走进一家管理规模超68亿美元、身后有80多家上市公司和165多起并购的基金

Almost no iconic company in the history of venture capital has come from anyone’s thesis area. No one was sitting on OpenAI when Khosla invested in OpenAI, before ChatGPT was really a thing, going, “Hey, my thesis is that LLM chatbots are going to be a thing.” No one’s saying that. The investment is, “These guys are really smart, and they’re working on something that could be pretty valuable.”

The number-one thing I’m looking for when it comes to the founder is: are you just doing this because you think being a founder is cool?

Mm-hmm.

Because everyone thinks being a founder is cool and high-status. I don’t care about that kind of founder. There’s information asymmetry between what most people in AI know—the Venn diagram of things people in AI know and care about—and the things that matter in the world. Who is at the very edge of that little, tiny intersection, where they’re able to capture a little bit of it, is pretty important.

I run such a cool company right now. If I get acquired, I will no longer run that cool company. Why would I want that?

So, Deedy, welcome to Sorcery. Sorry, DD—welcome to Sorcery.

So great to be here, Molly.

Congrats on the big promotion. I know you were just promoted to partner at Menlo.

Thank you. Thank you.

This is really cool. I want to know: between it all, you were a founding team member at Glean, and you’re now at Menlo. You are following the dream. You’re becoming an investor in Silicon Valley. What is your biggest bet right now?

1. Goodfire Opens the AI Black Box

When it comes to the big bets that we make right now, I’ll tell you about 1 company that I’m super excited about—we’re investors—and then just an area that I think doesn’t get enough attention, that I’m pretty excited about. The 1 company is Goodfire.

Mm.

I don’t know if people know about it. That will probably change, hopefully soon. Goodfire is a company that does what I like to call brain surgery for AI models.

Mm.

The technical term is mechanistic interpretability for AI models.

Oh.

These are researchers who founded their respective teams or worked with the very early teams in interpretability at Anthropic, DeepMind, and OpenAI. They came together to say, “Hey, I think if we have AI everywhere in the world today, there is no future that we want to live in where this is a black box and we don’t understand what’s going on.”

All explainability techniques for AI today are empirical: they’re looking at the results, they’re evaluating it. That’s not fundamentally explaining why a model does what it does. A good example is when ChatGPT—GPT-4o specifically—became super sycophantic.

Mm.

That’s not something they really caught in eval, right? That’s something they found out much later. What if you could peek into the brain of a model and find out what it was actually thinking, why it was actually saying the things it said?

I think they’ve come up with some really, really interesting, novel discoveries that are still pretty secret, and that I think could really change the game of how we think about AI in the future. So I’m really excited for that company and the future that it’s looking at.

And the name of that company was?

Goodfire.

Goodfire?

That’s correct.

I wonder what they found.

Can’t say.

You can’t say?

Can’t say.

Tell me.

It’s a super-niche scientific breakthrough. I think it’s quite interesting.

Okay.

You know, one of the interesting things about backing labs that can be quite anxiety-inducing is that they’re probably not on this revenue train where the only thing that matters to me is getting deals through the door and getting money. We think a lot about what the fundamental scientific discoveries are that we’re going after, and then when the right time is to productionize them into a go-to-market product.

I think I would say they have done a big part of 1 pillar of the 5 things that they’re looking to solve.

Mm.

Unfortunately, that’s all I can share for now.

Ah, classified. Bummer. We’ve only run into that when talking to Palantir, but this is another classified conversation, I guess.

Mm.

Okay, so you’re really into this company, and then also an area.

2. Labor Shortages Reshape the Economy

Yeah. In terms of areas, what I’m seeing is—I still think, and I know this is a very cliché thing to say, everyone says it, every single VC says it—but I mean a very specific thing when I say this. When I say we’re in the very early innings of AI, what I really mean by that is, if you really look at birth rates in the world, there are just a bunch of jobs that people can’t hire for anymore.

Mm-hmm.

It’s not just in America, Europe, and parts of the Western world; it’s true even in other places. If you look at what young kids want to do and have wanted to do for the last 10 years, they want to be YouTubers, coders, or have desk jobs, right? Primarily, people want to do desk jobs, or positions of power, maybe. No one does blue-collar labor. No one wants to do, say, a tax job. No kid is dreaming about being an accountant.

Mm-hmm.

10 years before that, I think there were some kids saying, “Look, if I get a nice, stable job, that’s all that matters.” I don’t think kids think that way anymore. What happens to the industries? They actually can’t hire.

I talk about this a lot, but the gap between what the Valley thinks about and focuses on in tech and AI, and what the world cares about, is pretty vast. I think that gap isn’t even close to being filled.

So when it comes to things like insurance, what do we talk about? The professions we know. Here’s the most common thing that happens in the Valley.

Mm.

We talk about finance and legal, because that’s where other Ivy League school kids go and work. Therefore, there’s an AI for finance, and there’s an AI for legal. People don’t even know what else exists, really. Tax—we know tax. But think about insurance brokerages, trucking, and logistics. People do a lot of things in the economy, in the world, that are not finance and legal.

Mm-hmm.

Just because you know a couple of private equity guys and a couple of lawyers doesn’t mean that’s all everybody does. So I think there’s a huge, huge gap to be solved there, and we’re only scratching the surface on what is really important there.

If you can’t hire these people, you rely on the technology to be able to fill the gap. You depend on it. You really hope it works. You’re not betting on some future where AI can solve math or coding in a specific way. You need this now. So that’s an area that I’m really excited about.

Wow.

Longer-term fertility efforts, I think, generally are pretty interesting.

Oh.

So.

Interesting. What do you mean?

Humanity, in the modern day, has never seen an extended period of time where the population has fallen systematically. We don’t know—and I don’t want to go on a pronatalist agenda; it’s not like that. I think the real point is, we do not know what the repercussions of that are for the economy.

There are some very smart people, much smarter than me, who would say a lot of our economic growth is built on the fact that consumption continues to grow. Consumption continues to grow in large part because the population in the world continues to grow. What happens when the population starts shrinking? You’re not going to buy double the furniture and double everything else, right? Consumption is somewhat limited in terms of variance.

So if the population’s not growing, where is the economy going to come from? Where is spending going to come from? What does that do to the world? What does that mean for humans if populations start declining?

I’m curious about that. I know there are a lot of sub-issues underpinning why that happens, but if I take a 10- to 50-year time period, that’s one of the things that I think is pretty interesting to look at.

So we’ll need furniture that breaks more easily, and we’ll need iPhone releases faster with more complicated adapters.

7 cameras—and then maybe people will buy the iPhone 20.

Why just have 3?

Why 3?

Go to 7. We should really be pushing the limits here.

Technologically, Apple’s not thinking hard enough.

