Keith recently rejoined KV and significantly changed his investment strategy. Before returning to the firm two years ago, he had not invested in any AI companies. Today, AI represents about 70 percent of his total investments.

Before I joined KV, I had invested in zero AI companies. Since then, in the last two years, I'd say it's about 70 percent of my investments are AI. Had I not rejoined KV, I think I would either miss the whole wave and been completely irrelevant or been reckless.

Keith credits his return to the firm with helping him navigate this technological shift successfully. He suggests that without being in that specific environment, he might have missed the AI wave entirely or made poor investment choices during the transition.

Keith and Vinod began their direct working relationship when Vinod joined the board of Square. Before that, they had an indirect connection through the company Slide. Keith would hear about Vinod's grand theories and ideas for reorienting businesses through others. Another partner, David, also worked with Keith at Slide. David was known for his blunt feedback. He once told Keith that his first revenue model was mediocre, which turned out to be an accurate assessment.

Keith learned a fundamental principle during his time at Square that influenced his approach to venture capital and building businesses.

The most important precept is that the team you build is the company you build.

This philosophy made joining Khosla Ventures a natural fit for Keith. He saw how Vinod's style of contribution resonated with his own way of thinking. This alignment made them a strong pairing when Keith decided to transition into venture capital.

Vinod believes that first principles thinking is the essential ingredient for a productive working relationship. This method allows partners to identify exactly where they agree or disagree without relying on vague arguments. When a decision is broken down into its core factors, the debate becomes a logical exercise. This approach often reveals that a choice depends simply on whether a specific premise is true or false.

If you can do first principles thinking, it is easy to know where you agree and where you disagree. It is not this hand wavy thing. We can debate those factors and it has worked out very smoothly.

Keith recalls a time last summer when he was struggling to make a decision on a specific investment. The choice felt close to the line until Vinod isolated the three specific attributes of a founder that mattered most for that particular business. Once those variables were identified, the decision became simple. By focusing only on the factors that truly move the needle, a complex problem can be solved with clarity.

Direct communication is a core part of the partnership between Vinod and Keith. They prefer brutal honesty over polite hypocrisy because it saves time and prevents people from guessing what others are thinking. This clear and succinct style also works well with ambitious founders who appreciate directness.

I prefer brutal honesty to hypocritical politeness. Being very direct saves a lot of hassle. Once you have that culture, nobody is guessing at what you think.

They spend very little time on firm operations or management, likely less than five percent of their interactions. Instead, they focus their energy on investing in the future and finding people who want to change the world. Management is often a distraction from these more exciting activities. They rarely have policy disagreements because their goals are closely aligned around the work of investing rather than administration.

Senior partners at Khosla Ventures spend significantly less time with Limited Partners than is common at other firms. This approach is consistent across the senior leadership team. Rather than focusing on fundraising or investor relations, they prioritize working directly with founders and building companies.

I spend less time with LPs than almost any other senior partner. And I think that's generally true of all the senior partners at Khosla.

The decision to limit time spent with investors allows for deeper engagement with entrepreneurs. This preference is driven by the belief that working with those building new things is more productive. It is also simply more enjoyable to spend time with founders than on administrative investor tasks.

Every Monday meeting begins with the existing portfolio before looking at new opportunities. The goal is to be a partner for many years to help a company reach its highest potential. In forty years of venture capital, Vinod has never called himself an investor. He views his role as a venture assistant to entrepreneurs. The focus is on helping a company change its trajectory whenever possible.

I've not once called myself a venture capitalist or an investor. I always say I'm a venture assistant to entrepreneurs trying to build companies.

Advising entrepreneurs requires earning that right by actually building companies. Many people who offer advice have never been inside a company. They lack the empathy required to lead one. A common pitfall for firms is being too nice to founders. This is like always saying yes to your children. While it feels good, it fails to push them to be their best. This hypocritical politeness is pervasive in the industry and can be very damaging. Strong founders do not want politeness. They select for the best feedback they can get and know when to disagree.

