Skip to content

Podcast

20VC- Why Cursor Is Dead | an AI Tsunami Is Coming & You Need to Prepare | Systems of Record Become Valueless Databases With Agents | Is This the End of Tech Private Equity With Jerry Murdock, Co-Founder of Insight Partners

Startup Funding | The Pitch

Source ↗ ← All highlights
  • Autonomous Agents Are The Tsunami
    • Autonomous agents are the core of the upcoming AI tsunami, not just incremental AI improvements.
    • Jerry Murdock calls the wave dangerous only when it hits the beach, emphasizing a systems-level shift toward agent-first design. Transcript: Jerry Murdock Do you? Well, the first thing I think about a tsunami is that it’s harmless when it’s out at sea. It’s only dangerous when it hits the beach. That’s number one. Number two, it’s messy. You’re going to get maybe an earthquake or two, notifying there’s something possibly coming. And it’s not just one wave. There’s these pre-peak waves and post-peak waves. But there’s an event coming that is more than a single product, and in this case, it’s autonomous agents. (Time 0:05:00)
  • Open Source Will Build The Agent Stack
    • Open source communities will build a “claw stack” for agents with orchestration across multiple LLMs and routing decisions.
    • Jerry compares it to the LAMP stack era and predicts model orchestration, open models rising, and ASIC chip specialization. Transcript: Jerry Murdock And I think that’s not obvious yet in the marketplace, because it’s only about two months old. These guys have been doing it for anywhere from two weeks to six weeks, roughly, that have brought in autonomous agents to actually write code. Harry Stebbings That’s the thing that I think is mind-blowing. Okay, so they have autonomous agents to actually write code. What does that mean for cursor, a 27 to $30 billion company that’s raised a lot of money? Jerry Murdock Yeah, I’ll tell you. So for most of the companies I just mentioned, their view, as they’ve told me is cursor is obsolete. That’s where the product is today. Now, my view is Yeah, but that team is really smart. They’ve got a lot of money and a lot of customers. They’ve got time to embrace autonomous agents, which is what I think they’ll do. And they’ve got shots to pivot and figure out what the next direction is. But in the AI business, you’ve got to be going where things are going to be. You can’t be thinking about yesterday. So I think those guys are going to have to quickly embrace autonomous agents. You mentioned utilization of OpenClaw. Harry Stebbings What do you think is the impact of OpenClaw more broadly that we’re maybe not considering enough? Great question. Jerry Murdock Look, first thing you got to do is look at OpenClaw and look at the community, the commitment to open source and the number of people developing for it. I mean, you can look at huge companies like OpenAI, and you can look at Anthropic putting massive resources. And then you look at open source, and you’ve got sheer numbers of people doing integrations, massive amounts. If that community keeps accelerating and growing, we’re going to see agents do incredible things that they don’t have today. When I think of an autonomous agent, first of all, we’re going to have to come up with what we call the claw stack or some form of stack for the autonomous agent. Back in, you may not remember, back in 2003, 2004, we were post 9-11. The world was miserable. People couldn’t afford to build websites with Sun servers and Oracle databases, which were expensive. But the LAMP stack came out, which was Linux, Apache, Web Server, MySQL database, and PHP front-end development tool. And that led to the explosion in 2004, 2005 of websites and ultimately commerce. Google went public in 2004 and rode that wave brilliantly. And I think in the same thing with autonomous agents that you’re going to find the open source community coming out with the stack, right? Right now you’ve got a reasoning layer that’s dominated by Claude and Codex and Gemini. I think what’s going to happen with the autonomous agents is ultimately they’re going to have an orchestration layer where they can have multiple different LLMs, you know, that they Can orchestrate on and triage workflows. They’ll triage workflows and say, Hey, for this part of the workflow, let’s use Claude, they’re expensive tokens, but they’re going to get the job done better. And then for this other part of the board, let’s use an open source model like DeepSea, Klama 3, whatever, you know, and they’re going to start from a big picture. Maybe this is going to happen as soon as orchestration gets solid enough, they’re going to determine where the workloads go for reasoning models, which is going to be super powerful. And my guess is this will lead to the rise of open source models more proficiently. And that’s going to lead to the rise of ASICS chips, because what’s going to happen with ASICS is you’re going to put the model on the chip. (Time 0:07:04)
  • Incumbents Must Embrace Agents Fast
    • If you’re a company like Cursor, pivot quickly to embrace autonomous agents or risk obsolescence.