They’re not. Let’s be real.

Yeah.

3. Deedy Builds Influence on X

Before we go way too far, because I want to go into the portfolio and your thesis and everything, and you have a really great technical background, we need to address something. You are ridiculously famous on X. What is going on here? How did you become so famous?

I wish I really knew the answer, but…

Like I was saying before, I can give you some theories on how this happened.

What’s your strategy? Okay.

I’ll talk about the journey of X. I’ll talk about strategy, too.

Mm.

I’m very open about this stuff.

Okay.

The first thing—I was telling a friend this just the other day—a lot of people now will come up to me and say, “Hey, man, great Twitter game. How do you do it?” And it’s concerning. There are people in positions of power where I’m like, “You really shouldn’t be the one asking me this question. You’re a billionaire. Why are you asking me about my Twitter game?”

Mm-hmm.

But people care about this stuff. I always say that there are 2 main things that I think people do wrong. I’ll start here, and then I’ll tell you my journey.

Mm.

The 2 main things are, 1, I think people are in it for the wrong reasons, and it’s so easy to fuck it up if you’re in it for the wrong reasons.

Mm-hmm.

Most people care about their Twitter game so much that they like other people hearing the sound of their own voice. That’s the primary reason people want to be on Twitter or be big on anything. They’re like, “I want to be heard,” essentially. I think that’s the wrong motivation. It doesn’t work.

That’s 1. Number 2, most people who fall into that first bucket—this is true for Twitter especially, not for other forms of social media—I just don’t think they have thick enough skin to endure it.

Yeah.

I know plenty of people who are the—I don’t want to say cookie-cutter, but kind of cookie-cutter—overachiever archetype who are like, “I studied all my life and I went to Stanford and I did all this stuff.” That archetype of person usually can’t take Twitter when it gets real.

Mm.

You can’t do Twitter without getting canceled a few times. You can’t do Twitter without getting hate a few times, without people saying the meanest things about you for the world to see. You just can’t do it without that. Most people don’t have thick enough skin to be able to endure that long term.

I’ll pause there, and then I’ll talk about what I did and my journey. I’ve always liked writing from a very young age. I joke that if I wasn’t in tech, I would probably be a writer or a journalist. I’ve always been writing since I was in high school and college. Some of the pieces I wrote were usually data explorations. It was like, “Oh, I found this interesting data. This is interesting.” And then other people liked it.

Mm-hmm.

The first big one was called “Hacking the Indian Education System.” This was when I was a freshman in college. I did this piece where I scraped all the data for essentially the equivalent of an AP exam or the SAT in America, but for India, and I found that a bunch of these exams, which are taken by 1 million people every year, were flawed in their grading. There were statistical anomalies and all this other stuff.

That went viral, and it got the feedback loop going of, “Hey, maybe I should write a little bit more.” So I did a bunch of these little things over time.

The second big reason that really convinced me this was something to take seriously is a touching anecdote. I love it so much. I didn’t just write publicly. There was this Facebook group of high school kids applying to college, usually international students coming to America for college, and they had a bunch of questions. Not everything is clear or online, and not everything is crystal clear to them.

So I would go and offer to do an AMA every year. One of the people who ran the group said, “Hey, maybe you should come do this. I think you’ve been helpful to me personally, so I would love for you to do that for the community.” So I did that for 4 years straight.

In the first year, the way these things go is hilarious, because there are always guys asking the douchiest questions publicly, because guys have no shame sometimes.

Mm-hmm.

And then it’s sad, because the group isn’t gendered. This was men and women. This one girl reached out to me in a DM, and it was just an essay. I’m like, “Holy shit. What’s going on?”

So I read this whole thing, and she basically said, “Look, I applied to college in America. No one asked me to do it, but I thought it would be nice, and I don’t want to stay in my country. I just don’t see a world for women here. It’s so hard to do X, Y, Z. I’m going to get married off at a young age. I don’t like this.”

And I’m like, “Okay, what college did you get into?” She’s like, “I applied to 1. I got into 1.” It was Penn M&T. And I’m like, “Dude, you should absolutely do that. Why is this a decision? You have the financial aid. You have everything you need to do that. Why aren’t you doing it?”

And she’s like, “My dad doesn’t want to send me.” And I’m like, “That’s insane. He probably has no idea what he’s talking about. What can I do to help? How would you like me to help?”

And she said, “Can you talk to my dad?” I was a 19-year-old kid. At that time, I wasn’t very comfortable talking to parents in general.

Mm-hmm.

I sort of froze up, but I’m like, “Sure. If that’s what it takes, I’ll talk to him.” And I swear to God, Molly, I got on the phone with her dad for 5 minutes. That’s all. And her dad was like, “I think you know what you’re talking about. I’ll send her to Penn.”

Wow.

So she went to Penn. She graduated. She works in private equity right now.

Great.

Great firm. It was a great story, but what that really taught me in the moment was that the asymmetry in the influence you can have on people’s lives just by sending a text is so high that it’s insane more people don’t do it.

Mm-hmm.

Long story short, I couldn’t do it when I was in big tech, because I got a lot of pushback. They threatened to fire me a couple of times for brand reasons.

Yeah.

But at Glean, I didn’t have that constraint anymore, and I was like, “Well, I’m definitely going to write more publicly, because I think it could help people.” So I made it a thing. Every day, I would write something that I thought was somewhat helpful to someone.

In the beginning, there were zero views and zero likes. I didn’t really care about that, but over time, there were people who were like, “Dude, thanks for sharing that. That was super helpful. I didn’t know.”

So that’s the long and short of it.

Mm.

Today, when I try to think of a tweet, the only question I’m asking is, “Is this helpful for somebody? Is somebody going to read this and go, ‘That’s kind of useful’?”

Mm-hmm.

That’s all. Everything that doesn’t meet that bar, I usually don’t tweet. There are obvious exceptions. People can pounce on me and find some examples where I—

Mm-hmm.

—kind of just knee-jerked and did something. But I would say 90% of them have to fit that bar. That’s what maybe has driven whatever has happened online.

Most of what I see in my feed—because I’m definitely active on X, but I like a lot of data-driven content—is the data-driven content that you put out, whether it’s charts, macro analyses, or that kind of thing. What is your thinking behind those?

I have a Ten Commandments–type list. When I started writing on Twitter specifically, I told myself, “It’s so easy to knee-jerk and do stuff, right? You scroll for a bit.”

Yeah.

You’re like, “Oh, man, this is a thing everyone’s talking about. I need to have my opinion. I need to put a take out there.” I found that paradigm to be not evergreen. This is just not a systematic way to do things.

Mm-hmm.

So I told myself, “Okay, what is my thing?” I have 4 or 5 areas of competency. Second, I try to stick to facts and not opinions as often as possible.

