a16z

Aaron Levie of @Box on how AI agents with wallets could create new business models for the internet: "I've always wanted micropayments for things online... I've appreciated why it's a bad business model, but I want it." "Now it's probably a good business model. Let's say you had paywalled, really proprietary data. You're Bloomberg. You're a medical institution. You could imagine there's maybe a hundred million agents that would be tapping into that data." "All of a sudden, a penny a transaction starts to make a lot more sense than it did when you were capped at 20 million [knowledge workers] that maybe were interested in your content." "We probably have new business models for tools and APIs... We probably want an agent to be able to have a budget and make any API call it wants." "All of a sudden you have this new revenue stream, which is these agents showing up, just transacting on your system." @levie at Daytona Compute 2026

Vídeo

How your AI agent can pay with your credit card, without ever holding it: 1. Your card stays stored in a trusted wallet. 2. When your agent goes to buy something, the wallet creates an agentic token: a stand-in card number that only works at that store, with limits like "don't spend more than $100." 3. Your agent hands the token to the store. 4. The store runs it through its usual payment processor, and the charge lands on your real card, with your rewards and card benefits intact. "AOL was a helpful set of training wheels for getting everyone online, but the open web ultimately won. We think the same thing will happen with agents as well." Full piece from VGS's @terryangelos on the Wallet Wars: t.co/RvC0pSRjGH

VGS's Terry Angelos and a16z's Alex Rampell on the one thing AI agent payments can't afford to break: the right to get your money back. Terry: "I asked Instinct to go and help me find these lights, and it found the lights at Walmart... when it came time to pay, Link [my payment wallet] didn't have a connection to Walmart. It can't just give my card over to the agent and over to Walmart." "If there's a problem... I call up my bank... And they say, 'Well, we don't see a Walmart transaction here... We just see a [Link] transaction.'" Alex: "One of the reasons why Visa and Mastercard have such impregnable positions is because they intermediate everything, and they set the rules, and chargebacks are very important." "It's a trust mechanism. I pay you for lights. You don't deliver the lights... You actually have rules by which I can kind of lodge a complaint." "That's high utility, actually even for the merchant... I'm much more willing to buy with a merchant knowing that if they don't actually deliver a product that works, I can get my money back." @terryangelos @arampell

Vídeo

AI agents can now book the hotel and order the cleats, as long as they're allowed to pay. That's set off the "wallet wars," with major players racing to connect to agents: - Stripe's Link: 300M users - Shopify's Shop Pay: 250M users - PayPal: 400M+ users How your agent pays matters. If a wallet pays the store with its own stand-in card, the purchase goes through, but your card rewards, purchase protections, and relationship with the store can get lost along the way. Say your agent books a United flight on your airline credit card. If the wallet pays United with its own virtual card, the money moves, and your miles may never show up. Agents also have to pick the right card for each purchase, and when one books the wrong hotel dates, someone has to make it right. Where it's all headed: new open standards from Visa, Mastercard, and Amex. "The best of all worlds is a future where my agent can use my credit card, without having my credit card." Full piece from VGS's @terryangelos on the Wallet Wars: t.co/RvC0pSRjGH

The wallet wars are here. Very Good Security chief strategy officer Terry Angelos joins a16z's Alex Rampell, his TrialPay co-founder from back in 2006 (TrialPay was later acquired by Visa), to discuss how agents are learning to spend money. "Agents have all of a sudden become significantly more helpful and useful because they effectively have arms and legs, which are payment credentials." VGS holds the payment credentials behind about 1/3 of all e-commerce, and now it's helping AI agents pay merchants. If you've bought anything online, there's a good chance VGS has already helped keep your card safe. In the last three months, Terry says agents went from "spectators to actors." Once an agent has your payment credentials, it can book the piano teacher it previously could only find and recommend. Now Stripe's Link, Shopify's Shop Pay, and PayPal are racing to connect to agents. It's all headed toward open standards from the card networks: agentic tokens that let an agent pay with your actual card. VGS went from zero agentic tokens to a few million a week in three months. 0:55 From TrialPay to Visa to VGS 2:05 How VGS vaults 1/3 of e-commerce 6:05 Why big merchants use 3-5 processors 9:55 Why AI agents need a payment vault 12:15 Agents go from "spectators to actors" 13:40 The wallet wars, old and new 19:55 Walled gardens vs open standards 23:55 How agents can pay with your real card 26:40 Your saved card goes multiplayer 29:40 How do you trust an agent's purchase? 32:35 Who pays when your agent gets it wrong 35:40 Why Amazon blocks agents 39:50 What's overhyped and underhyped 43:35 When merchants get agents too 45:40 Why agents could break card rewards 47:00 VGS's eras: vault, updates, network @terryangelos @arampell

