Software is no longer sold to humans, but to intelligent agents. The godfather of a $90 billion venture capital firm reveals: Cursor is going to "die", and NVIDIA faces a strong rival.

Jerry Murdock is one of the most influential venture capitalists in the past three decades. As the co-founder of Insight Partners, he currently manages assets worth over $90 billion. One of the most representative investments in his career was leading multiple rounds of financing for Twitter.

This veteran in the venture capital industry recently shared some profound experiences and insights from his more than three - decade investment career in a podcast. Murdock mentioned that many startups he has recently interacted with have expressed a similar view: Cursor is "dying."

Murdock pointed out that software procurement decisions are gradually shifting from being "human - led" to "autonomous agent - led." In the future, when enterprises choose software and technology services, it is likely that autonomous agent systems will automatically evaluate and make decisions, rather than the traditional human procurement process. If enterprises cannot adapt to this change, their competitiveness will be significantly affected.

In addition, he also believes that the open - source community and ASIC (Application - Specific Integrated Circuit) chips will play an important role in the next stage of AI evolution. This trend may even impact industry giants like NVIDIA. Meanwhile, AI may replace a large number of white - collar jobs in the next few years, and this change will inevitably trigger extensive discussions about the employment structure and the future form of work.

Looking back at previous technological waves in history, Murdock emphasized that in a rapidly changing era, what truly determines success or failure is often not the ability to predict, but the ability to seize opportunities and continuously adapt to changes. This is particularly crucial for investors and enterprises.

Below is the content of Murdock's podcast conversation. We have translated and trimmed the content without changing its original meaning for our readers.

Why is Cursor "dying"?

Host: This era is truly thought - provoking. Frankly speaking, I and even the entire generation of investors have a question: Has everything we've learned in the past decade become meaningless? Let's have a good chat about this. The other day, you used a tsunami to describe the impact of the current AI wave on us. Can you elaborate on the origin of this metaphor?

Murdock: First of all, it's important to note that a tsunami is not very dangerous before it hits. The greatest risk actually occurs when it rushes onto the shore. Of course, it does cause serious damage once it arrives. Moreover, a tsunami doesn't come in just one wave but in successive waves. The upcoming AI wave is far more than just a single product. In this context, autonomous agents will be the key to everything. In my opinion, autonomous agents are the essence of this tsunami, not just the general term "AI."

Host: So, what stage are we currently in? Is it just a forward - looking view that "SaaS is dead, and autonomous agents should take over" since it hasn't really arrived yet?

Murdock: I can't make long - term predictions, but one thing is certain: the pace of change is accelerating, and we must make pre - judgments to try to gain the initiative. The idea of "simply adding an AI module to the business process" is not necessarily wrong, and it may even have good short - term effects. However, only by making a thorough transformation with a truly AI - native mindset can enterprises become more competitive. The key lies in deeply considering the dynamics of the community driving this wave. Those emerging open - source communities are the real source of disruptive power.

Host: You have an amazing portfolio of business assets and have collaborated with many outstanding founders. What highlights do you think current news reports or investment analyses have missed?

Murdock: Focus on those truly AI - native startups, such as E2B, Eventual, Lotus AI, GetDynasty, and even Oven, which are already in operation. They are all using OpenClaw, NanoClaw, or self - developed agent systems. I think the market hasn't realized the power behind this trend because this technology has only been around for about two months. Moreover, these companies have only been using programming agents for two to six weeks, which is the most shocking part for me.

Host: What does it mean for Cursor when they use autonomous agents to write code? After all, it's a company valued at $27 billion to $30 billion.

Murdock: That's true. Many companies I've contacted have clearly stated that Cursor's products seem outdated. However, I fully admit that the Cursor team is very smart, well - funded, has many customers, and has enough time to explore its own autonomous agent technology. I think they will definitely do so. They also have the opportunity to adjust their direction and find their own future development path.

In conclusion, in the field of AI, we must focus on future trends and never be stuck in the past. Therefore, I believe that Cursor has only one choice, which is to quickly embrace autonomous agents.

The dividing line for all future products

Host: You mentioned the application of OpenClaw. What broader impacts will OpenClaw bring? What aspects haven't many people fully realized?

Murdock: Good question. Let's first focus on the OpenClaw conference and look at the community's commitment to open - source and the scale of developers. Big companies like OpenAI have invested huge resources in it, and the number of developers participating in integrated development in the open - source community is even larger. If this community continues to develop at an accelerating pace, we will surely see agents achieve extraordinary feats that are currently unattainable.

When it comes to future autonomous agents, the first thing to address is the so - called Claw Stack, which is a kind of infrastructure.