No political commentary. Try to minimize negativity in any way—

Mm-hmm.

—and not offend people for no reason, even though sometimes I clearly have missed that goal. Things like that.

Yeah.

I have a whole list of commandments. The reason the data thing always stands out to me is because I'm like, “Well, it is literally fact.”

Yeah.

So you can't—you can hate me for saying it, but you can't argue against the fact. More people should know that this is what the reality is, or this is what the data says, on something. That's why I do a lot of data-driven stuff.

What are your favorite charts?

Like, types of—

Name them.

Types of charts?

Yes. Tell me all your favorite types of charts right now.

Oh my God. I don't know if I have a—

You guys are investors in Carta.

They put out lots of charts.

Peter is so good at this stuff.

He's so good.

I actually met Peter in person recently.

Okay.

And he told me he literally just took the job at Carta because of the data access that he would get.

It's amazing.

And I'm like—

They're great.

You do such a great job.

Yeah.

I am so fortunate to have Claude Code to be able to do all my charts right now. I've written a big skill for how I like things to be graphed, and I'm very clearly opinionated about that—the exact color, what the themes have to be, and things like that. I put them into a skill. That's how I do my charts. I don't know if I have a favorite chart.

Can you send that to me so I can use it?

Yeah, absolutely. And I use really rudimentary tools. The other tool I use is the new thing on the Macs these days: Freeform. It's that new app that no one really knows about.

Okay.

It's kind of like a canvas where you can drag and drop stuff. It's kind of like Microsoft Paint—

Yeah.

—without the Paint. So you just drag pictures and text and stuff like that. I use that. I take the pictures from different places, and I put them together.

I use Google Slides a lot. That's how I make most of my graphics, and then I toggle between Figma and Canva.

I suck at Figma, admittedly—

Really?

—and Canva, so I know some people are really good at that.

It's a good skill.

Yeah.

4. Anthology and Menlo Invest Differently

Okay, to shift this back to investing, you spend 50% of your time on mostly Menlo portfolio investments, and then also on the Anthology Fund. Can you just explain the structure between those two, and then we'll go into one?

Right. So let's talk about the Anthology Fund first.

Mm-hmm.

The Anthology Fund is a fund that we do with Anthropic. It's a $100 million fund. The goal of this fund—it was set up, like, a decade ago in the AI world, but at the beginning of last year—

Oh my God.

—when Anthropic was a no-name company that no one really cared about. The goal was, “Hey, let's invest in companies around the ecosystem without making this a corporate venture fund.” Instead of having Anthropic do it, let's have Menlo do it, because we were the biggest investor in a bunch of those rounds.

We chose to do that fund as a way to say, “We just want...” There are 3 kinds of companies that meet that criteria. One is a fantastic early-stage team trying to build an AI company. Let's get them early and back them. It can either be a lead or a small follow check. Usually, the minimum we do is $100,000, but we go all the way up to leading those rounds.

The second is something of strategic importance to Claude or to Anthropic, essentially. So things like Turing—we talked about Turing. We talked about a company like Mercor. Various other companies are just extremely important to the ecosystem of Anthropic and maybe other companies as well.

The third thing is just iconic companies that are building on Claude, whether it's seed or not. Those are the 3 types of companies that we like, and that's how we invest out of the Anthology Fund.

Now, the Menlo Fund—the way we do our deals is that Menlo is generally a fairly low-volume shop. What I mean by that is we're not trying to do even 5 deals per partner a year. We actually have quite low volume, and we like to pick and choose the companies we work with very carefully.

But when we do pick, the reason we keep the volume so low is because we actually want to work with those companies to drive real outcomes. We want to have the time to be able to work with our portfolio companies. We think if you're doing 5-plus deals per partner a year—definitely 5, but I would argue even 3 or 4—at some point, 2 years in, you're going to have 6 companies that you're heavily involved with, and you're not really going to have time to do work for any of them.

We like to do that because we actually think it makes a huge difference when it comes to not just the classic venture stuff, but exiting those companies. Who is working on, “Hey, let's say 2 or 3 years later, this company's not going so well. Who's going to help you do an M&A motion?” Most founders haven't done that before. They haven't seen that. They can't learn that on the fly. These are time-bound things. That's when we come in. We do a lot of work to help get those companies a safe landing, find a home, and land well.

That's our main fund strategy. Recently, we've hired a bunch of tenured operators, such as myself, but also Tim Tully, who comes from Splunk and was CTO there; Joff, who was the chief product officer at Atlassian; and Matt Kroening, who started and sold one of the few cybersecurity unicorns that have been sold for above $1 billion in the last 10 or 15 years.

These are people who've built stuff, who know how to hire, and who know how companies operate. We think that's a big differentiator for how we view and want to work with companies. We're not just investors. We could build this company with you. That's the long and short of it.

You mentioned one of your portfolio companies at the start, but could you just break down your thesis and what you focus on?

5. Founders Matter More Than Thesis

2 things. The first thing is I have this huge gripe when people ask VCs their thesis because, honestly, I think all of them are semi-lying.

Okay, say more.

This is a hot take, right? All smart VCs know this, so it's not like they don't know this. But almost no iconic company in the history of venture capital has come from anyone's thesis area, right?

No one was sitting on OpenAI when Khosla invested in OpenAI, before ChatGPT was really a thing, going, “Hey, my thesis is that LLM chatbots are going to be a thing.” No one's saying that, right? The investment is, “These guys are really smart, and they're working on something that could be pretty valuable.” That's your thesis.

Mm-hmm.

The same thing with Facebook. No one had a thesis on social networks going, “Hey, guys, we looked at Hi5, we looked at Orkut, we looked at Google Plus, and we really decided you guys were the one.” It was a ripping product with a fantastic founder who had a lot of ambition, and we were like, “Okay, this could go somewhere.”

So that's the part where I think most people are semi-lying. We don't have—not every VC really has a strong thesis. They have a thesis, and I do have a thesis. But most of the investing we do is not just bound to that thesis, right? It's bound to whether there are incredibly—

Mm-hmm.

People use the word “smart” all the time, but there's a difference between smart people and founders who can run through walls with conviction on something.

Here's what I'm trying to get at. If you're a founder who has done all this stuff in tech, you've had a 10-year career, you could work at OpenAI and make millions of dollars, or you could work at Anthropic and make millions of dollars anywhere, and you choose to be a founder in something that has been your life's mission for 10 years, if it was really your life's mission for 10 years, I take that seriously.

I'm like, “Okay, why do you care about this? Why is it meaningful for you, and why do you think this would be a big thing?” That's my bet on teams. The number 1 thing I'm looking for when it comes to the founder is: are you just doing this because you think being a founder is cool?

Mm-hmm.

