Gurley describes Stitch Fix using machine learning and collaborative filtering to deliver items customers haven't seen yet via stylists.
“We try and learn as much as we can about the individual, and then we use machine learning and collaborative filtering, a whole bunch of algorithms that helps of our stylists, and we deliver something to you that you haven't even seen yet.”
Gurley identifies long-term memory as a major unsolved problem preventing AI from becoming true personal assistants.
“I do think this memory issue is a big, big deal, at least on the consumer side. So none of the major contenders today can remember who you are because it would require them.”
Gurley explains personalized AI memory requires rebuilding models per user, which is economically unfeasible.
“And it's because you'd have to redo the model for each human. And of course that makes no sense economically.”
Gurley invokes Spielberg's Jaws example to suggest AI expectations may exceed what can actually be delivered.
“I saw this inner great interview once with Spielberg when he was talking about Jaws. And he said, you know, I didn't show the shark until like 75% of the way through this thing.”
Gurley identifies long-term memory as a major unsolved AI problem preventing true personal assistant capabilities.
“And I mean, remember over five years, ten years to really become a personal assistant kind of thing that her represented.”
Gurley explains AI memory requires retraining models per person, creating an economically unsustainable structural problem.
“And of course that makes no sense economically. So it's actually a huge structural problem. There are complaints, you know, in Reddit about Character AI on this front.”
Gurley says Tesla replaced coded rules with neural networks using driver video as input and pedal/steering as output.
“And literally the videos are the input and the output is the steering wheel, the the brake, and the gas pedal. Yeah. Extraordinary.”
Gurley says AI companies operate three to four super PACs and insert staff onto congressional teams to influence regulation.
“They're putting people they're literally inserting people onto the staffs of the different congressmen and senators to try and influence the outcome here.”
Gurley notes Microsoft hands out $11 billion annually in RSUs, giving them advantage in recruiting AI talent.
“So that's over 11,000,000,000 a year. And this morning we woke up and read about them taking some employees from a hot AI company in the valley.”
Gurley says foundational model companies burning $100-200M yearly cannot possibly represent high quality capital allocation.
“100,000,000 a year? 200,000,000 a year? There's no way that's high quality capital allocation from my point of view.”
Gurley says personalized foundational models that can retune for individuals could wildly change AI direction.
“Any of that stuff could create a wildly new direction than what we have today.”
Gurley says initial LLM versions are structurally flawed around personal memory and RAG approaches don't solve the problem.
“One quick thing, I do think that the initial versions of these models are structurally flawed around providing personal memory and allowing someone to become dependent on one of these things.”
Gurley explains Strawberry requires 10X processing for linear improvement, meaning 10-100X inference costs for better solutions.
“in order to get linear improvement, you have to do maybe 10X the amount of processing. And this is all inference. So what are the implications of that?”
Gurley argues email integration for AI creates insurmountable switching costs for whoever achieves it first.
“Like, that's really, really powerful. And I think the switching costs are insurmountable if someone gets to that place first.”
Gurley calculates AI companies burning $20M monthly need $500M revenue run rate to reach profitability at 50% margins.
“if you're spending 20,000,000 a month, you're burning $240,000,000 a year. You know, if these companies in AI are 50% gross margin or whatever, you got to get revenue to twice that, you know, you're right at that 500,000,000 run rate before you could think about being profitable.”
Gurley explains Strawberry's logarithmic scaling means 10X to 100X inference costs for linear improvement requiring selective application.
“The implication being in order to get linear improvement, you have to do maybe 10X the amount of processing. And this is all inference.”
Gurley praises Anthropic's decision to open source their MCP data connection tool with GitHub and Slack integrations.
“They released several integrations already in the initial pack, with good GitHub and Slack and a few others so that you can start having your agents interact with these things.”
Gurley says Anthropic and investor Eric Schmidt have been pushing for more AI oversight and regulation.
“based on what I've heard and seen and read and listened to on podcast, Anthropic and their investors like Eric Schmidt have been some of the people that have been encouraging, you know, more and more AI oversight and AI regulation”
Gurley contrasts Dario and Sam's platitudes with Zuckerberg being transparent and in the arena on AI.