Many friends may not remember that in 2003 and 2004, after the 9/11 attacks, the entire software world was in trouble. People could no longer afford the high costs of Sun servers and Oracle databases. So, the LAMP technology stack, consisting of the Linux operating system, Apache server, MySQL database, and PHP front - end development tools, emerged. This led to an explosive growth in the number of websites between 2004 and 2005 and ultimately promoted the rise of e - commerce. Google went public in 2004 and perfectly caught this wave of the era.

I believe that autonomous agents will follow the same pattern. The open - source community will also introduce its own technology stack, overthrowing the current inference layer dominated by Claude, Codex, and Gemini. Future autonomous agents will form an orchestration layer to coordinate multiple large - language models and achieve hierarchical processing of workflows.

They will perform hierarchical processing of workflows. For example, for some workflows, they may use Claude, which has a high token cost but can do a better job, while for other workflows, they may use open - source models such as DeepSeek or LLama 3. They will plan from a macro perspective, and perhaps this can be fully realized when the workflow orchestration technology matures in the future. They will accurately allocate task loads to different inference models, greatly improving the overall system efficiency. I believe this will greatly accelerate the popularization of open - source models.

This will promote the rise of ASIC chips because the advantage of ASICs is that they can directly embed models into the chips. Compared with Intel's expensive general - purpose chips, ASICs not only significantly reduce costs but also can be finely tuned for specific workloads. Therefore, I believe that autonomous agents will have a profound impact on open - source models and ASIC chips, triggering a new wave of development. From a macro perspective, this is a technological revolution.

Host: My good friend Rory O’Driscoll from Scale has an idea that I really appreciate, which is that as venture capitalists, we must clarify our own positions. I hold a large number of NVIDIA stocks, so when listening to your analysis, I can't help but wonder: Can you specifically introduce the process of transitioning from NVIDIA chips to ASIC chips? And how will this trend become a reality?

Murdock: The real reason Huang Renxun acquired Groq is that the Groq team is well - versed in chip - level memory integration technology, and NVIDIA needs this technology to support ASIC chips. Therefore, I believe that the acquisition of Groq is not only about handling diverse workloads and achieving chip - level memory but also ensuring that CUDA supports ASIC chips. Huang Renxun must understand this well, and they know the industry's development trend best. So, I think the acquisition of Groq will help CUDA remain competitive in the upcoming ASIC wave.

Host: If CUDA can adapt in time when the ASIC wave arrives, can NVIDIA maintain its value? Or will it ultimately face value loss and a decline in valuation due to the expansion of ASIC chips?

Murdock: But it still depends on specific implementation, which is the essence of competition. Whoever can implement better and act faster will win. Many people criticize Meta for being behind the times in some aspects, but at least they have the courage to say no to Huang Renxun. "Sorry, we don't need NVIDIA's chips." Why can they say so? Because they have clearly decided to bet on ASIC chips.

Host: You mentioned the scheduling and diversion mechanism between different models, which makes me think of two questions. First, does this mean that models are becoming commoditized, that is, getting caught in a price war, with model manufacturers competing to provide cheaper and faster services? Second, by extension, this trend at the base - model level will also continue to penetrate the application technology stack. What's your view on these two trends?

Murdock: This question will be decided by autonomous agents, not developers. Agents are fundamentally different from developers. As developers, we humans take actions based on experience and judgment, such as deciding to build a certain system or choosing ASIC chips. However, autonomous agents have probabilistic characteristics. Agents will consider in a probabilistic way: Which is better, ASIC or NVIDIA chips? The simplest way is to directly call 10 sets of Python libraries in 10 independent sandboxes, write workloads, and then compare their performance. This is the future development direction, so we must pay attention to the increasing influence of autonomous agents in the development field.

Overreaction on Wall Street

Host: But when you think about it, although such a future is wonderful, what should we do? Like other investors, I've also bet on technologies at the agent layer. At the same time, I've witnessed Anthropic launching application - layer products at an amazing speed, including legal services and office collaboration tools. So, how can we ensure that our investment goals are both safe and reliable and won't be replaced by Anthropic's product iterations in the foreseeable future?

Murdock: Haha, that's why investors get high salaries. No one can tell us what is safe. Frankly speaking, 80% of my investment decisions have a return of less than 30%. The remaining 20% of the investments are the ones that truly create wealth and influence. The funds we manage are just a measure of an investor's influence. Especially in the current AI wave, there is no absolutely safe investment. The only thing we can do is to focus on who can implement better.

Host: Do you think the stock market's reaction to these new waves is too positive? Stocks of companies like CloudFlare, CrowdStrike, and Anthropic often soar by 10% after a single news release. Is this an overreaction?

Murdock: The problem you mentioned is actually a typical case of excessive market volatility. Do you still remember the huge bull statue on Wall Street? It's actually a concentrated portrayal of such group psychology.