Because everyone thinks being a founder is cool and high-status. I don't care about that kind of founder. There are some sniff tests to figure out who they are. It's not easy, but you can sort of figure out who they are. It's not like all founders who want to do it for status are bad.

They often do really well. But my belief is, the most important thing is that I want you to wake up 5 years later and be like, “I would be doing nothing else except this.” Whether your company works or doesn’t work, I need to believe that you believe that to bet on you.

Not for the least of reasons, what if you just run with the money? I don’t know. What if you’re a bad founder and you just blow something up into the sky? You blow it on this, you blow it on that, it doesn’t work, and you’re like, “Ah, whatever. I don’t know what to do.” We’re trying to avoid that kind of founder, right? We want a founder that really, really genuinely believes in why. So that’s number one on the thesis, I would say. But then I get into areas that I think are interesting also.

I want to get into areas, but I want to understand that a little bit more. How do you pressure-test those types of founders? What are the traits you’re looking for other than really wanting to build a company?

One thing I can easily say that we’re not looking for is that it is very hard for us to get to conviction on a founder or a company when we haven’t known them for a while.

Mm.

I think personality, like with—you know, VCs love using the word “shotgun wedding.” I don’t know. But it kind of does matter. You just can’t trust the nature of a founder when they’re pitching to you right before a raise. They’re selling to you, right? It’s really hard to get through that. People can be great salespeople.

What I really care about—and I can’t speak for the firm. I can say most of Menlo operates this way, but for me especially—is I’m looking to build that relationship way before you’re raising, and I want to understand what makes you tick. Half the time I’m not talking about work. I mean, we have to sometimes talk about work because it’s weird otherwise.

Yeah.

Like, why is this VC talking to me about my personal life? But I kind of just want to understand what makes you tick. What are your motivations? How did you grow up? What drives you? And that gives me at least a lot of confidence on whether or not you would be long-term successful.

And then on areas, what are the areas you focus on?

I’ll say one last thing before areas.

Okay.

One last thing I think that is important is that there are a bunch of founders that even meet that bucket. I think the biggest bottleneck today is the ability to hire.

Yeah.

The number one reason is there are plenty of really smart people who also care about something very deeply. The unfortunate truth is I look at them and I’m like, “But, man, who would work for you? I just don’t see you being able to tap into a network or even be super convincing when it comes to candidates and get people to buy into this mission.”

You almost kind of need to be a little bit of a cult leader as a founder, and not everyone is capable of being a cult leader. People who do it well can hire well and retain well. So that’s the second thing on the founder side.

6. The New AI Investment Map

On the area side, one of the things I mentioned before is, how can we apply all of these novel technologies to the industries that need it the most? The industries that no one talks about, the industries that are super, super boring, the industries where there is information asymmetry between what most people in AI know—the Venn diagram of things people in AI know and care about and read Twitter about—and things that really matter in the world.

Who is at the very edge of the tiny intersection between that, where they’re able to capture a little bit of that, is pretty important. A lot of people, and sometimes they’re successful, kind of back into it. They do the top-down research and find, “Hey, this is an area that’s important. Let me go attack it.”

Ideally, it’s just an area you know for whatever reason, and you grew up with it. You know it. Maybe your mom or your dad worked in it, and you’re like, “I know this area so well, and I know there are just so many efficiencies. I also happen to be really good at this AI stuff. Let me see how I can apply it.” So that’s one area. There are many, many such verticals like that.

Number 2, I care a lot about the fact that we’re at this point where research is not happening in academic institutions anymore. They just don’t have the money to fund frontier research. So how do you underwrite high technical risk, and what research problems could unlock economic value? And how do you find those teams? That’s an area that I care a lot about—fundamentally interesting research.

I would say the third thing that I personally care a lot about is that I look at these companies that are—how do I say it?—just beautiful product builders. I think that art is a lost art. People don’t give it enough credit.

A very overused example in venture, but I love the example of Granola, and I’ll give you why I like this example. No one thinks meeting note-takers are fundamentally interesting, right? No one is waking up every morning like, “I need to fund a meeting note-taker company.” No one. That’s not a thing you think. There are many of them that exist. There’s nothing fundamentally novel about the technology, and out comes Granola, and you use this product, and you’re like, “That’s beautiful. It just works. I don’t really have to think about it. It’s just kind of nice.”

That’s a lost art. Not many products are like that. Most products are really forced. It’s AI for this, so therefore you should use it for this. I’m like, “Dude, I’m not going to use it. What is it solving?”

So I think product beauty and taste—I know that’s also overused—is kind of, that set of ideas is really interesting. Also, because I truly believe that you—and for many products, here’s a theory: if your enterprise product can be PLG, then it must be, otherwise a PLG company will absolutely eat your lunch and destroy you.

It kind of ties back to a lot of venture capitalists these days talking about momentum and whether a 3-3-2-2-2 is enough, or you have to grow faster than that.

Mm-hmm.

I don’t really have an opinion on that. I think the reality is, if you are going enterprise, you are held back by these insanely long sales cycles, and the industry and the technology move faster than your sales cycle.

So not only in 2 years is your revenue weak, but your product is shittier because you’ve done all these shitty things to appease enterprise clients, which we all have to do, but you’ve also not adapted, typically, to what’s going on in the industry. You’ve made decisions that are irreversible based on how the industry has moved. Classic RAG solutions have now become agentic solutions, so those actually mean something technically, but people couldn’t evolve their product.

I do think that for the set of ideas where you can go product-led growth, they will absolutely destroy you in sales. You can do these long sales cycles, and then a product goes viral, and then they come into your enterprise sales call, and the buyer is like, “Yo, I’ve heard of this one. We should buy that one.”

That’s essentially how decisions are made in many cases in the enterprise.

Yeah.

So that’s another area I like a lot: tasteful products with a PLG motion.

Speaking of one that was recorded in this room, there was Cluely. You have a funny story about this?

I do have a funny story about this.

This is a hard pivot, but this just reminded me: product, viral, okay, Cluely.

A lot of people have a lot of opinions on Cluely. I actually think that—I find myself having to defend Roy a lot, not because I think everything he does is right, but I actually think he’s a very thoughtful guy and has a very—not brash exterior, shall I say.

My story is, when Interview Coder became a thing, prior to Cluely, that was what Cluely was, I think, prior. It was called Interview Coder. It was a tool for cheating on your tech interviews.

I remember seeing this somewhere—I forget where—a small post. I broke a bunch of my Twitter rules, and I’m like, “Dude, this is hilarious that this exists.” I hate interviews. I hate tech interviews with a passion, and the reason I hate them is because I’ve seen so many people that are, I call them, hardcore strivers. All they do is game the interview process, and they’ve had fantastic tech careers.

