What I read and found interesting of late. The title ‘Interleavings’ comes from publishing. Interleaved books are those sold with blank pages inserted between printed pages so as to facilitate note-taking. This was a common practice in the 16th and 17th century, but has disappeared now. Thus my term ‘Interleavings’ for my notes and thoughts on the digital texts I enjoyed reading. Here are my interleavings from the recent past.
Podcasts
1/ Marc Andreessen doubts whether AI will disrupt early-stage VC
Link to podcast.
A little late to this podcast. Nothing special as such in the podcast, though this passage, on why he thinks early stage VC will not be significantly affected by AI, stood out. I quite agree. Early stage venture is all about identifying talent and picking people. It is also a place where in picking people you have to sometimes cast off past patterns / not be prisoners of patterns. My colleague Karthik said once: “Venture is a pattern-matching business, but the only stories worth telling are those built by exceptions”. Even if AI becomes the best pattern matcher, it will assign those weights and biases to past patterns. Yes, you can program it to reduce those weights and biases, and assign more importance to present context / state, but humans will likely for a long while, be the best real-time weighers of state, context, and past patterns all together at once.
“Marc: What I’m about to say may just be wishful thinking on my part. I might be the last Japanese soldier on the remote island in 1948 by saying what I’m about to say. I’m going to attempt fate. But I’m going to say, look, so much of what we do on the early side in the first five years is really very deep evaluation of individual people, and then it’s working with those people in very deep partnership. And this is one of the reasons, by the way, that venture doesn’t scale well, particularly venture doesn’t scale well geographically. The geographic scale experiments tend not to work. And the reason is just because you end up having to be face to face with the people for a long time, both during the evaluation process, but also during the building process. Because in the first five years, these companies generally aren’t on autopilot. You actually work with them a lot to help make sure that they do the things that they’re going to need to succeed. There’s a part of this that is very, very deep. Interpersonal relationships, conversations, interactions, coaching, by the way. We learn from them, they learn from us. It’s a lot of back and forth. We don’t come in with all the answers, but we have one lens because we see a panorama. They have another lens because they’re in the specific details a lot more than we are. And so there’s tremendous interpersonal interaction that happens. Tyler Cowen talks about this, I think he calls it project picking.
Certainly talent scouting would be another version of this, which is basically like, if you look back over hundreds of years for any new area of human endeavor, you almost always have this thing where you have very idiosyncratic people who are trying to do something new. And then there’s some professional support layer of the people who fund them and support them for the music industry. That’s David Geffen finding all the early folk artists and turning them into rock stars. Or it’s David O. Selznick finding the early movie actors and turning them into movie stars. Or it’s the guys sitting in a cafe, a tavern in Maine 500 years ago, figuring out which whaling captains are going to be able to go get the whale there. You know, it’s Queen Isabella getting the pitch from Christopher Columbus in the royal chambers and saying, yeah, that sounds plausible. Why not? There’s this alchemy that has developed over time between the people who do the new thing and then the people who sort of enable, support and fund those people. Let’s just say, like, there’s no guarantee that this continues, but that’s like a 400, 500 year endeavor, honestly. Probably it’s thousands of years old. You probably had tribal chieftains 2,000 years ago, 3,000 years ago, sitting around a fire, and the young warrior would come up and say, I want to go take a hunting party into this other thing and I want to see if there’s better game over there. And the chief sitting around the fire trying to figure out whether to say yes or no. So there’s something very human about that. My guess would be that that continues. Having said that, if I meet the algorithm that can do that better than I can, I will instantly retire. We’ll see what happens.”
2/ Roelof Botha on the two trackers they send out every week at Sequoia
Link to podcast.
Good podcast, though for regular Sequoia-watchers, some of the content is a repeat of what they have heard / read. To me, as a VC here in India, what stands out is Sequoia’s obsession to enhance themselves in every aspect of venture craft, from access (leveraging AI) to better picking by honing their decision-making through a) recognising cognitive biases + structuring their dealmaking process to factor these biases in, as well as b) combating these biases by using cognitive decision-aids such as their monday trackers etc.