Vídeo

AI could help defend communities that lack the resources to fight off cyberattacks. a16z's @AnneNeuberger on the mission behind the Securing America's Infrastructure Foundation, starting with rural hospitals: "America has 1,800 rural hospitals... Those hospitals lack the skills, the people, the money to ensure that they are not disrupted due to cyberattacks." "AI makes a real difference in being able to economically defend networks, to find vulnerabilities..." "A, can we build an easy open-source AI harness to enable us to find and block cyberattacks?" "B, stand up an effort to take the code on IoT devices like internet-connected MRI machines to patch those." "And thirdly, when there is an incident, adopt the model that New York City and New York State used, where they set up a security operations center that was able to support multiple smaller communities in responding to an incident." "So those are the three parts we're thinking about in helping critical infrastructure leverage AI to secure the critical systems Americans rely on every day." Full announcement introducing the Securing America's Infrastructure Foundation: t.co/NXg2JpmUWx

Vídeo

The full conversation with @AnneNeuberger, @jai_ramaswamy, and @GoldsmithOnGov on SAIF: a16zpolicy.substack.com/p/securing-ame…

Northwood Space CEO Bridgit Mendler on building the infrastructure for 1,000 SpaceXs: "I wanna see 1,000 SpaceXs. And I wanna see them not take 25 years to get there. I wanna see them get there in five years." "I really believe that the best outcome for humanity is being able to have a real, open, unfettered, democratized space future." "Being able to build infrastructure that introduces capabilities that so many different, not just companies, but countries can stand on, is really exciting." "Think of cloud infrastructure and what it enabled for all of the internet-based companies. So many companies would not have been able to achieve that global reach if they had to think about deploying their own fiber and standing up internet exchange points all around the world." "I know it sounds like a ludicrous thing to say, 1,000 SpaceXs, but if you said that about cloud, maybe it doesn't sound so ridiculous now." @bridgitmendler w/ @itsmoislam

Vídeo

Compute scarcity, in one chart Old GPUs are turning into appreciating assets

Five bangers from Nathan Lambert: 1. China's compute gap: "You talk to people [in China], and you're like, 'Oh, I heard an average OpenAI researcher has five thousand H100 equivalents that they can access at any time,' and like their jaw just falls to the floor. They're just like, 'I can't comprehend.'" 2. China's talent pool: "The Chinese labs are very technically excellent... To say that China is anything but equal on talent, you're gonna have a hard time because they just have so many people being trained at such a high level." 3. OpenAI's superpower: "[The frontier labs] have just taken the most ambitious bets the earliest in order to get there, and that's why OpenAI has done things like kickstart the reasoning model revolution." "And I think that kind of breakthrough and that beautiful plot with the inference scaling laws is peak OpenAI. They're really good at getting this plot that just changes how everyone thinks about everything." 4. Underestimating the competition: "The Chinese are so good at building models as well. I think there's a bit of cope where people are like, 'Only we are good at building models.' There are so many companies. There's, like, 20 to 30 companies that are building good models." 5. The AI risks that are already here: "[The lack of preparation] is kind of the thing that worries me more than, say, RSI is coming and unlocking new risks. There are risks that are here and that are gonna continue to diffuse." "Figuring out how to get society to prepare is very hard. I'm of the opinion that we need more eyes, we need more science, so that society feels informed." @natolambert

Now that AI agents can own an outcome from start to finish, how do you sell them? 1. Sell the platform: License the agent to a company to run in its own operations. No existing contract sets the price, so there's more upside, but new budgets take time to approve and results are harder to credit to the agent. 2. Sell the service: Take over work a company already outsources, the way Red Ventures sold policies for insurers and got paid per policy closed. The budget already exists, so deals close faster, but the highest-value work usually runs in-house, which caps the upside. 3. Go full-stack: Become the business, like buying debt portfolios and collecting on them yourself. You keep all the value the AI creates, but you also take on the licensing and compliance. Full piece from a16z's @dhaber on long-running agents in high-value customer journeys: t.co/aHaatcizcd

Nathan Lambert of @trillium_labs says Chinese AI researchers' jaws drop to the floor when he tells them how much compute OpenAI has: "The Chinese labs are very technically excellent. They are more compute-starved." "Without a doubt, they have a comparable talent pool to draw on... To say that China is anything but equal on talent, you're gonna have a hard time because they just have so many people being trained at such a high level." "What they don't have is compute." "You talk to people there, and you're like, 'Oh, I heard an average OpenAI researcher has five thousand H100 equivalents that they can access at any time,' and like their jaw just falls to the floor. They're just like, 'I can't comprehend.'" @natolambert @lishali88