There is a distinct difference between being proactive and reactive. Keith views his role as a consigliere who helps the principal and offers direct feedback. Other firms provide capital and then get out of the way. Both models aim to back bold ideas, but the level of involvement varies significantly.

Our craft is to be the partner in building the company. I look at my role as being the consigliere to the founder. Sometimes the consigliere tells you that's a bad idea and sometimes they tell you that's a great idea.

Even the most successful founders benefit from having an advisor. Professional athletes have shooting coaches or fitness coaches. Take someone like Max Levchin. Vinod was the first investor in his company. Even after the company went public, they still have quarterly phone calls. Max wants that second opinion and the pressure to think harder about things he might be ignoring. Great entrepreneurs use mentors to prod them. This helps ensure they are not ignoring critical parts of their business.

Keith looks for founders who excel in one of two specific ways. The first is finding someone who ranks in the top one basis point on a single dimension. They might be the smartest person or the most tenacious person. The goal is to identify a founder with a non-zero chance of changing an entire industry or even the world. Most people will never do that. When you meet someone who might, their specific talent stands out immediately.

One of two traits. Either I meet a founder and on some dimension they are the best I have ever met in my life. It can be different. They can be the smartest person, they can be the most tenacious person, they can be the best assessor of people, they can be the most strategic. It is just like top one basis point on some dimension.

The second indicator is a rare overlap of different skills that usually do not go together. Keith mentions Max Levchin as a prime example of someone who is both a first-rate technologist and a first-rate business mind. This combination is extremely rare. Jack Dorsey is another example, possessing a high-level design sense, technical ability, and business strategy. These rare combinations often lead to exceptional success.

Identifying elite talent often happens remarkably fast. While high intelligence usually becomes apparent within the first few minutes of meeting someone, other essential traits like grit require looking for specific evidence in a person's history. These traits are best revealed through stories that show how someone handles pressure or obstacles.

One of my favorite gritty founders told me the story of when he was working at Uber. His team was going to run a triathlon on Saturday. He wanted to be part of this team and fit in, even though he had not trained at all and did not own a bike. He rented a city bike and completed the triathlon. That shows so much grit. That is all you need to hear.

Another example of this determination involves a founder who passed out from stress during a high school spelling bee. Instead of quitting due to the embarrassment or the physical toll, he returned the following year to try and win. These specific dimensions of character are rare and provide a much clearer picture of a founder's potential than generic interview questions.

Finding the right founder is about identifying exceptionality in at least one dimension. It does not matter if a founder is incomplete or lacks experience in specific areas like management or marketing. These gaps can be filled with venture assistance as the company grows. The more critical factor is the founder's learning rate and their ability to process information. A founder must be open minded to new ideas while remaining disciplined enough to reject bad ones.

I often take positions I don't believe in just to test how the founder's thinking about something. If a person listens to me all the time, I will almost never invest with them. I know they are not critically examining.

Interviewing should be freeform rather than mechanical. Standard interview questions often lead to rehearsed stories that mask a candidate's true abilities. To find the best people, you must put them in situations they have not encountered before. The goal is to see past the polished storytelling to understand how they would actually solve problems in a new context.

This philosophy also applies to the early team members a founder hires. Many boards make the mistake of looking for candidates who look good on LinkedIn or come from large, established companies like Cisco. However, the skills required to sustain a massive brand are entirely different from the skills needed to create one from nothing. A startup needs people who can handle the zero to one phase of growth rather than those who have spent a decade managing incremental changes.

If you have been marketing at Cisco, you know nothing about creating a brand. You know how to incrementally sustain a brand. That is a very different skill set.

Venture partners can provide the most value by giving young founders the confidence to trust their own instincts. Often, a founder might feel pressure from a board to make a safe, conventional hire that does not actually fit the startup's needs. Having an experienced investor tell them they are right to say no can be the permission they need to stay true to their vision.