Roelof: “We have an internal tracker here at Sequoia. We send it out every Monday as a reminder to be rational and level-headed as we look at late-stage private companies.
There’s about 690 public tech companies that we include in this basket. And we look at the current median multiple, enterprise value to revenue multiple, of this entire basket of companies. And right now, it’s sitting at the 60th percentile of the last two decades….in addition to that particular tracker, we have a sheet that we hand out that summarizes all the investments we’ve made to date in the current fund that we’re investing. And it’s a useful way to just reflect on what is the quality of the companies we’re assessing today. How does it measure up to the companies we’ve agreed to invest in?
And admittedly, there are mistakes in there and there’s some good decisions in there. But it’s a useful mechanism because humans are very good relative decision makers. Now, I’ve read some interesting behavioral economics research on this, where if I show you three different homes, and two of them are Spanish and one of them is Tudor, you’ll probably pick the nicer of the two Spanish homes just because you have a comparison.
If I showed you one Spanish and two Tudors, you might choose the nicer of the two Tudors. And so humans are just a little bit stuck in this relativism in how we make decisions. And so if you can widen the aperture of things that you put in that consideration set, I think it helps you make better decisions.
“Otherwise, you might just think about how good is this company relative to what else we’re looking at today or what else we’ve seen in the last month. You step back, you look at a wider set and it helps you really think about what quality means. Does this company have the potential to become a legendary company in the Sequoia language?”
More excerpts of what I found interesting here.
3/ Varun Mohan, Windsurf AI, on the changing role of the software engineer
Good read / listen for context-gathering around understanding how a AI-native high velocity startup founder thinks and executes (though I didn’t find anything dramatically new). His view on AI and coding in the light of GenAI’s rapid improvements in coding were interesting. Broadly 1/ Engineer’s role moving from coding to reviewing. 2/ Alpha in coding now moving to prioritisation / what to build (now what does this imply for PMs?!)
Varun: “One of the key pieces that we recognized was, with this new paradigm with AI, AI was probably going to write well over 90% of the software, in which case the role of a developer and what they’re doing in the IDE is maybe reviewing code. Maybe it’s actually a little bit different than what it was in the past.
…
The goal of AI has now changed a lot in that it is now modifying large chunks of code for you. And the job of a developer now is to actually review a lot of the code that the AI has generated.
…
I think when we think about what is an engineer actually doing, it probably falls into three buckets, right? What should I solve for? How should I solve it? And then solving it. I guess everyone who’s working in this space is probably increasingly convinced that solving it, which is just the pure, “I know how I’m going to do it” and just going and doing it, AI is going to handle vast majority, if not all of it.
In fact, it probably actually, with some of the work that we’ve done in terms of deeply understanding code bases, how should I solve it is also going to get closer and closer to getting done. If you deeply understand the environment inside an organization, if you deeply understand the code base, how you should solve it, given best practices when the company also gets solved.
So I think what engineering kind of goes to is actually what you wanted engineers to do in the first place, which is, what are the most important business problems that we do need to solve? What are the most important capabilities that we need our application, our product to have? And actually going and prioritizing those and actually going and making the right technical decisions to go out and doing it. And I think that’s where engineering is probably heading towards.”
4/ Robin Dunbar of Dunbar’s number fame, on the different relationship numbers that govern your life
Link to podcast.
Robin: “So you have these series of layers going out from you, which increased in size, but the quality of the relationship and the frequencies that you contact them gets smaller and smaller, smaller, lower and lower as you go further out. So these layers occur at very, very specific numbers. Core is really this 150 number, Dunbar’s number. But within that, there are a series of layers at 50, 15, 5 and the smallest one, 1.5.