“And Zuck was down in the weeds in the meet. He's in the arena. Yeah. Being super transparent.”
Gurley contrasts Zuckerberg's technical transparency with Altman and Amodei's high-level platitudes about AI.
“Dario and Sam talk in these high level platitudes about how this stuff's gonna cure cancer and we're all gonna not have to work anymore.”
Gurley warns AI companies receiving server capacity for equity don't recognize their true COGS in real-time.
“They're trading equity for server capacity. And so those companies are running on a day to day basis with a high COGS, but they don't know it”
Gurley says AI consensus is unprecedented in venture history with even Fortune five hundred fully committed.
“And if even if I go back to the mobile wave or the Internet wave, there were plenty of skeptics like within the Fortune 500.”
Gurley says founders' brains get warped when not paying cash for costs, forgetting to account for them.
“And my suspicion, having lived in the venture industry for twenty five years, is your brain gets warped in that environment. Like, you don't think about, oh, shit.”
Gurley clarifies Manus operates exclusively on US models, has no foundational model, and never used Chinese models like DeepSeek.
“first of all, they've only operated on US models. They're a wrapper company. They don't have a foundational model and and they've only operated on top of US models.”
Gurley notes 50% of AI researchers are Chinese-origin and China now leads in AI patents, questioning immigration restrictions.
“some people say 50% of AI researchers are of Chinese origin. And I believe now the patent count in AI coming out of China is larger than The US.”
Gurley identifies Microsoft, Amazon, and Apple as US tech companies not on AI bleeding edge.
“The reason I think that's an interesting data point when you think about The US and, you know, are several big tech companies that seem to be not on the bleeding edge of AI.”
Gurley notes China has ten open source AI models in hyper competition versus US proprietary approach.
“Wild. And by the way, I mean, not to divert too much back, but right now China has 10 open source AI models.”
Gurley says OpenAI's burn rate is five times Amazon's or Uber's, unprecedented in venture capital.
“And when we did at Uber, we had the same kind of burn rate, but Open the Eyes got a burn rate that's five times that size.”
Gurley says competing in top AI categories now requires willingness to lose a billion dollars per year.
“So every one of the most interesting AI categories, you may have to be willing to lose 1,000,000,000 a year just to compete”
Gurley says unprecedented capital at every AI project creates uncharted competitive territory.
“now we're in a weird world where the amount of money that gets thrown at every project, I kind of lived through this with Uber Lyft because both companies had billions and billions of dollars”
Gurley argues those who craft bespoke career paths can view AI as jet fuel rather than threat.
“I think you view AI as jet fuel, as something that's going to make what you're capable of even better.”
Gurley argues workers must understand AI's impact on their industry, comparing it to farmers adopting tractors and drones.
“Like if this technology has the ability to impact your industry, you need to know about it.”
Gurley warns alienating trade partners could result in China serving global AI markets while America is isolated.
“I think this could even happen with AI models where there's a fence around America and China's serving the rest of the world.”
Gurley says AI makes safe jobs riskier and pursuing passion a better career defense.
“if you were put in a safe job that you don't truly love, I would encourage you to consider this alternate door that I'm representing, to maybe find something you're passionate about and get away from that risk.”
Gurley cites Mark Cuban distinguishing people who use LLMs to learn faster versus not learn at all.
“The the people that use LLMs to learn faster and the people that use LLMs to not learn at all. I thought that was profound.”
Gurley argues AI becomes a superpower for learning and productivity when combined with high agency careers.
“Like, you can learn faster than you could have ever learned before. You can get more things done Yep. Like with fewer people.”
Gurley advises being the most AI-enabled version of yourself using tractor versus donkey analogy.
“One, in any role in any field, be the most AI enabled version of yourself you can possibly be.”
Gurley argues those who best understand AI's impact on their role won't be laid off.
“if there are 40 people in your org all doing the same thing, and you understand how AI affects that role more than the rest of them, you're not getting laid off.”
Gurley quotes Cuban's distinction between using LLMs to learn faster versus not learning at all.