Evaluating a person's rate of growth is difficult during a first meeting. A single meeting provides only one data point. In geometry, an infinite number of lines can pass through one dot. To understand where someone is going, you need to see their trajectory over time. One way to do this is to look at their learning rate over a specific period. For founders, three months of history is often enough to see if they are learning quickly.

For YC founders, the most important question I can ask the partner working with the company is how much they have learned in the last three months. Three months is enough to tell if they have a high learning rate or not.

When you cannot see someone's past, you can test how they think through unfamiliar problems. This reveals their natural problem solving process. Putting a person out of their usual context makes it easy to see how they approach a challenge. For example, ask a candidate how they would use a small amount of seed funding to investigate three new ideas. Their choice of ideas and their plan for evaluating them over six months will reveal their true potential.

If I gave you a seed amount of funding to investigate three ideas, which three would you pick? How would you go about evaluating them over the next six months? It is a pretty simple test. You can tell a lot about a person just based on how they answer that question.

Ethics is a non-negotiable trait for any founder. While many skills can be learned or supplemented, a lack of integrity is a fundamental flaw. Finding out if someone is ethical usually requires checking references. Often, founders who seem unfriendly or highly disagreeable are actually the ones with the strongest moral compasses. Their intensity and willingness to disagree often stem from a deep commitment to their principles.

A lot of the highly disagreeable, intense founders actually have a very strong moral compass because they believe in their principles.

In contrast, some people in Silicon Valley shift their political views whenever it is convenient. This lack of consistency suggests they do not stay true to what they really believe.

The ability to recruit is a vital dimension for any founder. The first ten employees are critical because they replicate themselves and set the standard for the next hundred. Even if a founder is uneven in some areas, they must be able to convince people to chase their ambition. Success requires telling a story that convinces investors, customers, and employees to join a journey before anything is proven.

Your first 10 are going to multiply by 10. You have to convince people that you haven't proven almost anything, and you've got to convince people to come along on the journey with you.

Keith recalls meeting Brian Chesky and knowing within minutes that the Airbnb team was special. Similarly, Vinod describes the decision to invest in OpenAI as a unique case based on talent density rather than a business plan. At the time, OpenAI was a nonprofit with no revenue or product plan. It was such an outlier that the firm sent an apology letter to its investors explaining the massive initial investment.

This was the only critical density of research grade people that could possibly pull off AI. That was the investment hypothesis. It's the only time in the 20 year history of Khosla Ventures we sent an apology letter to our LPs when we made the investment.

The success of the OpenAI investment was also rooted in long-term relationships. Knowing Sam for a sustained period allowed the investors to see traits and potential that were not obvious to the broader market. This combination of talent density and personal trust superseded the need for a traditional corporate structure or revenue model.

Seed stage investing often follows a predictable pattern. Hot deals usually form around people with impressive backgrounds rather than the strength of an idea. For example, if talented employees leave a high-profile startup to start their own company, they will likely see a lot of investor interest immediately. However, these consensus bets rarely outperform the true outliers.

At the seed level, consensus hotbeds are generally around people, not around the idea. If you have an engineer and a designer leave a hot company tomorrow, that is going to go at a high price. But a seed investment around undiscovered people, the consensus ones aren't going to outperform the outliers at all.

The biggest returns often come from sectors that the broader market ignores. When Rocket Lab was starting out, almost no one wanted to invest in space. Vinod was able to buy a third of the company for five million dollars, and it later became a massive success. The same was true for OpenAI and Commonwealth Fusion. At the time, these investments did not make sense to the rest of the venture community because the fields were not yet popular.

Rocket Lab was one of the best investments anywhere. I don't think they would have raised money from anybody. Nobody wanted to invest in space. We bought a third of the company for 5 million dollars. It was the same for OpenAI. You were the only venture investor in 2018 that committed because it didn't make sense.

