In a candid conversation with host Zubin Mirzadeh Khan, venture capitalist Vinod Khosla critiques the prevailing short‑term focus of many Silicon Valley founders and investors, urging entrepreneurs to build a durable, differentiated advantage that can survive rapid AI model advances. He argues that venture capital and board pressure often amplify short‑run revenue obsession, which can leave companies vulnerable when powerful AI models render existing features obsolete. Khosla distinguishes between incumbent SaaS firms and early‑stage YC startups, recommending incumbents leverage AI as a partner while startups should aim for radical, high‑risk bets that create new markets rather than incremental improvements. He emphasizes a “vision‑first, tactics‑later” mindset, using a Mount Everest analogy to illustrate the need for stepwise, goal‑driven progress. Khosla discusses the psychology of AI adoption in large enterprises, noting that outsourced services adopt AI more readily than internal teams that fear job loss, leading to high pilot failure rates. He categorizes coding‑AI companies into three groups: (1) AI‑built software engineers, (2) tools that augment engineers, and (3) no‑code platforms that enable non‑engineers to program. He predicts a future where each professional works alongside multiple AI “interns.” The conversation also touches on early AI‑driven drug discovery (Isomorphic’s first clinical trial slated for next year) and operational scaling, suggesting companies aim for $5‑10 million revenue per employee rather than traditional lower ratios. Throughout, Khosla shares personal work habits, his 70‑year‑old perspective, and his involvement in diverse domains such as fusion, public transit, AI, and drug discovery.
Chapters
today. So if you're ready, let me call up Zubin.
Please put your hands together. Zubin Mirzadeh Khan. Zubin. Thank you for doing this. Thank you. Yep. Take it away. All right. That was quite an intro. For time's sake, I will welcome my first guest on the stage. I cannot see him, but Vinod Khosla. Vinod. Um, I wanted to start with, um, maybe what do you think is the worst advice being given to entrepreneurs in Silicon Valley right now?
Um, I think too many people are focusing on the very short run. Uh, the worst advice I think people get is go start revenue without sort of even having a full plan. So you have to build a long-term plan, then, then a bunch of short-term tactics in the context of the long-term plan. I think a lot of the long-term planning is going away. So I like to joke, uh, whenever I see YC startup or a new batch, Before the batch is over, which is only 90 days, half the companies are already obsolete because the models came out with something new. And these— so I, I think focusing on what's your real differentiated advantage over the long term that can sustain in the face of what everybody else will be doing is probably the most important thing.
Do you think venture capital makes this problem worse? There's a lot of folks in venture capital right now that are rewarding companies for getting out of the gates really hot. There's a lot of talks of triple, triple, double, double is dead. Are we egging it on?
I think that's silly talk. Personally, you want to see what differentiated value you're going to create. And that's exactly the kind of thing I think is bad advice. It's very short-term focused. Not based on, you know, if you're a Grammarly and you know the models are coming and your revenue's gonna disappear, you have to do something different well in advance, not go for just the revenue.
Mm-hmm.
So I, I do think much more thoughtful about what your, uh, long-term plan is 'cause the models are getting incredibly powerful and there's a lot of short-term, I have this feature, the models don't do this. Well, they, they don't do this today, but tomorrow they will. So thinking through strategically where you want to be is very, very important. And it's missed by a lot of sort of short-term VC advice.
Mm-hmm. Maybe can I explore those two sets of folks? One is the incumbent companies, the Grammarlys of the world, and then the other is maybe the YC founders. On the incumbent side.
Mm-hmm.
If you're Grammarly or Salesforce or any SaaS vendor that has been around for a while, how do you react to the models getting so good? Like, aren't they running into the classic innovator's dilemma?
Um, I, I, I, I view it differently. The job of a tech startup is to start with a technology advantage that's significant. So it lets you build a large customer base. Once you have a large customer base, that becomes a sustainable advantage. So I always say get your foot in the door with technology, but have long-term differentiation with your customer base. And then the, the big models, even when they're good at— get good at whatever you're doing, will wanna work with you. You know, Anthropic or OpenAI are happy to, sell their models to any of the coding platforms, whether it's Cognition or Cursor or look at your favorites.
What's interesting to me, I'm curious to get your reaction, is my observation is if you go talk to a Fortune 500 CIO today, or if you talk to the CEOs of any of these big tech companies, they're quite on their front foot about this technology. Like they're not lagging, they're not laggards like they typically would be. Yeah, they're quite on the front foot. And so it is interesting to see that in this wave of a technology shift, all of the folks that you would traditionally expect to be late adopters are just as much pushing the pedal to the floor as—
Well, no matter what kind of company you are, whether you're a 100-year-old manufacturing company or a new tech software company, your board of directors is asking you, what are you doing in AI? Yes, everybody's forward leaning. I would say many of them are doing incremental things as opposed— this is the non-AI companies or the non-big tech companies. They're doing incremental things as opposed to radical things. And there's always room, especially with the technology as radical as AI, to reinvent a space. So I generally will prefer taking more risk and do something that will sustain over a long time than do something incremental where there'll be 6 players doing the same and the big models will be doing the same. So, you know, most startups have a trade-off between the, the risk you take And then if you minimize risk, you're going to minimize the consequences of success, and it's easily overtaken. I always prefer larger risk and larger consequences of success so you build a bigger, better company.