“Mark Cuban the other day tweeted, he said he said there's two types of people. The the people that use LLMs to learn faster and the people that use LLMs to not learn at all.”
Gurley advises being the most AI-enabled version of yourself in any field, citing tool evolution.
“in any role in any field, be the most AI enabled version of yourself you can possibly be. There's tons of anthropologists that have written about how we evolve with our tools.”
Gurley argues open source creates hypercompetitive dynamics where innovations are immediately absorbed across competitors.
“And if you have 10 companies all doing open source, the minute an idea is shared, it's available to the other 10, and they absorb it immediately.”
Gurley argues AI is a jetpack for people running their own lane with bespoke passion and goals.
“For that individual, is AI a threat or is it a jetpack? I actually think it's a jetpack because now I can go learn about anything I want.”
Gurley says AI is a huge threat to people following a conveyor belt career path without differentiation.
“I think those people, AI is a huge threat because they're like, they don't really know why they're there.”
Gurley compares AI adoption to tractor versus donkey farming saying understanding AI better than peers prevents layoffs.
“you can just imagine a farming competition between a guy with a tractor and some drones and the other guys got a plow and a donkey.”
Gurley used AI to analyze best nonfiction conclusions, discovering eight of ten take orthogonal directions.
“And in reading through that list, I noticed that eight or nine of the 10 took an orthogonal direction to the book, so it didn't summarize.”
Gurley argues almost any job can be approached as an artisan would, not just creative fields.
“I have this interesting theory where almost any job you could think of it as being an artisan.”
Gurley argues top performers in any field understand nuance, which AI struggles with.
“Or do you think of, and I would say the best in those And fields probably so what is it they do? They understand nuance in their field quite a bit.”
Gurley quotes Cuban dividing people into those who use LLMs to learn faster versus those avoiding learning.
“He said, there are two types of people in the world, those that use LLMs to learn faster and even more, and those that use LLMs to avoid learning.”
Gurley claims this is the best time in history to chase dreams due to AI learning tools.
“the, the methodology in my book, the, the AI tools that are out there for learning and connecting have never been this is the best time in the history of the world to go chase your dreams.”
Gurley reports 98% of VCs focus solely on AI and non-AI company valuations halved in five weeks.
“98% of venture capitalists are only looking at AI and they're AI all day long, they don't wanna see another business and and and in the past five weeks, the valuations on the non AI companies have been cut in half”
Gurley estimates 98% of venture capitalists are exclusively focused on AI investments right now.
“From where I sit, 98% of venture capitalists are only looking at AI and they're AI all day long, they don't wanna see another business”
Gurley says non-AI company valuations have been cut in half in past five weeks due to VC focus on AI.
“in the past five weeks, the valuations on the non AI companies have been cut in half and so that mentality is gonna be reinforced even more and so there's just no oxygen.”
Gurley claims 98% of VCs focus only on AI and non-AI valuations were cut in half in five weeks.
“98% of venture capitalists are only looking at AI and they're AI all day long, they don't wanna see another business and and and in the past five weeks, the valuations on the non AI companies have been cut in half and so that mentality is gonna be reinforced even more and so there's just no oxygen.”
Gurley warns people who chose safe jobs without loving them are sitting ducks for AI technology.
“I think you're those are the people, oddly enough, that are kinda sitting ducks for this new technology that's coming along.”
Gurley argues artisans differentiating at the edge of their field are the last thing AI will replace, not rote work.
“The way you're differentiating yourself is what's out on the edge of that field. And that's the last thing the AI is coming after.”
Gurley argues artisans differentiate on field nuances not in LLM models, which AI attacks last.
“And it's the subtle nuance in any field that's not baked into the LLM models. And so this is where I go back to this notion as an artisan.”
Gurley cites rumors of thirty billion dollar Nvidia investment into OpenAI, calls modern venture capital sport of kings.
“And so, there's rumors today of a $30,000,000,000 investment from Nvidia into OpenAi on top of everything they've already already raised and spent.”
Gurley cites rumors of thirty billion dollar Nvidia investment into OpenAI, calling it sport of kings venture capital.