There is a massive opportunity to build AI workers that handle entire professions rather than just assisting humans. Instead of focusing on copilots where people might get in the way, the focus has shifted toward autonomous systems. There are dozens of startups now building AI oncologists, mental health therapists, chip designers, and structural engineers. The goal is for the technology to fully do the work.

As many professions as there are, there is that many opportunities basically to fully do the work.

This strategy avoids the limitations of the copilot model. While most early AI tools were designed to help humans, the more powerful approach is to let the AI operate independently. This focus has led to a portfolio of companies that aim to replace specific professional workflows entirely.

Loyalty is also a core principle when working with these founders. This is not just a business decision but an ethical one. When leadership crises occur, the role of an investor is to stand by the entrepreneur. If there is a disagreement, it should be handled privately. Public support remains a priority when you believe in the principles of the person you funded.

This is how it should be. It is about ethics. And if I disagree with the entrepreneur, then my job is to sit down, tell them why and agree that we agree or disagree before taking the right position publicly.

Exploring artificial intelligence requires looking beyond the standard transformer models used by large labs. There are many promising avenues, including category theory in mathematics, neurosymbolic techniques, and diffusion models. One of the most significant frontiers is the development of real world models that understand the physical world. Vinod has no doubt that these systems will eventually be embodied in robots and function with high precision.

I have zero doubt it will work and work increasingly well. It will be embodied. Robots will understand the physical world.

A critical component missing from many current models is intuition. Vinod describes an experiment using gaming data to train an AI on general intuition. When shown half of a video featuring soldiers in combat, the AI was able to predict their movements by replicating the human intuition required to escape an attacker. This level of understanding goes deeper than just processing text.

While many startups focus on general customer support, they often fail to address the problem of hallucinations. For industries like banking, insurance, or mental health, even a small percentage of false information is unacceptable. Success in these sectors requires architecting systems specifically for zero hallucination rather than just trying to incrementally improve a standard large language model.

If you are a bank, if you are a Visa or MasterCard or insurance, you cannot afford to hallucinate an answer, even if it is a low percentage. The winning customer support thing will be something that does not hallucinate for a class of applications.

Keith explains that he had zero AI investments before rejoining Khosla Ventures. In just two years, AI has come to represent about 70 percent of his new deals. He believes that if he had not returned to the firm, he would have either missed the AI wave or been reckless in his approach. By rejoining, he was able to learn through osmosis during partner meetings where most of the discussed companies were AI-based.

Had I not rejoined KV, I think I either missed the whole wave and been completely irrelevant or been reckless. Neither one's good because what I was able to do when I joined is learn by osmosis.

When Keith first started making AI investments, he relied on partners like Vinod to provide air cover. They helped him understand the quality of the technical teams and how differentiated their approaches were. This support allowed him to develop a taste for what works in the sector. He now learns how these companies are built by joining their boards and seeing their specific challenges firsthand.

AI companies are growing at unprecedented rates. This shift is like the four minute mile. Once one company proves that rapid growth is possible, it sets a new standard for everyone else. In the past, reaching $10 million in revenue during the first year seemed impossible. Now, it is a goal that founders must seriously consider. This speed forces leaders to rethink every part of how they build a company.

Once you see someone runs the four minute mile, then no company should have an excuse for not growing rapidly. You see all these enterprise companies going from 0 to 50 million, you start asking questions. What are the limiting factors? You start with the question of why not versus thinking it is impossible.

Standard roles like product management do not always fit this new model. Traditional product managers talk to customers and build 12 month roadmaps. However, AI technology moves too fast for a long term plan. New research papers and capabilities are released every month. This means companies must be more flexible. Some organizations even pair their research teams directly with customer acquisition teams to stay aligned. This is a major departure from how tech companies were built in the past.

The idea of PMs does not make sense in a rapidly emerging technology field. What do PMs do? They go talk to customers and they create a sequential roadmap over the next 12 months. Well, if the field's evolving and the capabilities are evolving every month, you can't have a 12 month roadmap. That makes no sense.