And then beyond the 150 going out further, there are layers at 500, 1,500, the last possible one is 5,000 because 5,000 identifies the number of faces you can say whether you’ve seen them or not before. In other words, it’s the distinction between a complete stranger. I’ve never seen that face in my life and somebody you’ve seen but probably, for most of them, you don’t know much about. You don’t even know their name maybe.
But then the 1,500 layer within that, which actually turns out to be the typical size of tribes in hunter-gatherer societies. The distinction there is that you can put names to faces. As you come further in the 500 layers, the layer of acquaintances, many of the people you work with will be in that layer. In other words, you might go and have a beer with them after work. You might even spend a weekend away for some big work-related project with them, but you would never invite them back to your home for a big party.
Your 150 layer, well, that’s what I always call your bar mitzvah, weddings and funeral party. These are the people who will turn up out of obligation to you on that big once-in-a-lifetime event. You may not know they’re there in the third case, the funeral party, but believe me, they will be there. And then as you come in lower, so the relationships getting stronger and stronger.
The 15 layer is your sympathy group, people who you would feel extremely upset about if they die. That’s been known for a very long time. I didn’t discover that. And then inside that is this layer, five is the number’s layer, which is what we call the shoulder to cry on friends. And that layer turns out to be very important for you. It usually consists of two family members and two friends plus an another one per my design.
That number turns out to be the one that has the most influence on your general health and well-being. So the best predictor of your mental health and well-being, your physical health well-being, and even how long you’re going to live into the future from today is predicted by the number and quality of friendships you have in that layer. The five is an average. So if you only have three don’t panic yet, because introverts tend to prefer to have fewer people but have stronger friendships. As a result, extroverts tend to prefer slightly more people but have weaker friendships, those kind of effects. But on average, very much five. So as I’d like to say, if you really want to live forever, just make sure you’ve got five friends.”
Above diagram via Robin Dunbar, from his book ‘Friends’.
Patrick: And what about the 1.5. That seems like an odd number at the end of the cycle.
Robin: I always use to sort of come out with this joke when I was giving talks because I actually thought it was a joke, which was, well, just look at these series of circles and how very tight, the kind of fractal mathematical structure is. Each layer is three times the size of the layer inside it. And remember, these layers count cumulatively. So your 150 layer includes the 50 people in the layer inside. It’s not an addition. But just look at the regularity of these structuring, of these numbers, and surely, there’s a layer missing, if you project backwards.
Because you only knew about the five, we thought the five is the smallest number. If you project backwards, there’s a number missing and what is it? Well, it’s 1.5 and everybody would say, “How can you have 1.5 relationships?” The answer is obvious think about it. Half the population has one very, very close friend in that circle. And the other half of the population has two, what might the difference be? Well, the answer is gender. It’s obvious.
And the question is, why are women having two? And the answer is they have a platonic friend called the best friend forever. BFF. And that’s nearly always another woman, not 100%. 85% of women at any one time, have a best friend forever because the social psychologists who’ve explored this question have told us that constant reason that women always hear this because their romantic partner who statistically, obviously, is usually a male is completely useless at emotional-ish. You need somebody else who’s better qualified to deal with the emotional issues to sort of deal with that part of your life and the only person that falls into that little bracket as it were, is a girlfriend. The average is the male’s one and the women’s two that makes the 1.5 of that layer.
The layers are created by essentially the amount of time you spend doing stuff with different people. So obviously, the inner core layer, these layers in your social network have very specific frequencies of interaction you must have to keep the person in that layer. So the people in your five shoulder to cry on friends layer, you have to see at least once a week. That’s the minimum. The people in the 15, sympathy group layer next outside that you have to see at least once a month. If you see them less, those relationships decay inexorably.
It takes about 3 years for somebody to go from being a member of your 15 sympathy group to falling out of your 150 and joining the layer of acquaintances, which includes your favorite barista that you get your glass of coffee from every morning on the way to work and that kind of thing. So you have to invest time in relationships in proportion to essentially the emotional closeness. And it turns out that emotional closeness of a relationship is highly correlated with the time you invest in it.