And I’m like, “Dude, you guys know nothing about building product. How are you a director at X company?”

Mm-hmm.

There are hundreds of these people, especially in my—I’m, like, 30. I can’t remember. I’m in my 30s.

I see people who are peers, and with some of them, I’m like, “You are not good at your job—”

Mm-hmm.

…and somehow you are so senior, and the reason is because you could game tech interviews.

Yeah.

“I think it’s not representative of your skills.” So I see this product and I’m like, “I like it.” I like what it stands for, which is that these interviews suck and they have to go. So I tweeted it, saying, “Cheating is really bad. Absolutely don’t do it. Especially don’t use this tool. It’s really bad to use this tool. This is the tool in case you ever use it by mistake. Don’t use it.”

Oh my God.

Something to that effect. And that blew up.

Mm-hmm.

Clearly, there are some people who are like, “Oh my God, you have no morals,” whatever. And then some other people who are like, “Yeah, I mean, I hate tech interviews too, so kudos to you for doing that.” And that, I think, was the first time that anything related to Cluely or Roy went super viral. That post had several million views on—

Mm—

…on Twitter. But then Roy took it on his own, and he—

He took it on his own.

He’s a force of nature.

Yeah.

Before I knew it, within a month, I was asking him for—

Allocation?

No, not allocation. But I was asking him, “How do you do all of this?”

Yeah.

Um—

That’s funny. So Roy, I think he started Interview Coder at Founders, Inc. That’s where we’re recording. I try to keep that secret, but I love this space, so gotta give them some credit. Specifically on your areas of focus—so infrastructure—what else are you looking at?

Pretty broad.

Mm-hmm.

One of the many reasons I chose to join Menlo specifically was that I wanted to keep the scope very broad. So I look at AI, SaaS, and infra, which people would argue is everything.

Isn’t AI the same thing as SaaS now?

I guess. AI is the same thing as a lot of things now.

Yeah, isn’t AI everything?

I think everyone’s, by default, an AI investor.

Oh, this is a good point. Carta put out a recent report that Series A companies that are AI-enabled or have AI in their name are getting a 30% premium.

Mm-hmm.

I can’t wait until the next report comes out to see where that delta is, or how much it’s changed. But if every company is AI, how are we measuring this?

Oh, man. I would—

Every company’s AI. It should be.

Well, I think the stat that you cited was a 30% premium, right? You said—

Mm-hmm.

I’ll tell you that maybe I take 10 to 20 pitches every week, and in the last year, I’ve seen the ups and downs of different buzzwords go in and out of fashion in a way that’s hilarious. There was one time when every company was like, “Well, our differentiator is that we get data, and then the data flywheel feeds back into fine-tuning, and that’s why we’re different.” The same guys, 1 year later: “Yeah, yeah, that same thing, but instead of fine-tuning, it’s RL.”

Mm-hmm.

And I’m like, “What do you know about RL? Tell me a little bit about it.” And silence. “You know, it’s reinforcement learning. You know reinforcement learning?” “Yeah.” I’m like, “Well, I don’t know if I know, but what do you know?”

So what I’m trying to say is, founders—many founders—are coached to say the right things in a pitch, and maybe they should be. Maybe this is the game that they’re playing. I would say that one of the reasons that when we look at deals, me and, say, Tim, we’re both very technical. We both code with this stuff all the time. One of the powers of that is I don’t think I know much of anything for sure, right? But because we code with these things, we’re not embarrassed to ask the dumb questions.

Mm.

A lot of VCs are, because they don’t code, right? They don’t actually understand what they’re talking about on the tech side.

Yeah.

I can ask them, like, “Hey, when you say RL, do you mean this? Do you call this a library? Or what are you actually doing? What are you actually using for that?” And if you prod deep enough, I would say for at least 90% of companies, they’re actually just changing a prompt on an LLM.

Mm-hmm.

Right? And then they’re pitching all of this other stuff that they hope exists or want to believe could be the future of what they build, but that’s not what they’re building currently.

If you ignore all of that and look at the few companies that I think know what they’re talking about and are very deep, I think the premium is, in some sense, justifiable because it’s such a constrained talent pool. The number of people who actually know what they’re talking about when it comes to several parts of the AI stack, if you call it that, is so small that if you know what you’re talking about and the VC genuinely believes you’re working toward an economically viable direction, I think the premium is justifiable. And that’s another one: I would say at least 70% of AI companies fall off because the TL;DR of their pitch is, “I just want to build models. I’m not sure why it’s useful.”

Mm-hmm.

The few times you have the AI guys who know what they’re doing, they know what they’re going to apply it to and how this is economically valuable. To pay a 30% premium for that? Yeah, I’d do that. We’re betting on an outcome that’s 10X, 100X, especially at the stage I’m looking at, which is primarily seed through B. But in this case, I’m thinking seed and A. Yeah, I would 100% pay that premium.

Yeah. My steelman to this would be that the premium for actually good companies is much, much higher than that.

100%.

Because when competition comes into play, pricing goes out the door, and you just want to pay whatever you can pay to win that deal, and hopefully it’s somewhat responsible. But I would say that’s probably part of it as well.

Yeah, we try to keep more discipline around that. I’ve seen some of these billion-dollar seed-type rounds.

What’s your investment range? Do you have parameters around check size and valuations? How do you think about that?

I would say all parameters are guidance, and they’re rough.

Mm-hmm.

At the same time, despite those rough parameters, it’s so hard to justify $1 billion. I just don’t see how people can write a $1 billion seed round unless—here’s a whole other hot take—it’s one of those situations where the money’s not coming from the fund, right? It’s one of those SPV-type structures where it’s like, okay, I’m kind of going to put in $1 million of a $100 million round and raise $99 million from other people. I’ll put my name on it—

Yeah.

…but I’m effectively taking none of the risk here. I’m just getting all the marketing value of being the big, bold VC who’s backing these egregiously long-term bets. So that happens. But aside from that, I don’t know how you can justify putting it out of your own fund.

Mm.

It doesn’t make that much sense. I don’t want to put a hard number on the seed that we’d go up to, because sometimes we break that, sometimes we do exceptional stuff, but definitely not $1 billion. Even a $100 million seed, yeah, I can see that. I can see that in some cases. So that’s how I would put it.

I was having a really fun conversation with Anne, who works at Elad’s fund, and we were talking about this because there are some companies that I am close to that are in very competitive markets, and their competitors are raising billions and billions of dollars. And so now the competition—and the metric for success and winning—is who can raise the most money. But you’re a software company. What are you gonna do with 400 years of runway? I don’t know how it ends. Because I think these momentum trains go for so long, and then you’re kind of left with, “Okay, we raised a boatload of money at a crazy valuation. Now we gotta back up into it.”

