):

Content Warnings:

Transcript