Elad Gil: Silicon Valley’s Most Dangerous Startup Advice

There's a bunch of conventional wisdom in Silicon Valley that I either think is always wrong or mostly wrong. An example of that is you always need a co-founder. Michael Dell didn't have a co-founder and Jeff Bezos didn't have a co-founder.

Your thing isn't working and you stick with it too long. There's radical openness to trying things that didn't exist three or four years ago. And that's really important in a way that I think few people really understand.

Wow. They keep inviting me back. I don't know why. So, thanks for keep on raising this big ass fund. So, we know needing to invite you back. So, you know, we've been here for three times.

Yeah, I think there's multiple ways to start a company. So I don't think there's like a single prescriptive approach or like a single right thing. I think fundamentally there's two or three different ways that you can approach things that are still pragmatically true. One is obviously being very customer centric. Sometimes is building for yourself as a customer. That's Braintrust, you know, Anker Goyal's company where it's an eval and prompt and you know etc tool and he built that for himself effectively as a potential user that's something he kept beating over and over. You can also do that for specific customers. So one approach to the world is doing that and being very customer-centric or honestly self-centric. Second approach which I think is really interesting in today's day and age which I think is very unique which is I'm starting to see people do these AI-driven rollups or buyouts where they're effectively buying companies and then expanding margin dramatically using AI. And there you need three skill sets. You need the ability to buy things. You need the ability to operationally change them, which is often the work part, the change in management. And then you need to do the AI stuff on top of that, so you can actually expand margins and make it a better business and make it more software-like. And then, you know, there's a third, which I know most of you is kind of a grab bag, which is you know, you iterate on different things, you invent things, you run it by customers, you know, you try customers. That's kind of almost a traditional approach and it's a little bit more of a winding road in some cases where you come up with an idea and then you try embed it and you know see what people think.

And so I think that part is dramatically faster and better and then I'd also say faster in two ways one is you can build things much faster which makes a big difference but the second way is customers are very interested in trying stuff in a way that I haven't seen in my lifetime and because of AI every CEO has the edict you need to do something in AI what are you going to tell your board about AI and how are you going to do AI and so there's radical openness to trying things that didn't exist three or four years ago and that's really important in a way that I think few people really understand. And so I remember when I was diligencing I invested in Harvey the AI legal company really early in their first round and then I led their series B and for Harvey I remember calling their customers as due diligence and I call these big law firms and I asked them and usually law firms by the way are really bad adopters of technology right they're very slow they're hesitant to change. There's a big security reviews like many legal laptops are completely locked down, right? You can't download things on them without like CIO permission at the law firm. So I call these people running these law firms, you know, the 10 biggest law firms in the world and I asked them what is the biggest change you expect the party and why you adopting it and all the rest. And even two or three years ago the insight was so interesting and unexpected where they said we think this will augment associates to the point where we'll be fewer of them for the same book of business where people can grow our book of business quite a bit and we grow partners from associates. So who are going to be our future partners? Are we going to have enough associates per partner to actually have the right number of partners? Like it's part of the training. It's part of choosing people. It's a selective criteria. It's the filter. It's like whatever you want to call it. And I thought that was fascinating as an insight and this is insight like 3 years ago or whatever it was 2 years ago. And so I do think there's other aspects that are really overlaid on this that are like super fascinating.

100%. And I take it a step further, which is even a couple years ago. So I think there's a bunch of conventional wisdom in Silicon Valley that I either think is always wrong or mostly wrong. And an example of that is you always need a co-founder and you look at the biggest market caps in the world and Michael Dell didn't have a co-founder and Jeff Bezos didn't have a co-founder and you know you go through company by company. Steve Jobs was a dominant founder. There was two founders but he was really the main one although wasn't of course built everything initially but you kind of accompanied my company and either it's unequal in most cases until YC right YC and Google was roughly equal though Larry had a little bit more stock even.

No co-founder right so you look at many of the big Bill Gates co-founded with somebody but then he really became the sole founder when Paul Allen left so over and whoever is either unequal or solo for many of the biggest things. Now there's lots of people who've been incredibly successful in some cases is equal in some cases unequal. If you actually look at cap tables as companies go public you actually realize that most sins were unequal and it was really YC that normalizes equality. So anyhow there's a lot of conventional wisdom that I just think is wrong.

Yeah. No I think winning is important for startups. People often ask me what is the single biggest determinant of culture in a startup and I say winning you know it's not the kombucha it's not the ping pong table it's you know not the number of GPUs you have although that is exciting it's winning but anyhow back to the original question the companies at least that I've been involved with that have tended to work have tended to work early and it's not like we're going to grind for 70 years and then the thing will work and it's usually something starts working really fast.