Yes, investors think simply: Since I don't understand technology anyway, I'll just wait and see. Some panic - sell, but most just hold their money and wait. The current market situation is not so much a panic - selling as a situation where buyers on the sidelines are still hesitating, thinking that the current market is not cost - effective. I also don't know when it's the best time to heavily invest in CrowdStrike and further drive up its stock price. So, I can't give a definite conclusion on whether this is an over - adjustment or a waiting period for many investors. In short, the only certainty now is uncertainty. No one knows who will win and who will lose. That's my understanding.

Host: Has there been a similar precedent in history? Those heart - wrenching moments when, in the face of an unknown situation, people would rather wait and not bear the consequence of a 40% plunge in a week?

Murdock: In March 2000, 26 years ago, technology stocks tumbled by 30% or even 40%. If a company's quarterly performance did not meet expectations, the decline could even reach 50% or 60%. From that March to the following summer, the market was in a slump.

By the way, we still successfully completed several IPOs at that time, but the valuation figures were significantly reduced. We were struggling in pain until the 9/11 attacks, which dealt the final blow and completely destroyed market confidence.

At that time, we thought that since we were not doing Internet investment and consulting, we should be fine. After all, as early as 1999, we felt that there was a bubble in the Internet industry and predicted that Internet companies were bound to collapse. We asserted that such a model could never be sustainable. The business foundation was weak, the number of dial - up Internet users was insufficient, the weak network transmission capacity could not support business operations, and the underground fiber - optic cable laying was far from enough. Sure enough, the bubble finally burst. However, the tsunami - like impact not only engulfed Internet companies but also dragged all software companies into the whirlpool. We all struggled in despair for several years.

Host: Are we at this critical juncture again? Do you think the stock market's slump will last for a long time, or will this situation be different?

Murdock: The specific details will definitely be slightly different, but the key lies in the speed of change. For example, many people think that the E2B company only does sandbox technology, but most technical practitioners don't pay much attention to the differences between different sandboxes and only regard them as underlying tools to ensure the safety of agent programs or codes. In fact, it is also a productivity tool. When building sandboxes for agent programs, there will be a qualitative difference in running speed.

For example, the critical point of human perception of delay is about 400 milliseconds, so most sandboxes on various systems need to control the delay within this time. The response time of E2B's sandbox is only 70 milliseconds. At that time, I exclaimed: This is incredibly fast! More importantly, agent programs can accurately perceive the delay, which is the key. When an agent needs to start tens of thousands of sandboxes within a few seconds, guess which one it will choose?

Host: We've mentioned autonomous agent programs several times. Will the emergence of such programs make traditional record - keeping systems lose their value, or will it integrate the existing distribution network and greatly increase its value?

Murdock: It depends on the specific situation. I've been investing in Carta, a dedicated record - keeping system in the stock field, for a long time. It has great potential, and I think they make very good decisions in all aspects. If future stocks are tokenized and still use the Carta system, its value will definitely increase exponentially. After all, Carta has the equity structure table and has the opportunity to assist in managing the tokenization process. However, if the tokenization wave bypasses Carta and chooses to build a new record - keeping system, then the future value of Carta is hard to say. So, the key lies in the execution ability of Carta's management team and their ability to grasp future trends.

Host: Regarding the current ecological situation of Salesforce, which camp would you classify it into?

Murdock: Many enterprises build their businesses based on the Salesforce record - keeping system. So, to judge the fate of Salesforce, we need to observe the health and development trends of these enterprises. If these enterprises collapse one after another and the number of closures is enough to weaken the intrinsic value of Salesforce, then people will naturally think that Salesforce is worthless.

However, in the history of enterprise development, Salesforce is like Mount Everest, a towering peak of 8,000 meters. It won't melt away overnight and will surely exist for a long time. The key is how we quantify its value. To analyze this problem, we need to clarify the entry point: How much value do those enterprises built on top of Salesforce actually have? If these enterprises start to decline, then the value of Salesforce will shrink.

Host: The responsibility of investors is to judge the value and future value growth. However, the traditional criteria for measuring value, such as revenue, growth rate, and profit margin, are now questionable and debatable. In this context, how would you re - define the criteria for judging value?

Murdock: All considerations about value cannot be separated from the time dimension. Judging from the development trajectories of most software manufacturers, when new technologies first emerge, they usually expand their market share first and then shrink. This is the long - established development inertia of enterprises, and this inertia is still operating step by step.

I don't think traditional software manufacturers will suddenly disappear from the market. In reality, it's more likely that people's panic causes fluctuations in their stock prices. The fundamental value of an enterprise will depend on the ability of its management team to adapt to the new situation. For example, those manufacturers with high - quality data systems can increase their value if they can coordinate these data with AI agents (especially autonomous agents). On the contrary, those slow -