Patrick: I’m so fascinated that I think in your research, you show 60% of people’s time they spend with maybe 40% with the closest five and another 20% with the next 10. Time it just strikes me is the resource. It’s the resource that we have to allocate thoughtfully and building those groups, time is the only way to do it, conversation and activity.
5/ David Senra on Jensen Huang’s whiteboard obsession
Link to podcast.
Enjoyed this episode by David Senra where he looks at Jensen Huang’s life through Tae Kim’s book ‘The Nvidia Way’. Lots of interesting snippets that give you a glimpse of how Jensen Huang thinks (a lot closer to Steve Jobs in a way) and acts. A good glimpse into high strategy, especially how to create the conditions and environment that sets you to win. Read ‘Three Teams, Two Seasons’ to get a glimpse of this. Oh, and fun fact: Jensen Huang has a favourite brand of marker that is sold only in Taiwan.
David: “Their first product, right, is a flop and it’s called the NV1. And so Jensen is actually doing like this postmortem, and he realized they made several critical mistakes with the NV1. So I’m going to read a section from the book to you. Some of these critical mistakes with the NV1 start from positioning to product strategy. He says they had over-designed the card, stuffing it with features no one cared about. The market simply wanted the fastest graphics performance for the best games at a decent price and nothing else. The NV1 could simply not stack up against other cards that were more narrowly designed.
This is what Jensen said: “We learned it was better to do fewer things well than to do too many things. Nobody goes to the store to buy a Swiss Army knife. It is something that you get for Christmas.” Says—this is what James Dyson said: “People do not want all-purpose. They want high-tech specificity.”
*
David: “Some people like Bezos wants a six-page memo. Jobs wants a demo. Jensen wants the whiteboard. So the whiteboard represents both possibility and ephemerality. I can’t pronounce that word, I’m sorry. The belief that a successful idea, no matter how brilliant, must eventually be erased and a new one must take its place. One of the reasons Jensen likes to use the whiteboard as the primary form of communication in meetings is everybody must demonstrate their thought process in real time in front of an audience. With the whiteboard, there is no hiding. I lost count how many times in the book they’re talking about how central, like, how important whiteboarding is to running NVIDIA.
And actually, one of the things that Tay Kim was gracious to send me, he sent me this Wall Street Journal write-up, a review of the book written by this guy named Ben Cohen. Ben has some great writing here in terms of like what he learned from reading the book and the importance of the Whiteboard. It says, “Instead of cloistering himself in a private office, Jensen prefers to work from conference rooms. He does his best thinking at the whiteboard, which he uses so frequently that he has a favorite brand of marker that is only sold in Taiwan.” And so even when he’s traveling, they have to travel with this specific brand of marker, and they have to travel with whiteboards.”
Link to excerpts from the podcast I found interesting.
Articles
6/ Lenny Rachitsky: To write consistently, write only what you are curious about
Link to article.
Lenny: “…Initially, my heuristic was 80% writing about what I’m energized/curious/excited to write about and 20% writing about what people want me to write about. These days, it’s actually 100% what I’m curious about. I now never publish anything only because I know it’ll do well.
This is actually a very important lesson I’ve learned: In this line of work, individual (viral) posts come and go, but it’s all about how long you can stick with it. One viral post: easy. One post every week for five years: much less easy. You need to prioritize stamina over anything else. In other words, play ‘infinite games’.”
7/ Elliot Jaques’s time horizon – strata fit
Link to article.
Beginning to see Elliot Jaques pop up off and on, most recently on the Roelof Botha episode with Mario Gabiele / The Generalist (see above). Here is an excerpt from an old but good piece on the social scientist. (He passed away in 2003)
Art Kleiner: “But if everyone agreed on the value of jobs, what did that value depend upon? Dr. Jaques was stumped until, one morning, three shop stewards burst in to tell him they’d figured it out. The critical difference had to do with time. Factory floor operators were paid by the hour, junior officers by the week, managers by the month, and executives by the year. Within two years, Dr. Jaques had refined this insight to the concept of “time span” — the value of every job could be measured by the length of time it took to carry out its longest-running assignment. (He also called it a “by-when,” his name for the explicit or implicit deadline embedded in every task.) A maintenance operator on a factory floor might wrap up all tasks within a 24-hour period, but a purchasing manager might need up to three months to finalize a contract, and a marketing VP might take two years to plan and implement the introduction of a new soap. The longer the time span, the greater the amount of “felt-fair pay” that was appropriate to earn.