Hiring and compensation also require a new approach. Big tech companies pay research talent very high salaries, similar to professional athletes. Startups cannot always match this cash. Instead, they must find people with a missionary zeal who care more about the work than short term pay. They might also look for undervalued talent or "diamonds in the rough." Building an AI company from scratch requires a different financial plan to handle these unique talent needs.

Building an AI-centric company requires a fundamental shift in how teams are structured and how success is measured. Traditional systems of record, like ERP software, are becoming unbundled into specific modules like finance or procurement. In the past, these systems were judged by their feature lists. Today, the focus has shifted toward reducing the human headcount required to run these functions. For instance, a small business lending company named Slash reached 150 million dollars in annual recurring revenue with only a single person in their accounting department because they used an agentic architecture rather than a traditional one.

The benefit is very different than a feature list. If you don't have the right substrate to have it operate under an agentic architecture, you're not going to have huge success.

The traditional moats that protected older software companies are also evaporating. Defensibility used to come from building hundreds of complex integrations, which created a significant barrier to entry. New AI tools can now perform those same integrations in a fraction of the time at a much lower cost, making incumbents like Mulesoft far more vulnerable. This change forces even successful companies to rethink their foundations. Keith points out that even high-growth companies like Ramp, which started just before the AI boom, must constantly evaluate which parts of their original playbook should be discarded to remain competitive over the next decade.

This environment prizes rapid learners over individuals with long resumes. Vinod notes that many industry experts are actually experts in a version of the world that no longer exists. Because every function from computer architecture to marketing is being reinvented, the ability to learn quickly is more important than historical experience. This shift even affects how companies sell software. While top-down sales were once seen as a difficult way to build a company, many CEOs now feel intense pressure from their boards to adopt AI. This has created a unique market pull where executives are eager to buy AI solutions even before the long-term impact is fully proven.

Success in financial services has been a consistent theme throughout the history of the fund. Every fund managed has returned its total value through a single financial services investment. This track record includes being early investors in major players like Square, Affirm, and Upstart. While they were second investors in Stripe, the sector remains a primary driver of their overall performance.

In every single fund we've had we've returned the fund solely on one. Just on one financial services investment.

Keith mentions Aven as a company that could be the next major success on the level of Stripe. Although the firm is heavily invested in AI and robotics, their excellence in financial services remains a defining characteristic of their investment strategy.

The finance sector has been slower to adopt AI strategically than other industries. Hallucination concerns often make companies hesitant to fully integrate the technology. Despite this, some firms are using AI to radically change their efficiency. Keith points to Figure as a prime example. They use AI to process home equity lines of credit in a single hour, whereas the traditional process takes weeks. This allows customers to consolidate debt onto new cards with much lower interest rates.

AI is what makes it possible to go so fast, which makes it fit in the right way into the process.

While companies like Ramp are also using AI aggressively, iconic fintech companies are rare in the United States. Truly massive fintech firms only emerge every two or three years. Keith expects the next generation of these companies to rely on AI in a very significant way. Beyond finance, there is also a strong focus on sustainability and energy, which remain high growth areas for investment.

AI is set to change the paradigm of manufacturing. By applying AI to the complex processes typically handled by thousands of engineers, companies can reduce labor costs significantly. This shift makes it possible to bring manufacturing back onshore. While robotics plays a role, the real advantage comes from using AI to run the entire system more efficiently.

An iPhone assembly line would have a few thousand manufacturing engineers. If you can do that function with AI, then you have a pretty large advantage manufacturing onshore.

Beyond the factory floor, there is a massive opportunity to replace existing supply chain software with AI-driven solutions. In the defense sector, the need for innovation is driven by global competition. Supersonic aircraft are becoming essential because adversaries are already using them. Vinod notes that early investment in these areas was once rare, but more venture capitalists are now following the trend as certain companies succeed.