Right yeah this actually ties into competing schools of thought in Silicon Valley around are there enough founders in the world so the YC school of thought is there should if you just have more founders more magical things will happen and on the alternative viewpoint of the world and I think both are partially correct is there's only so many markets that are open in a given moment in time and therefore there's only so many markets in which you could build a big company and they're open because of a regulatory change they're open because of customer behavior they're open because of technology shift right right now we're going through a simultaneous massive shift in technology and a massive shift in buying behavior so many more markets are open right now than I've ever seen in my life.

There's a few there's also a lot of stuff that took off in ways that people didn't anticipate, right? Lope, but also ChatGPT. It was launched Thanksgiving week. This is true. is a buried launch November 2020. It was like oh it's nobody's gonna pay attention to the sea house launch it during Thanksgiving. So like what does that mean in terms of our ability to predict what's going to be compelling? So today we have three to four players let's call it three players who are kind of dominating the frontier and we'll say four there's obviously you know OpenAI there's Anthropic there's Google and then we'll put all the Chinese open source models into their own kind of bucket right so let's say there are four kind of big categories do you think that this is a stable play or do you anticipate this shifting call it in a five-year frame.

Actually sorry you're right I mean if you want the if you want the beans and what's happened. Yeah. As in Grok is the best of that.

And you know, so I wrote a post about this maybe three or four years ago where I tried to predict the foundation model market and at the time what I thought would happen was that you'd end up with three players each of which would be aligned with a hyperscaler and it'd be kind of a captive hyperscaler model relationship where the hyperscaler would fund it and then the key on that platform. And so I thought it'd be Google for itself with GCP and then OpenAI and Microsoft because it would be deep relationship at the time. Then I thought Anthropic maybe partners with Amazon or something. And then I thought Nvidia would be the primary funder of open source. And if you look at every technology wave visually a big company that funds and monetizes the open source. So during the 90s for Linux sells IBM where they spend billions of dollars effectively on Linux development and really monetized it by selling it through services. And I got it wrong. I got it partially right and partially wrong. I got it right in terms of thinking be a oligopoly market. And the reason I thought it would be a oligopoly because of the scale of capital needed to do it. So eventually you just don't have new entrance for people to compete. But I got it wrong in a few ways. One is Meta ended up being the US-based funder for the time being and China the Chinese government became the other major source of economic funding for open source models and then I got it wrong in terms of the models are partnering more aggressively across right and is partnered with GCP and Amazon Microsoft is working with multiple models is you know Google is working with Anthropic so I it's much more of a overlapping world than what I anticipated and everybody kind of crossed it's actually honestly closer to what happened in China across a few different waves of like the big Chinese companies effectively funding the versions of Uber and you know other sort of next-gen applications. So I do think the natural structure assuming one of the companies doesn't truly lift off in terms of model capabilities ahead of everybody else's oligopoly that seems like the natural state for now. I do think one interesting thing that's emerged is the importance of the harness relative to some of these applications like Claude Code. I think a big shift there that I observed was when o1 came out for OpenAI, there was less switching than I anticipated because some people argued it was a bit of a better model, but people had already been using Claude in a harness and was really optimized for using Claude and so fewer people switched than I expected would. It could be the model was only incrementally better or it could be a variety of thinking. So an open question to my mind is how much does the harness matter for stickiness now versus the core model and if that transition happens and that has really deep implications in terms of how you think about things and then the other piece of it that's the interesting question is when do the certain models asymptote and functionality and if you hit the asymptote then other aspects become more important in terms of switching or not if you don't hit the asymptote and it continues to be what's the best model again with this overlay is it significantly different enough that you'll abandon whatever harness you're And then the question is like what's the enterprise version of the harness?

Yeah. No, I think there's tons of room. I think the question is more what is the closeness between the models and what is the utility you get through things that ancillary and the ancillary things include the harness. I think includes brand actually.

Yeah. Yeah. No, seriously, that matters. So, I think it's a couple different things. You just get used to having things configured a certain way, different types of prompts or you know, everything is kind of set up for you. And I think it's a recent phenomena. I think it's very new and that's why I think it's not discussed very much as like a thing.

At this stage, you might as well just go work for a foundation model company, you know.

Yeah, we all get absorbed when they work on a railroad.

It's very durable. So I think there's a few different types of models, right? So there's the core language foundation model like the LLM, etc. That's mainly what we talk about. And I think two or three years ago, everybody thought, okay, this is going to go agentic over time. And for agentic things, you need some form of persistent memory and you need to have these sort of actions. You need to be able to tie into these different to integrate with these APIs, you know, like I think it was kind of clear what to do and it was just a matter of time to go do it, right? We knew reasoning was coming because we'd have definitely seen reasoning in other contexts and you know so I think a lot of it has been reasonably predictable and we knew like I remember talking to Eric Stein Burer who runs Magic and three years ago he was talking about super long context windows being incredibly important for code because then you could put the whole codebase in and you could work across it and you know so again I think a lot of the smartest people had all these ideas and announce and similarly I think there's a very clear road map in terms of what's to come for other areas I think there's lots of really cool stuff to do. So are you know what are the best models for physics simulation what are the best models for materials what are the you know you kind of go through whole by circle or area by area there's super interesting stuff to do now that may be severable for economic value.