    • Jerry warns incumbents must adopt agent-native approaches despite existing customers and cash runway. Transcript: Jerry Murdock And they’ve got shots to pivot and figure out what the next direction is. But in the AI business, you’ve got to be going where things are going to be. You can’t be thinking about yesterday. So I think those guys are going to have to quickly embrace autonomous agents. (Time 0:07:52)
  • Agents Will Route Workloads Between Models
    • Agents will triage workloads across models, choosing Claude for critical tasks and cheaper open models for others.
    • This probabilistic benchmarking by agents could drive commoditization of models and favor orchestration layers. Transcript: Jerry Murdock They’ll triage workflows and say, Hey, for this part of the workflow, let’s use Claude, they’re expensive tokens, but they’re going to get the job done better. And then for this other part of the board, let’s use an open source model like DeepSea, Klama 3, whatever, you know, and they’re going to start from a big picture. Maybe this is going to happen as soon as orchestration gets solid enough, they’re going to determine where the workloads go for reasoning models, which is going to be super powerful. And my guess is this will lead to the rise of open source models more proficiently. And that’s going to lead to the rise of ASICS chips, because what’s going to happen with ASICS is you’re going to put the model on the chip. You know, ASICS is going to be a lot cheaper, a lot more tunable for a specific workload than an expensive chip from Jensen. So I think that we’re going to see massive development of autonomous agents influencing open source models and ASICs chips. But in the big picture, that’s a revolution. One of my dear friends is Rory O’Driscoll from Scale. Harry Stebbings And he says, the thing I love about Harry is the statement, that’s great, but what about me uh which i specialize in as a venture capitalist i have a shitload of nvidia and i’m thinking Through as i listen to you can you just play out for me that migration from jensen chips to a6 chips and just like how that actually plays out in reality well it’s the thing that wasn’t mentioned Jerry Murdock As to why he bought Grok. He needs that capability so that he can handle ASICS chips eventually, right? I mean, because Grok, those guys know how to put memory right on the chip. And so that’s automatically an ASICS chip today. And so I think Grok not just about handling different types of workloads and getting memory on the chip. It’s about making sure that CUDA can also support ASIC chips. They know it’s coming. They absolutely know it’s coming. And so I think the Grok acquisition is going to help them make sure that CUDA is viable for the ASIC explosion that’s coming. Harry Stebbings If the ASIC explosion comes and CUDA does migrate with it, does NVIDIA retain their value or does it still denigrate because they have a leakage of value with the A6 chip expansion? Jerry Murdock Depends on execution. This is the thing about the game, you know? Who’s going to out-execute? Who’s going to get it more? A lot of people like to criticize meta here and there about being behind the game, but they had the balls to say no to Jensen. Sorry, Jensen, we don’t need you. So why did he do that? Harry Stebbings Because he’s betting on ASIC’s chips. No question about it. You also mentioned the routing and the triaging between different models. It makes me think a couple of different things. One, is that not just like a commoditization of models, where it’s a race to the bottom on price and who can deliver the cheapest, fastest? Two, aligned to that, what if then we just see models do what we are seeing, which is eat into the application stack? How do you think about those two elements? Jerry Murdock The answer to that question is going to be decided by the autonomous agent, not developers. The agent is different than a developer. If you and I are developers, we’re going to go and do something based on what our experience tells us, what we think. Okay, let’s go build something here and we’re going to use an ASIC chip or whatever. The difference is an autonomous agent is probabilistic. So the probabilistic nature of an agent is going to say, I don’t know which is better ASICs or use one of Jensen’s chips. Why don’t I just go out and get 10 Python libraries, run them in 10 different sandboxes, work the workload, and then to see which one performs better? (Time 0:09:55)
  • Agents Become The New Buyers
    • Buyers will shift from humans to agents which act as employees with identity and credentials.