Yeah. There’s not much rocket science to be said. I think more capital can sometimes unlock velocity, and if you draw a graph of capital and velocity unlocked, or acceleration unlocked, it starts looking like this for most companies. If you raise 15, you get about 80% of it for an early company. If you raise 100, you’re getting maybe 90%. But honestly, sometimes it doesn’t even look like that; it looks like this. Because if you raise too much money, we’ve seen companies where this happens: A, there’s an illusion of success, so the people aren’t hungry anymore.

Yeah.

“We’re already successful. We’re a billion-dollar company. Can we even fail?” Right? That’s one. Number 2: they start—I call this the $50 lunch problem.

They start eating fancy; they start doing fancy stuff.

Mm-hmm.

They have money to blow. When you have money to blow like that, again, the hunger dies. You're not actually working that hard. You're kind of chilling. You're already a part of this anointed company, so people don't work as hard.

Number 3, it's hard to recruit. How do other people see the upside when you're already raising at whatever price without any product? Some people will believe it, but most people won't. So, for those kinds of reasons, I actually think the curve kind of looks like that: when you raise too much, it's actually pretty negative.

You should raise the right amount, and history has told this story many, many, many, many times, right? It's nice for a moment to be a part of a hot company that's raised a lot of money, but real businesses are built over a long period of time. I think one problem with venture, again, is that it's a very tech-bubble-y thing: we ignore a lot of amazing companies that just grew slowly over time.

I can name a few, I guess, but there are companies that have hundreds of millions of ARR that most people in core Twitter tech are like, "I don't know. I don't know what that company is." And they're there just because it grew slowly, it grew over 10 years, and it was just a boring company. The media will obviously disproportionately report on the crazy stuff, so that causes an issue.

To summarize, I think it's generally bad. In some cases, I can see the logic for raising a large amount, and that is when you feel like you need to make a splash in the market. 1. You need to compete on the dollar value of equity offers.

So maybe you've gone to an RSU model, and you're a billion-dollar company. You're losing people to Anthropic because Anthropic is paying $4 million a year, or whatever amount they want to pay, and it's hard to justify a future value of your equity to an employee. But if you raise at a future price, then you can be like, "Well, your paper value is actually worth this much." And it's kind of competitive. So those are some good reasons, but there are too many companies that have over-raised, in my opinion.

Yeah. I also don't know how they would hire that fast. It's like, how are you gonna hire that fast? You're gonna triple, double? It doesn't make sense. It's really interesting to see that, and it's interesting to see that in non-AI companies as well.

I'm actually curious—though I guess this is off topic—but I don't know how much they're raising, and I also don't know what they're gonna do with it. But I do see a world where... Well, there are 2 things to call out, not specifically for those companies. Some people also over-raise because founders get secondary when they over-raise.

Mm.

So that's just a whole other topic. Generally, those are really poor incentives in the Valley. We've seen that happen. We've heard rumors about all sorts of things where that happens. That's really bad, because the founder basically exits and goes like, "Well, who cares if I over-raise? I'm out."

Yeah.

"Like, I kind of got the bag."

Mm-hmm.

So that's bad. The second thing, I think, for those specific companies, maybe they can use the money to blow it on ads and try to get users on board. I see some logic maybe for consumer companies. I see more logic for one of the prediction markets over, say, "Hey, I'm a seed-stage AI company and I just wanna train," whatever. So that's just my view.

7. The AI Scaling Roadmap

Mm-hmm. To go back into the infrastructure piece of AI, where are we in the evolution of it? Can you walk through it a bit? I know with Turing, I worked closely with them, and so I know the AI research accelerator place. It went from sweatshop data labeling to synthetic data. Now they're really focused on—to your point earlier, this is a thing—reinforcement learning. That's now the focus. So what comes after that, and how did we get here?

Let's talk about the easier one, which is how did we get here? Most people know most of this, I think. I'm not repeating that much interesting stuff.

But we trained on all of the internet data. We trained large language models on all of the internet data, and we were like, "Okay, turns out this can actually do interesting things." It sort of generalizes outside the obvious training set. I'm like, "That's interesting." That was GPT-3.

Then people realized, "Okay, well, we can RLHF this data," which is a form of reinforcement learning, "and make it like a chatbot." So it doesn't just generate random text; it can actually pretty easily generate text that makes sense in a chat conversation. And so we got to, like, "Okay, we basically sort of solved the Turing test overnight." Amazing.

Then the next step was, all right, let's train a bigger model. There was a big-model phase, right? So people were like, "Okay, GPT-3 is kind of small in the grand scheme of things. Let's go big." The scaling laws were invented, or empirically studied: How does size affect quality? And people scaled this stuff up, right? A lot of money was spent scaling this stuff up.

Then people saw it and went, "Well, it turns out that the scaling laws are proportional to the amount of data you can train on, so you can't just arbitrarily make the model super big if the dataset is just the internet." Maybe you can augment that with other kinds of proprietary internet data and do all kinds of other stuff, but order-of-magnitude-wise, you're not 10Xing the data.

Mm-hmm.

Right? You're kind of stuck with that kind of magnitude of data. Therefore, it doesn't make sense to make the models bigger because you're not getting more performance on that dataset.

Then came the RL world, right? The RL world was: Maybe I get proprietary data by paying for it on very specific things, and I find interesting mechanisms to use a smaller amount of data more effectively and get domain expertise in a bunch of these different areas. So, hey, can I buy a bunch of coding data to get good at code? Can I buy a bunch of finance data to be good at Excel? What are the things of value in the world, and how do I get that data?

So then you see companies like Mercor, like Turing, come in and sort of fill that need, especially as Scale AI left a big hole after the acquisition. So we're in that part of the cycle, and then the question really is, okay, well, how far does RL scale? The scaling laws of RL are not intuitive because you're depending on this manual process of getting data from people. That's not like the internet. You can't just 10X that overnight. So that's 1 bottleneck.

The second thing people are thinking about is, well, okay, now we have these big models, and we have RL. Here are a couple of things that are on people's minds, I would say, and this is not exhaustive. I'm just thinking about it out loud. One is RL is kind of a shitty paradigm for learning. Karpathy obviously talks about this a lot.

It takes a lot of samples to learn some very basic stuff because you only get a reward at the end. You don't actually understand things as they're happening. So, like, if you do RL for a game like chess, it's kind of pointless. This is not how humans would even come close to working. We sort of understand the incrementally good moves, and then we know we're closer or further from a result.

RL is not a perfect analogy, but it's like I only know at the end that I played a shitty game if I lost. So how can we get that improvement mechanism to be better? That's 1. The second thing is data. How do we get better, more interesting data, or learn from fewer samples? So that's the sort of second piece that's interesting.