Maybe two questions rolled up into one here. So I mean how I think that over the last 6 weeks there's probably more encroachment from the foundation models into a number of like different domains right like you know thought going out for profit thought for legal thought for finance and obviously of all the coding related things. So I guess the two questions here are like how do you as a startup if you could start companies how do you think about what is it in the effective blast radius of like what these models can do and then secondly I guess related to that is what are the durable modes that an application or a product can have and I think one of the things you said in the past which actually resonated with me a lot was that data is maybe overrated as a mode everybody talks about we have proprietary data and we can you know do better like context window stopping and that feels a little thin to me and an ultimate. So two different questions folded up into the same thing there.

It's great. We'll be all, you know, we'll be all at the edge of the cemetery because I got a little bit distracted.

Yeah, exact. So, I think if you look at durability, there's two ways you can be durable. And by the way, if you look, you know, the whole wrapper around a model thing, I don't really buy because if you look at a SaaS company, is this a wrapper around a SQL database and a lot of things that those things just bite, right? I think usually what you need is you need to build a multi-product company. So there's three theories. One is there's a system of record? You're basically where core data about a thing is stored in all its attributes. And so, you know, that's Workday for all the people working at your company or whatever. And so then you can build a dozen applications around it and it's very durable. That's one hypothesis in terms of what creates durability. I think the single biggest thing is just can you build a dozen different products that say you're cross-selling roughly to the same customer or user. They're deeply integrated and therefore you have multiple workflows and I think then you're very defensible. So for example, if Harvey has two dozen different workflows for different legal applications and use cases, that's defensible. If they're just, you know, bust a child with a legal document and not very defensible.

It's a lot more doable today because you can just crack that stuff out. Yeah. It's what's the surface area of your product? And I think it's very under discussed and it's often discussed as a revenue driver, which it is. And you know you have different wedges in that you can cross-sell against. You can cross-sell to the same account. It's hugely defensible, right? Because usually what happens is a founder will start a company, they'll build one product and they'll win because the product is 10x better and that'll get them distribution and then they keep going and they forget the fact that now that they're the incumbent and they have all these customers, they can build a product that's 80% as good as the best thing and still win because they're just...

They've worried on the security review, they've worried on the purchasing, they've gone through procurement, they've gone through everything legal. And so it's really easy to start cross-selling, in which case it becomes very defensible, right? It's really hard to rip out a dozen things instead of just one thing.

Well, Teams is such an interesting example of that, right? So yeah, it's actually an interesting example of what you said. Yeah.

So the early days of a startup, the primary thing you're competing with is other startups. And then incumbents often used to have five to seven years to react literally because they could just cross-sell to everybody. And so that's Slack and Zoom versus Teams, right? They were both like this and then they flattened out and then Slack had to sell. And the reason is because Microsoft bundled it in Teams, gave it to everybody and they cut off their growth and they won. And so in the olden days, like last decade, you had five to seven years of fighting other startups and then an incumbent would come in and if you could escape that incumbent, you won. Now maybe there's less time for the incumbent to react because you can iterate so quickly on code and the markets are open and everything else. But the incumbent can also build share way faster even if they're getting in their own way. And so there's this interesting dynamic that that timeline should shorten on both sides. So we'll see what happens there.

They're fast moving. They're also good companies. Yeah, look at companies and they have the distribution and traditionally, you know, probably the biggest headache for them is they in that five to seven years, and maybe they have two or three years, right? But there is time.

You know, I think there's two types, there's three types of failure mode. I think one type of failure mode is your thing isn't working and you stick with it too long. And it's back to like you should eventually iterate into something really working. And if it's taking two years and nothing's working, you should rethink. In most cases, not all. Again, there's counter examples. The second thing is there's really bad advice given to things that are working from people who've never had anything work. So what used to happen is you'd get advice from founders who'd seen product market fit to founders without it and they say go hire a sales team and scale really fast and burn a lot of money and do it, and that was terrible advice. But it goes the other way. If something's really working and people tell you to stay as lean as possible, don't hire executives, don't scale, that is awful advice. And that's where a lot of companies break for a while. And other people, competitors who are going to scale, come in and it's a very common pattern for things that are working that founders don't actually build out their company as a product and they don't really go for it in an aggressive enough manner. So, I think that's a big failure mode. And another failure mode is spending money on stuff that just doesn't matter. You know, you start training a crazy model on something instead of just testing something sheer with an existing like SaaS lab and just seeing if people want it, you know.