    • Jerry predicts pricing will move to consumption models where agents request and pay based on sandbox/compute usage. Transcript: Jerry Murdock Right now, all software is eventually purchased by human beings. Software is going to be purchased or used by agents. And what’s going to happen is an autonomous agent becomes an employee. You give it credentials, you give it identity, and then it’s up to the agent to make the decision. You will review them like an employee and say, hey, what did you buy? What did you spend? What did you accomplish? And you will manage that autonomous agent just like an employee. You need to prepare for that. If whatever you got invested in, is that future going to be done and controlled by agents? (Time 0:22:09)
  • White Collar Roles Face Early Displacement
    • White-collar jobs that input data (assistants, junior devs, marketing ops) are most vulnerable and hiring will slow first.
    • Small and medium businesses will adopt agents earlier than enterprises, causing uneven labor displacement. Transcript: Jerry Murdock We’re about two and a half years from the next presidential election. I’m here to say that that is going to be a major issue on the next presidential election. It could decide the election. It’ll be one of the most important things because there’s no easy answer. There’s no question that anyone that inputs data into a computer that does scheduling, that, an executive assistant or people doing marketing, those jobs are ultimately going to Be better done by autonomous agents, white collar jobs. People have been saying that for years. I mean, this is nothing new. The question is, when was it going to happen? And now with the advent of OpenClaw and NanoClaw and all the other autonomous agents that are coming, they’re the direct threats. First of all, not to the people with the job. First of all, to the next person in line, the next junior developer you want to hire, the next executive assistant, the next marketing person. So the first thing you see is people stopping or slowing down hiring of white collar employees that input data and use a computer. That’s the first thing. Second thing, depending on the speed at which these autonomous agents evolve, because there’s a lot that they need to do. They’ve proven it’s exists that they can write code and they can do anything you want from a scheduling perspective for your life. So they already can replace a human being. (Time 0:25:44)
  • Firms Should Use Agents To Invest Smarter
    • VC and PE firms should build and use autonomous agents to analyze markets and vet opportunities.
    • Jerry recommends evaluating founders by how well they deploy agents and using agent-driven data as a decision input. Transcript: Jerry Murdock So if I was starting from scratch, the one thing I could imagine is having incredible data. I could vet on a new opportunity for the white space for where these new AI companies are going to go and say, wow, the data says this guy’s on the right market. He just needs to execute. And then you evaluate not just the person, but the quality of how they use autonomous agents. That’s going to be the deciding factor for venture capitalists and startups. (Time 0:34:11)
  • Insight’s Misstep With Early Europe Expansion
    • Early Insight mistake: starting an ill-timed European office with a famous hire that shut down after six months.
    • Jerry says they weren’t mature enough and should have stayed co-located in New York to learn and focus together. Transcript: Jerry Murdock Oh that we hadn’t done in the early days of insight the big mistake, we hired some famous guy to start Insight Europe. And after six months, we shut it down. It was a mistake and a half. We were not mature enough to do that. And Europe was something, well, you know, we loved Europe. Early on, Fund One had an investment called SLP in Paris. Fund Two had a company called MetaQuatro in Spain that went public. That was a big win. We did deals in Portugal. We did deals in the Czech Republic in the 90s. So we thought, oh, let’s do a fund. Big mistake. Too much difference. We needed to stay in New York together. We needed to focus. We needed to learn from each other. Having people in Europe was too challenging. That was a mistake that we made. (Time 0:42:40)
  • How Jerry’s Fast Bet On Twitter Paid Off
    • Jerry’s quick decision to back Twitter in 2009 was a high-conviction, rapid move despite partner skepticism.