And then third, I would say, is: Are there other ways to scale intelligence within test-time compute paradigms? That's sort of the agent stuff that we're seeing.

Mm-hmm.

Which is, I mean, I guess, obvious now, but not that obvious before: You can sort of chain these things together and make it call external tools, and that could be way more magical than just saying, "Hey, reason and come up with an answer." So those are a couple of things, and then there's more fundamental work on, say, memory and understanding.

When I say “understanding,” I mean reliability of models. For humans—I hate using the human analogy, but it’s easy to understand—if you ask a human a basic fact, if I ask you 1 + 1, you say 2. You don’t sometimes say 4. That’s a bad example, but there are many cases where models should be more deterministic if they understood the principle involved, but they’re not. They’re very undeterministic because they have no core structural understanding of things.

How do we fix that? That’s one. Context windows are another. People are working on expanding those. So there are just a bunch of these things that are pretty interesting and could unlock a lot.

I like Anthropic’s framing of the end goal. One of the nice end goals to have, which is quite cool, easy to measure, and interesting to follow, is the economic Turing test. Can I pay an AI X amount of money to do a task that I could pay a human to do as well, and I wouldn’t know the difference?

Mm-hmm.

That’s the economic Turing test, and can I drive that value of X higher and higher and higher? That’s an interesting framing, which I think is what the actual goal—one of the goals—might be here.

How does the evolution of this, and trying to think of what comes next, inform your investment decisions?

It’s a really tough one. This one I struggle with all the time, right? It’s so hard to predict the future of where these models go. I don’t think anybody, even at the labs, has a single opinion. People have different opinions, right?

A common opinion—I think this is the most sensible take in many ways—is, “Hey, we have this graph of the amount of hours of work that a model can do in some structure, and we see that graph kind of go up and to the right in some log-linear curve, or whatever.” Maybe that can keep happening. It’s been happening for a while. Hopefully, we can keep it going.

That’s one view, in which case, if you keep extending that, then at some point, a lot of the work that humans do is automatable. That’s one belief. The other belief is, “Yeah, this doesn’t really scale.”

There are some novel discoveries that AIs can do today, but they’re kind of pseudo-novel in the sense that if you really had a human go discover that and look at the data for that long, they’d probably discover it. It’s just not that interesting a type of discovery. The way to think about it is, if the branch of knowledge is like this, it’s kind of a twig on one side. It’s cool, but it’s not a branch.

How do we get to, “Hey, can we invent branches of knowledge?” Some people think we can never get there. I don’t know. The current techniques are not good enough to get there. So my TL;DR to that is I do not know the future. I do not know how to answer things like AGI 2027. People have so many views on this stuff. I just don’t know.

Mm-hmm.

I can only reason from the information I have, which is, look, there are some jobs that are really important that people can’t hire for, and the current techniques are getting pretty good at doing most of it, if not all of it already. So that’s a problem to be solved.

I don’t see a big lab coming in and solving that problem, so maybe invest in a company that does something like that. Or I talk about, say, the OpenRouter investment. It’s not rocket science. I’m like, well, there are many models. People want to use all of them. New ones come out. It’s kind of frustrating to find a new one and figure out how to integrate with it.

What if one API could give you access to all of them? It’s not rocket science-y at all, and it’s not the most insane novel technology. But we look at that, and I’m like, well, that’s obviously something we need. Why not make an investment there?

Mm-hmm.

Right? So that’s kind of how I would say I think about the investments. I just don’t have a big long-term view that I’m confident about.

Well, might I suggest you go to Kalshi and check out their prediction markets? They have one for the best model for the end of the year—

Yeah.

—the end of the month.

It’s Google.

You think it’s Google?

Well, that was—that’s what the market says.

That’s what the market says?

That’s what the market says.

Gemini.

Yeah.

8. AI Rewrites the Job Market

Yeah. Another one is jobs and tech layoffs. That one was really interesting. I just had Michael Barton from Coatue on a couple of weeks ago, and we were talking about that, and then, of course, cascading events, lots of layoffs. Amazon is planning on laying off up to 30,000 or so, but I think it came out at 16. This is more in their robotics division—

Mm-hmm.

—because they want to bring in robotics. But yeah, we’re going to see a shift in any sort of technological wave: new jobs, old jobs, things just changing, rescaling, all that kind of stuff. So maybe Coursera comes back.

I would love that. You know, I was an intern at Coursera—

Really?

—when I was a sophomore or a junior. I love that company.

Yeah.

Maybe it does.

Yeah.

Wait, I’m curious. What did Michael say? What was his take? I have my take already, but what was his take?

His take was there are 2 camps for this. There are people who believe that AI will create more efficiencies and fewer jobs.

Mm-hmm.

There’s another camp that says AI will make your job much, much better, and so you’ll hire more because you can have more firepower.

Mm-hmm.

So there are 2 things there. He also says—quote-unquote, don’t take this as investment advice or anything like that—but his other take is that layoffs are a positive indicator for the health of the business. We saw this with Facebook. This is something Brad Gerstner had talked about for a while. It’s like getting fit.

Hmm.

All these years, revenue and growth were going up 20%, and employee count was going up with that. But now we’re learning you don’t need to scale employee count with that. The businesses are becoming more efficient, so the result will be much better-looking companies with higher margins. So that’s kind of where it landed.

I’m smiling because I was going to say it’s surprising we’re realizing that now. I have this thing I wrote that went pretty viral. Again, this one was another viral one, which is—

A banger.

It was a banger. I said something that I thought was the most obvious thing for most engineers in the Valley, which is that most engineers don’t do shit. We know this.

Mm-hmm.

I know them, right?

Yeah.

We all know them.

Yes.

We know the jobs. We know the companies they work at.

Mm-hmm.

And we know how many trips to Hawaii a year they go on. So obviously, they don’t do anything, right? When people tell me, “I’m so surprised at the Meta layoffs or the Google layoffs,” or whatever company layoffs—

Yeah.

—I’m like, “Well, do you know the people who work there?”

I’ll tell you one other funny anecdote. In the early days at Glean, I remember I was talking to Arvind in the kitchen, and I was telling him, “One of the things I just love about Glean is I feel like I’m working so hard, and there are so many things on my mind. I’m so in the zone.” At Google, there were people I knew who just didn’t write code for an entire year, and they’re still there.

In classic Arvind style, he looks at me, he smiles, and he goes, “A year? I knew people who haven’t written code for 10 years, and they’re ICs.” That’s just Google. My point is not to pick on Google or anything, but this happens in plenty of other companies.

Mm-hmm.

My take on these companies is, sure, in the tech sense, I think they’re getting leaner and more efficient. Elon sort of proved this with Twitter.

Right.

There are many bloated companies.

Yeah.