In the early 1980s, he codified his findings. The true fit between a person and a job, he has concluded, depends on the match between the “time span” of the job and the potential capabilities of the person.
At the heart of the Jaques work is this double helix of human capability in organizations. On one side of the helix are the “categories” (as Dr. Jaques calls them) of people’s ability to handle cognitive complexity. Each of us is born with a certain potential ability to handle complexity. By the time we come of age (at, say, 18), if we’ve matured to that potential, then we can handle assignments of three months, a year, two years, five years, or more. This “time horizon” is more or less hardwired into us (not just in our minds, but in our beings, Dr. Jaques would say). Some people start out higher than others. On the bright side, we all continue to mature all our lives, making occasional palpable leaps in our ability about every 15 years, as we cross a threshold into the next level of capability. (If you realize that you can suddenly handle tasks that seemed unfathomable before, you’ve probably made such a leap recently.)
Of course, we may not fulfill our potential; we may be blocked by (for example) a physical accident like a stroke, the kind of emotional baggage that leads to neurotic self-destruction, a decision simply not to strive for success, or sheer lack of opportunity to develop our skills — which is one reason hierarchies are so important for many people.
That brings us to the other side of the double helix. Just like the time horizons (for people), the time spans (for jobs) break naturally, according to Dr. Jaques, into eight levels, which he calls “strata.” The fit between time-horizon levels and strata determines how comfortable we will feel at various positions in a hierarchy.”
8/ Dwarkesh Patel asks what is to AI as the motor car was to petroleum
Link to article.
Dwarkesh has been trying to sensemake developments in AI through his eponymous podcast. In this article, he lists down all the open questions in his mind. I didn’t really understand all of the points (truth!) he raised. Still it was very useful to get a glimpse of his mind and what he sees as the big unsolved questions. Re his query on the industrial-use case of AI (what is the motorcar for AI’s fuel?), I am reminded of Tim Harford’s book ‘Fifty Things That Shaped the Modern Economy’ and the chapter on the dynamo. He writes how the old factories were all based around steam power and needed a central driveshaft to relay steam power. It was only when newer factories / layouts replacing the driveshaft with wires to individual workstations was enabled that electricity’s power could be used by every worker. Today perhaps it is the other way around. AI can be accessed by individuals easily but it perhaps need group or org adoption for its power to be fully realised.
Dwarkesh:
“What is the industrial-scale use case of AI? Between 1859 (when Drake first discovered oil in Pennsylvania), to 1908 (when Henry Ford invented the modern automobile), the main use for crude was as kerosene for lighting. What is the ultimate industrial-scale equivalent use case for AI?
*
How transformative would the AIs of today (March 2025) be even if AI progress stopped here? I’ve personally become more bearish about the economic value of current systems after using them to build miniapps for my podcast: I can’t give them feedback that improves their scope or performance over time, and they can’t deal with unanticipated messy details. But then again, the first personal computers of the 1980s weren’t especially useful either. They were mostly used by a few hobbyists. They had anemic memories and processing power, and there just wasn’t a global network of applications to make them useful yet. Another way to ask this question: imagine that you plopped down a steam engine in a hamlet from 1500. What would they do with it? Nothing! You need complementary technologies. There weren’t perfect steam engine shaped holes in these hamlets; similarly, there aren’t many LLM-shaped holes in today’s world.”
9/ Jake Knapp + John Zeratsky on the Founding Hypothesis
Link to article.