But if you're a founder and you're taking larger risk, the trade-off to that is that you could go to zero. And that's a very daunting idea relative to maybe being one of 6. You build a decent business, maybe you can sell it. Um, there's a trade-off.
Well, every founder should decide why they're in that business, right? If you want a small outcome, that's perfectly reasonable goal. Um, and there's money available for that. There's enough VC investors who want 2x or 3x their money, and that's— so as a founder, you should be clear what your goals are, and based on that, pick the VC who's gonna help you do that. And there's every style of VC, um, now there's so many of them.
Mm-hmm.
So, uh, I generally think the small risk often turns into the large risk of undifferentiated, and then you can't sustain the business. So you, you get a chance of losing everything either way. Mm-hmm. One, you may be taking a technology risk of whether you can do something hard. The other, you may not be taking large technology risk, but then you're taking a very large market risk.
Mm-hmm.
Uh, and you'll be undifferentiated. And then as a small company, you can't win against the big guys.
Um, when you're a founder, let's go to the, let's say the AIYC founder looking for their niche in the market. It strikes me as very difficult to be strategic about what market you're in, what the models are doing when you have 10,000 different things that you have to do, and really all you're thinking about is how do I stay alive right now?
I disagree with that. I think the best founders, even at the YC stage, are thinking about what's the long-term vision and then what are all the tactical things they can use to get started. There's a great article on our website, it was written by some professor at INSEAD, Vinod Khosla, of all places, who interviewed 400 founders. And he roughly says, founders tend to be very— and this is a phrase I use— be obstinate about your goals, but be really flexible on your tactics. At every small step, you gather resources for the next step. The analogy I give is if you're climbing Mount Everest, you don't go to the highest peak. You first go to base camp and then Camp 1, Camp 2, Camp 3. If you went for the highest revenue or the fastest revenue or the highest peak, uh, it doesn't lead to the top of Everest. Mm-hmm. So you have to be strategic about what paths, uh, people are taking. In fact, one of the things that bothers me is too many boards give advice to founders to chase revenue without thinking through what the right stepping stones are to get to Mount Everest.
You said business school of all places. Can you double-click on that? Why do you say of all places business school was where, where that article came from?
Um, it just, I wouldn't have expected an entrepreneurial article out of INSEAD in France, not a particularly, if it had come from Stanford Business School, I'd understand. Fair enough. Um, I'm biased. Fair enough.
Um, can I go back to the point that you were making about the board telling CEOs, go adopt AI? That, that, that has been happening for the last couple of years. Everybody's using Anthropic or ChatGPT on the consumer side personally. They see the magic of it. Then they obviously know that their companies need to use it.
Yeah.
And it strikes me that there is infinite money in budget coming out of these conversations where every big company wants to just use AI right now.
Yeah.
And if you were asking me, why are so many companies working today? It's because there's basically budget flying outta the sky because everybody is pushing down to use AI. I don't believe that is sustainable. I think at some point there'll probably be pressure to use AI thoughtfully in a smaller set of projects. But I'm curious, like, when does that party—
I happen to disagree. Yep. Um, here's the way you should think about it. Uh, if you're a large enterprise, take any large enterprise but not a tech Silicon Valley company, um, there's lots of things going on, pressure from the top to adapt AI and show the benefits of AI. What I find is when a service is already outsourced, very easy to adapt AI. So if you're BPO or IT services or others, you're going to see rapid adoption because nobody inside the corporation is threatened. If it's inside the corporation, they're saying, oh, my job may be at risk. If you're doing an AI accounting function, people, the accountants are like going to make it fail. So there's proactive efforts to sabotage the adoption of AI even though the orders from the top are coming down differently because they're afraid for their jobs. So one has to be really careful the psychology of how one approaches these, uh, situations. Uh, and that's why you see stories like the recent story, I think it was from MIT, that 90% of these pilots are failing.
Mm-hmm.
It's because of things like that. On the other hand, almost all the coding startups are working fairly well because you are just selling a programmer more productivity, the coding. Yeah. Yep. Uh, and, and, and they don't care very much that new junior programmers are not being hired.
Uh, so you think the coding companies are working because it's not replacing people, not because it's transformative necessarily?