There’s a power law of who gives value in terms of engineering. There are some people who do so much they keep the company alive in any good, successful, big company, and there’s a long tail of people who basically do nothing.

Mm-hmm.

So in the big tech sense, yeah, I agree, I guess, with Michael on all of his takes. But anyway, coming back to the actual question of AI and, I guess, job layoffs, I agree with him. I think there’s some section of jobs that have the pattern of: they will be augmented, and you can hire more.

We see that when we look at it. I always look at history with these things, right? There was a time when most companies didn't need—didn't have—a tech team. There was nothing to do in tech. Then we had software, and people were like, “Well, we clearly need a person to run—you know, whatever. Call it the website, call it a couple of other things.”

Then we had a search engine, so people were like, “Oh, shit, we should do some marketing. We need another marketing team to focus on search ads or whatever.” Technological shifts also create a lot of jobs. But at the same time, you could imagine one of those companies being like, “But we have a robot for the factory, so maybe fewer people in the factory.” This has happened through all big technological evolutions. There have been some new jobs created and a lot of old jobs that go away.

Broadly, I think no one wants to do the old jobs that go away, usually, and most people want to do the new jobs that come. Most people would argue that factory jobs are much harder and more annoying than a desk job where you write tweets. I think so.

Mm-hmm.

In this case, I think one of the fears of AI is: What is the calculus on that balance? Is it really bad in the sense that we're going to take away too many jobs for the amount of jobs that we add?

I don't think it really applies to tech. I know people say, “Hey, is AI taking tech jobs?” I think that's more of a laziness, zero-interest-rate phenomenon in getting the business to be more efficient. But then you look at the rest of the economy, and I'm like, “Oh, yeah, there are so many jobs where people don't really do anything, or whatever they do, an LLM could clearly do better.”

Then you have factory robotics, all these other kinds of Waymo-type AI companies. On net, the population is also declining, so in some sense, we don't need as many jobs in the long run. My prediction is, look, there's going to be some period of pain, as there always is with a new technology. Some jobs will go away. People who have those jobs will be very upset.

But in the long run, humanity finds a way, typically and historically, to find new things to do and new areas that they want to focus on, and that's where they're going to go. But in the middle, there is some pain. I'm sure there was a factory worker in the Industrial Revolution who was 35 with 3 kids and went, “Shit, now you have a robot. I hate you. I hate robots. I hate the guys who make robots.”

That's fair for him to think. Eventually, I'm sure he found something to do, or maybe he had a little bit of a struggle. So I imagine something similar will happen in the AI world.

And maybe less toxic fumes.

And maybe less toxic fumes.

Well, I want to ask this question. We went over what you look for in investments and all that kind of stuff, but I want to know what informs you, who is someone that you admire most, and what have you learned from them?

9. Arvind Satyanarayan's Founder Lessons

I think somebody I admire the most is probably Arvind Satyanarayan. I'll tell you the reasons I admire him. Number 1, I love the Japanese philosophy of respecting your craft. You do work because that's what people do. You don't do work to win. You don't do work to be number 1. You don't do work to have a high ego about what you do.

I think, in a weird way, amongst people I've worked with, Arvind absolutely embodies that. He told me something that made no sense at the time, and I thought, “Of course he would say that.” It was in the early days, and it stuck with me. He said, “Deedy, don't complain about hard work. Hard work is a gift, and if you wake up and you're doing the job that you do, and you can work hard, you don't know how lucky you have it. Most people can't do that.”

At the time, I thought, “You're the boss. Of course you would say that.” But he also works. Despite having a paper net worth of multibillions of dollars, he works more than almost anybody I know at that level. He has no joy in ego stuff, events, or podcasts. He hates them. I know he hates them.

Damn it.

He goes to them because he has to, because he's a CEO and the face of the company. He doesn't like that stuff. He doesn't buy flamboyant things. He drives shitty cars. He lives in a house that—we joke—you could have lived in 15 years ago in your career. He doesn't upgrade it.

He doesn't care about any of that stuff. He cares about the fact that he can work hard every day and do what he does. He just loves it so much. Another quick one on that: At some point, he told me—two other stories, because I'm going on an Arvind rant.

Mm-hmm.

At one point, we had a serious acquisition offer that we were considering. It was a big acquisition offer. We would have all made out very well. I remember we were driving back to San Francisco. This is the kind of guy Arvind is. He could have done anything, and he was like, “Deedy, will you drive me to the city? I would rather not spend extra money on an Uber or drive myself.” And I'm like, “Okay, boss. You got it.”

We were driving up to the city on 280, and I asked Arvind, “Man, real talk, how are you feeling right now? We have this acquisition offer. This works. You have validation that you can be out with 1 billion dollars right now, liquid, and you've made it. You've done a 2nd company that's a unicorn. Very few people have that. How do you feel?”

He looks at me, and I thought he was being pretty honest. Sometimes CEOs have to lie and just say what you want to hear—what you have to hear. But he said, “Deedy, I run such a cool company right now. If I get acquired, I will no longer run that cool company. Why would I want that?”

That was genuinely what was going through his mind. He was like, “From a personal level, why would I not want this job? I don't want to be some exec schmuck at some company that bought me. Who wants to be that?” And I'm like, wow. I don't know many people who could seriously have that thought, and I would believe that they truly believe it, except Arvind. I love him for that. That's one lesson.

Last lesson. I really like this one because I see this happen in so many other people. In the very early stages of Glean, I was talking to him about, “Hey, man, have we thought about, when we grow, what our defensibility is going to come from? What if this company decides to compete with us? Don't you think they have a pretty good advantage? Or right now we care about hiring, but are you thinking about ramping up the sales team?”

All of this stuff was always in my mind because I'd read enough startup shit on the internet that I had views. This is how I—

Crazy.

... like to see, like to think about it. And he looks at me, obviously having done this way more than I have, and says, “Deedy, you have a lot of questions. They're pretty good questions, but they're not the right question. At any given point, you have 1 question you're trying to answer, and you want to go get that answer. That's all that you need to make something work. Don't overthink it.

“The 1 question we have right now is not any of that shit. All we need to know is: Do customers love this product? Is the answer yes or no?”

I'm like, “No.” He's like, “That's all you need to work on. That's all. None of that stuff matters.” And I'm like, wow. The clarity you need to have to be that person is so hard, and I don't see that kind of clarity in most—forget founders, for sure—but most people.

Even in their day-to-day lives, I think we just jam our heads with shit. “Oh, my God, all these things and concerns and questions and anxieties.” There's only 1 or 2 things that really matter. Go do those 1 or 2 things. So I love him for that kind of advice.

My God. That is some wisdom. That's a great way to end it, too. Thank you so much, Deedy.

Thank you, Molly.

This was really fun.

Thank you for having me.