“there's rumors today of a $30,000,000,000 investment from Nvidia into OpenAi on top of everything they've already already raised and spent.”
Gurley cites Brett Taylor as example of founder who repots himself across technology waves from maps to social to AI.
“He's just always right there as the thing changes and on top of the next thing.”
Gurley notes radiologists have increased in number despite AI automation speeding up reading scans.
“I think there was someone at TED talking about the fact that the number of radiologists has gone up, not down, despite the fact that you can read these things so much faster.”
Gurley says SoftBank threatened to fund competitors if founders refused money, forcing preemptive rounds.
“And there's an implicit threat that, at least at SoftBank, Masa was very direct about and not implicit. That if you don't take it, I'll give it to your competitor.”
Gurley criticizes Dario for being both top cheerleader and doomer while raising billions and cashing out employee shares.
“Dario, the guy leading a company that's raising billions of dollars and cashing out billions in secondary to his employees is simultaneously the top cheerleader and the top doomer.”
Gurley argues learning has never been easier or freer than today for high-agency people.
“For people that have high agency and want to learn, there's never been a time in the history of the world where that is as easy and free as it is today.”
Gurley says most Silicon Valley people have innate FOMO about new technology, installing tools immediately out of competitive fear.
“most people that live in Silicon Valley just have this innate FOMO built into them about what's next.”
Gurley says Brett Taylor became an AI expert in three years despite not being one previously.
“Well guess what like three years ago he wasn't an AI expert but he became one.”
Gurley notes AI company burn rates increased from risky million monthly ten years ago to billions yearly today.
“Ten years ago, like, it was super risky to burn 1,000,000 a month. You know, today, these companies are burning 5,000,000,000 a year.”
“The biggest problem is if you're a AI skeptic because you're never trying to figure that out. And the truth of the matter is humans have evolved with their tools forever.”
Gurley quotes Mark Cuban dividing people into those using LLMs to learn faster versus skip learning.
“Mark Cuban said there are two types of people in this world, those that use LLMs to learn faster than they ever could before and those that use LLMs to skip learning altogether.”
Gurley proposes using AI to simulate working in different careers for a week each to test fit.
“you could use AI and just imagine yourself in a career for a week. Like, cause you could just talk to AI every day.”
Gurley argues competing without AI is like farming with shovel against modern equipment, zero chance.
“If if I ask you to compete with a farmer that has modern farming equipment and I give you a shovel and a hoe and an oxen, there's zero chance you're gonna be competitive.”
“And I mean so at this recent TED, I just went to Neil Cotyal who argued the supreme court case on the tariffs and won.”
Gurley says being the most AI-productive person in your field makes you indispensable and sought-after.
“And if you are the most AI productive human in your field, you're not getting fired. Like, you're you're the one they're asking all the questions of. Like, what's this capable of?”
Gurley argues that for proactive continuous learners, AI is a jetpack that enables them to achieve more faster.
“For that person, AI is a jetpack. Like they can now do more things faster than they wanted to do, and they can achieve more than they were able to before.”
Gurley says being the most AI productive person in your field makes you indispensable and sought after.
“if you are the most AI productive human in your field, you're not getting fired. Like, you're you're the one they're asking all the questions of.”
Gurley argues that becoming the most AI-productive person in your field makes you indispensable and sought-after for expertise.
“And if you are the most AI productive human in your field, you're not getting fired. Like, you're you're the one they're asking all the questions of.”
Gurley predicted three years ago that AI incumbents would use regulatory capture against open models, but says he underestimated how aggressively they would pursue this strategy.
“Almost 3 years ago, I believed open-models would become the key threat to the AI incumbents & that unfortunately they would lean into regulatory capture as a response. I underestimated the fervor with which they would use this non-business approach. https://t.co/oFDclTqJDq”
Gurley says he predicted three years ago that open-source AI models would threaten incumbents who would respond with regulatory capture, but underestimated how aggressively they would pursue this strategy.
“Almost 3 years ago, I believed open-models would become the key threat to the AI incumbents & that unfortunately they would lean into regulatory capture as a response. I underestimated the fervor with which they would use this non-business approach. https://t.co/oFDclTqJDq”