Yeah. Today I, I think that's the reason. Hmm. Now I do think the bigger opportunity is to replace people to build software engineers. Uh, in fact, there's two really large— there's really three categories of coding companies. There's people who are building software engineers, people who are building tools to assist software engineers, and then people who are building coding for people who don't code at all, who don't know what SQL is or databases. Uh, you know, Replit, which we are investors in, is one of those. They don't, they don't sell to IT departments. They sell to a Salesforce, like have every salesperson become a programmer. And program their function. Um, and, uh, I was told like even at a company like Zillow, 250 people in the, uh, sales and marketing side are coding now when they'd never coded. Cognition, on the other hand, is going after billing full programmers. And I think that's long-term the most important business there is. Mm-hmm. Because you're not a tool to help. And tools— and Cursor's doing tools. I suspect Cursor will switch to building programmers. They will have to because that's where the larger markets are. But revenue ramp is much faster if you're building a tool somebody can easily adapt. But I don't want to get too detailed. I think the same is true whether you're doing financial account— AI accountants, AI HR people, AI salespeople. Uh, the same issues apply.
Do you, um, we have yet to see all these models have consumed all the data that basically exists in the world. Now we're making up data for these models to consume. Um, we're throwing infinite money at these models to continue to push the scaling laws to limits that we haven't seen before. When do we get, in your opinion, the true, call it early scientific breakthroughs, let's say like, uh, curing cancer or something like that? Like, how far away? I know Sam has talked about that being maybe some point in the future.
Yeah, just last night I was talking to Demis at a conference I was at from Open— DeepMind, and he said Isomorphic, their biology drug discovery startup, is going to have its first drugs in clinical trials by the end of next year. Now drug trials are a very long process, but you know, it's starting to happen.
If you were building an AI company today, how would you build it differently than if you were when you were building Sun?
Look, that was a long, long time ago.
Okay, then if you were building a company like 10 years ago.
Yeah.
10 years ago.
I would— so what is the big, big— if you're planning a large market, and growing fast. Previously, you had to learn how to manage and scale people. Management was super important. I was just talking about a company that has $100 million in revenue, ARR, and they have in their finance function one person at $100 million in revenue. It's the management problem. When you have very few employees becomes so much easier. I'd architect a company to have $5 million or $10 million per employee instead of $200,000 or $500,000 per employee. That's fundamentally, if I'm building a big company, that's what I'd architect towards.
When So I understand the point, but the companies that I see, so like let's take Bipple, Rubrik's coming on stage after this. Their headcount grew 30% from 2 years ago to today. If you ask like Parker Conrad from Rippling, his response to this would be, well, Jubin, the minute that your competitor starts to go and hire a huge go-to-market team, raises a bunch of money, capitalizes the business, and then hires all of these folks that you now don't have. That's the minute that you go and hire all of those people? Like, what's the—
Look, it's hard to generalize across all types of businesses. Each business is different. So I won't generalize, but I would— maybe a simpler way for everybody to think about it, whatever person you're hiring, whether it's an accountant, a salesperson, an engineer, think about for the next 5 years, a good model is every single professional in your company has 5 AI interns and they do the work and you supervise these AI interns. It's, it's like the perfect model to adapt AI. Every time you needed 10 engineers, have 3, and each, each of the 3 have 5 AI interns., and these 3 engineers will supervise them. That's the way I think for the next couple of years, because these AIs aren't perfect, is how one should think about it. If you're a doctor, you have 5 AI fresh graduates from Stanford Medical School, you're not gonna let them alone, let them do the work with patients, but you supervise their diagnosis or prescription or whatever. So it doesn't matter whether you're an accountant and, you know, the CPA firms are doing this, with AI accountants. This is why junior hiring has slowed down so much according to a recent report from Stanford. Eric Beaufort did a report. Fresh graduates are not being hired because people are thinking of better models. Now you're not going to day one come in and replace a senior accountant or a senior engineer or an architect. So I think that's a good model for people to think about.
When founders come to you struggling, what do they tend to come to you with? What are the struggles that folks tend to share with you?
Again, those are as varied as you can—
Are there themes?
Huh?
Are there themes? Are they market? Is it market? Is it company building?
Early stage, it's almost always a product-market fit problem. Solving the problem? Are we getting traction? The next step is always how do you scale and how do you hire the right people in at that stage? Because you don't have an experienced team, uh, but you have fast scaling after product market fit. It's almost, almost always people issues. That's what they struggle with. And then at some point it's organizational structure to scale. But it varies a lot by business. So it's one of the things I've learned is not to generalize.
Why are you like— how much longer do you want to work?
Only 25 years.
Like you still seem incredibly energetic and passionate about your work.
Yeah, I still work 80 hours a week easily, and I love what I do.
Do you actually work 80 hours a week?
Oh, I probably work more than 80 hours a week. Yeah. Look, I just turned 70. Warren Buffett is 95. I have a good role model.
So you want to do an— you actually want to do another one?
I love what I do. I'm working on fusion. I'm working on public transit. I'm working on AI. I'm working on drugs, drug discovery. I work across all these areas, and I have to keep up with the latest developments in each of these areas in addition to traditional areas like Rubrik, you know, pretty amazing. Bipples coming up, but we were just talking 10 years ago, we got involved with the company. The company, Bipple, I guess this must be public data anyway, has created 4,000 millionaires. Think of how many people in one company have been able to buy houses because of the company. So that's an introduction for Bipple.