Jake Knapp who came up with the Design Sprint methodology, teams up with Jake Zeratsky to write how founders can lay out your pick or approach into a framework they call a founding hypothesis, which asks them to list out the problem they are going after, the likely customer, the approach, and finally their differentiation (see image below). It is a handy way to structure the startup idea, and forces clarity early on. The article also has a few more bells and whistles, including the process of running a ‘Foundation Sprint’ to structure the pick better and arrive at a common language and buy-in for all founders. Certainly worth a read, especially for early stage founders.
10/ Bob Moore, Crossbeam, on picking the space to play in
Link to article.
Third-time founder Bob Moore covers why his older ventures failed, and why the last one has a fighting chance of success, in this article. Has a detailed section on how he and his cofounder arrived at the startup idea looking at both their affinity for the space, as well as ability to win in the space.
“First, he had to whittle down his list of roughly 100 miscellaneous entries — which ran the gamut from B2B SaaS tools to an escape room franchise.
Too often, founders — especially those earlier in their careers — fail to take their own strengths into account when choosing an idea, merely chasing market trends or their own passions. When Moore started RJMetrics, he and Stein were fresh off a two-year stint in venture capital, hungry for a startup of their own. But when they quit their investor day jobs to get RJMetrics off the ground, they didn’t have a strategic game plan beyond wanting in on the heating-up analytics Market. “We went in as mercenaries. We wanted in on this startup game and big data was increasingly a thing. I knew how to write a damn good SQL query. But that was all we had,” says Moore.
Looking back on it now, Moore thinks one of the contributing factors to RJMetrics’ shorter shelf-life was that the founding duo lacked a sense of where the market was headed. “When we started RJMetrics, I didn’t have strong conviction. It may have been something that lit up my brain, but I didn’t even know the word ‘dashboard’ until after I started the company. We didn’t know business intelligence as a market — we didn’t even know who Gartner was,” he says.
Years into building RJMetrics, this absence of a long-term vision became glaring in the product strategy. “Because we weren’t steering with a solid strategy, we pursued all these product micro-optimizations and landed ourselves in a place where we created a company that creates some value for some people. Our growth was stuck on a local maximum. RJMetrics became designed by committee. We didn’t have a core conviction that a certain specific thing ought to exist because that’s where the market was going. That wasn’t there,” says Moore.
So when Moore was assessing ideas for his third startup, he knew he needed to be honest with himself about not only what he was excited to build, but what he was uniquely positioned to build well, given his chops as a multi-time SaaS founder. To do that, he devised a simple founder-market fit pre-screen — a mental matrix with two dimensions of interest and ability:
• Intellectual interest in the problem. How much fun would I have doing this? How much would it light up my Brain?
• Founder aptitude. What is it about my specific experience that makes me predisposed to being good at solving this problem?
“This helped Moore quickly rule out the oddball ideas and pipe dreams. “This exercise took my list of 100 down to 10 or 15 ideas,” he says. With the remaining handful of ideas that landed in the top right quadrant of his mental matrix, Moore wanted to get some outside opinions. But he chose to do things differently for the validation step: He bypassed the classic customer-driven discovery model and instead ran his remaining ideas past fellow founders.
Moore felt that founders could offer more nuanced perspectives — and widen any preconceived notions he might have had about each of these ideas. “Founders are special. They need to develop an extremely high level of empathy and understanding for the needs of people across multiple personas, and also understand a baseline grasp on how markets evolve over time and what makes for something that’s more durable and versatile,” he says.
With the founders, I wasn’t asking, ‘Would you buy this?’ But instead, ‘Would you start this company? And what are the things that you think you might bump into?’ “Rather than narrowing these ideas down into the scope of how a persona would use it, I was able to broaden my horizons of what they could become,” says Moore. “I wasn’t interested in having my next thing be anything other than an IPO-scale business — I wasn’t optimizing for a ‘build it for a few years and sell’ situation. I thought I had one more startup in me and I wanted to see how far I could take it.”