    • He closed within 30 days, pushed partners to join, and it became a watershed consumer investment for Insight. Transcript: Jerry Murdock One correlation is you can look across all VC funds and look at their vintage, whether it’s a 99 vintage or 2004, 2005 vintage. And the one correlation you can make to success was timing. If you did a fund in 2005, 2006, you were in great shape to take advantage of the mobile thing that started happening in 2008 when Steve Jobs got 10 million subscribers from AT&T. That’s when mobile became real. So if you had an 05, 06 fund, you had a chance to get some of the best, most iconic companies. But if you start a fund in 2009, you’re gonna miss being early in Twitter, you’re gonna miss being early in Facebook, you’re gonna miss being in Uber, you’re gonna miss all that. So timing is the most important thing, whether you’re early stage or late stage. You can’t apply it to just an overall general strategy. Harry Stebbings Depends on the timing of when you did the investment. Do you think now is the best time ever to be starting a new fund? Absolutely the best time. Jerry Murdock You’re in a sea change. Humans are no longer going to be the decision makers about software. It’s going to be autonomous agents. This sea change, this tsunami is so different than anything that’s come in the past that people that embrace this new model can embrace it from scratch right now, can have a huge advantage Over people that have been successful with older models that don’t quite move fast enough, because they’re already rich. They already made a lot of money. It’s hard to get old dogs to move fast. Harry Stebbings How do you feel about getting back in the game then, Joe? You’re doing this show with me. You’re talking about auto-organizations. Seems like you’re an old dog moving quite fast. Jerry Murdock Here’s the secret. Here’s the real secret. I never left the game. I retired from inside. I retired from negotiating term sheets. I retired from board seats. I retired from internal investment committee meetings. I retired from having dinners with LPs and answering LP phone calls, but I didn’t retire. I haven’t done a single investment on my own that someone didn’t bring me that I love and I trust. Every investment that I’ve done, which is over a hundred, has come from someone close to me and said, Jerry, I need your help to help me think about this thing. And through that process of trying to help someone else, I end up in it. And that’s been fun, because it’s an experience that you’re not doing on your own, your experience you’re doing with people you care about. Harry Stebbings Final one before we do a quick fire. Do you believe the future of venture is like the go big or go home we see andreason race 15 we have insight a mega platform we have the mega platforms of gc and light speed is it go big or go Home now that’s the only way i played the game and i mean i went big on my things you know when i invested in twitter in 2009, there was 30-something people, no revenue. Jerry Murdock My fund were like, are you kidding me? It was a watershed moment that my partners allowed me to make that investment. What did you see? What I saw, and this is really interesting, like our early investments, the idea of Twitter, which I call the status update for the world, was so brilliant, so over the top. It was far better than anybody on the team’s ability to execute against it. So I think Twitter could have been infinitely a more interesting company. But VCs got involved and created politics and they fired Jack Dorsey early and they created issues. And the company was pretty messed up from that trauma of firing one of the founders. And at the same time, they were very excited and the community was growing, the community was helping it. And for me, I saw an incredible idea that wasn’t easily replicated. And there was enough momentum to it that this was for me a no-brainer. And it was a bet that I put my entire reputation on the line. I got Jeff Horne and Devin Parekh to jump in with me on this. And I pushed really, really hard. We did that investment less than 30 days from the time I met the team and the time we closed. (Time 0:45:04)
  • Distribution Beats Tech Alone At Scale
    • Scale and distribution matter hugely; crossing hundreds of millions to a billion users is a key defensibility factor.
    • Jerry favors OpenAI for its vast active user base as a structural advantage over rivals with product depth but less reach. Transcript: Harry Stebbings It depends what stage you get me a shot in. I’m getting you a shot in OpenAI at 500, and I’m giving you Anthropic at 380. I’d take them both. Jerry Murdock I’d take them both, but I think for me it’s OpenAI because I got 800 million people doing it. You got 800 million people doing it. (Time 0:52:26)