Keith highlights the geopolitical necessity of these technologies. The country faces threats from adversaries and must maintain a technological edge to preserve its way of life. Technology serves as a magic wand that allows the government to achieve more with fewer resources. The government needs to embrace these advancements to thrive in a shifting global order.

The country needs to take advantage of the best and brightest in technology and the cutting edge technology or we are going to sacrifice our way of living to people who do.

Engaging in political discourse on social media often stems from a sense of responsibility. Success in technology can lead to a large following. This provides a platform that can be used for more than just business. It is about using that influence to share important ideas or challenge bad ones while the opportunity exists.

I do not want to die one day in regret that I did not use my audience to proselytize about ideas and things that I find important. If I can surface new ideas or rebut bad ideas, I want to finish my life thinking I did the best I could to have influence.

The drive to participate in these debates comes from wanting to avoid the regret of staying silent. Having a platform creates an opportunity to change opinions and shape the conversation around topics that matter.

Engaging in public debates on X requires a balance between intuition and evidence. Keith notes that while he often knows the facts of a situation from his past involvement in politics, his day job limits the time he can spend on research. He simply states what he thinks without feeling personal stress from the interactions.

There are times when I happen to know the answer. I used to be very involved in politics. I know a lot of details, but I do wish sometimes that if I wasn't doing a real job, I would be spending more care in marshaling more evidence and probably be more effective.

Constructing a powerful argument takes significant effort. Some individuals even hire assistants to help with research for their online presence. Without that extra support, it is difficult to be as effective as one might want in public discourse.

Keith once followed a strategy of responding to bad ideas on the internet to ensure a record of the truth existed. This habit started at Square, where he and his team dealt with a consistent critic. Eventually, this turned into an internal meme. Keith suggests that if no one responds to a false claim, people and even AI models might start to believe the misinformation is true.

I am going to combat every bad idea on the internet with the worst idea. If someone puts a bad idea or something wrong and nobody responds, people think it is true. LLMs are going to pick it up and think it is true.

In contrast, Vinod spends very little time on social media, usually less than an hour a week. He only engages when he sees a significant violation of principles or blatant dishonesty. For example, he recently criticized a prominent investor for supporting price controls on interest rates. Vinod believes that people should speak up when they see bad ideas rather than being dishonest for the sake of political favor.

Keith used Twitter as a business tool during his time at Square. He read every tweet about the company to find customer complaints or success stories. This allowed him to route issues directly to the right people. However, he admits this level of immersion can become a difficult habit to break once it is no longer necessary for the business.

The conversation turns to political integrity and the danger of changing views for convenience. Vinod describes himself as an independent who left the Republican party over climate change. He argues that many leaders change their political affiliations based on who is in power. This lack of core values is a significant problem in modern leadership.

Doing things for convenience is a bad idea. Now I realize some CEOs have responsibilities to others than themselves, but some will change political affiliations and if the next president is Bernie Sanders, they will become liberal again. That I hate.

Tech has historically operated far from the influence and scrutiny of Washington. This distance allowed the industry to invent itself and thrive without early interference. However, the relationship between technology and politics is shifting. Many leaders now face the responsibility of representing customers and employees, especially those involved in government contracting. This new reality makes it impossible for tech to remain politically neutral or isolated.

Regulating an emerging technology like AI is particularly dangerous. The government often lacks the accuracy to regulate something that is still rapidly evolving. There is a concern that excessive or state-level regulation could hinder progress while global competitors move forward. Vinod argues that the United States is currently in a techno-economic battle with China. He believes it is essential to prioritize winning this race to maintain economic superiority.

Economic superiority is up for grabs and we gotta win or we'll live under President Xi's rules.

A few years ago, many people were naive about the geopolitical threats facing the tech industry. A bipartisan effort has since shifted the mainstream perspective. There is now a broad consensus that the race for AI dominance is critical. This environment is fostering a sense of patriotism within American-centric companies and investment firms. People are increasingly drawn to work at organizations that recognize these global stakes.