Vídeo

How AI for the enterprise is evolving, from owning a task to owning an outcome: 1. Assistant: handles one task in seconds, like sending a follow-up email. Companies pay for it out of their software budget. 2. Agent: handles one interaction in minutes, like a customer's refund call. It takes on work people used to do, so it gets paid for out of the labor budget. 3. Long-running agent: works toward a goal for weeks or months, like getting an injured client the biggest settlement possible. Because it owns the result, the company selling it can price on the outcome it delivers. Full piece from a16z's @dhaber on long-running agents in high-value customer journeys: t.co/aHaatcizcd

Nathan Lambert, former post-training co-lead at the Allen Institute for AI, now launching Trillium, a nonprofit open AI lab, joins a16z's Lisha Li to discuss whether open models can really keep up with the frontier. When he tells researchers in China that an average OpenAI researcher reportedly has five thousand H100s to work with, "their jaw just falls to the floor." In his view, Chinese labs have the talent but are mostly short on compute. He also thinks the distillation story is overblown. Based on the research, he puts its edge at a few months, and says "there's a bit of cope" in thinking only American labs can build good models. A lot of frontier progress outside math and code comes down to labs buying 1) data and 2) RL environments. It's a multibillion-dollar market, and some of it is sloppy enough to teach models to reward hack. That's a big part of why Trillium exists. Nathan wants to give away the environments other people would sell, and show exactly how his team uses AI to build AI, including where the models still get stuck. His bet is that the more people can see how this works, the better prepared everyone is for what's coming. 0:55 Can open models keep up with the frontier? 2:05 What it takes to reach the frontier 3:55 Open weights and the fast-follower playbook 8:25 Chinese labs: equal talent, less compute 9:55 Why distillation is harder than it looks 14:55 Is distillation theft or competition? 16:20 The lab flywheel vs owning your weights 21:45 Can general models beat every specialist? 26:30 AI at 1% of today's cost in 4 years 27:15 The RL env boom and reward hacking 31:30 What Trillium will build in the open 35:30 Breaking into AI in the age of AI slop 38:05 Open-sourcing how AI builds AI 44:25 Why Nathan isn't pre-training 49:45 Agent swarms and why lab secrets leak 54:25 Fast takeoff vs risks already here YouTube: t.co/001eb2u6Rh @natolambert @trillium_labs @lishali88

Vídeo

Until recently, AI agents mostly handled short tasks, like one support call or one email. Now they can stay on a job for weeks or months. That matters for "relay race" work that runs through a long chain of people. Every handoff in that chain is a chance for something to fall through: a slow callback, a dropped follow-up, lost context. Take a personal injury case. Today it moves from intake to investigation to the demand letter, with different people at each step. A single case agent can work it from the first call to settlement, chasing witnesses, records, and bills, and keeping the client and attorney updated along the way. The same idea applies to processes like mortgages, insurance claims, and enterprise sales. Because the agent owns the end result, the company selling it can charge on the outcome (a closed loan, a settled case, a signed deal) and move past pricing on the labor it saves (paralegal hours, call center seats). Full piece from a16z's @dhaber on long-running agents in high-value customer journeys: t.co/aHaatcizcd

a16z crypto's Eddy Lazzarin on why AI makes the world more secure: bugs are finite, and the cost of finding them keeps collapsing. "Making comparisons between models from a year ago versus today... Some people were coming up with estimates like it's... 99% cheaper for intellectual capability per dollar." "Pair that with a frame we've talked about before, which is that the number of bugs in software is finite. Then we're asymptotically approaching zero bugs. Maybe we actually just hit zero at some point." "If that's the case, we are net improving the security status of the world instead of eroding it." "This is not a battle of wits where you have a system, I have a system, your system is smarter than mine, therefore I am vulnerable to you... That relies on there being vulnerabilities, and vulnerabilities can be removed." "The primary resource we dedicate to mitigating it is a thing that's getting radically cheaper every couple months. That's AI." Dan Boneh: "Companies actually are already doing this. They're spending a lot of money on scanning their code base with the latest models... Every time there's a software update... Every time a new model comes out, they scan the software again." @eddylazzarin @danboneh

Vídeo

a16z crypto's Eddy Lazzarin and Stanford's Dan Boneh on how AI's math leap could turn a decade of trust-building in cryptography into one month: Eddy: "We specifically rely in some sense on time... We put these schemes out and we wait. And we wait because what we believe is happening is that experts, like cryptographers and mathematicians, are testing and trying to break these things." "Because of the pace of progress on AI's mathematical abilities, there is a conspicuous absence of time... A whole new well of cognitive ability to scrutinize has appeared." Dan: "In the past, when someone would make an assumption and build a system based on it, we'd have to wait like 10 years before that system could be deployed." "Now we can just let a model run for a month, try to break it, and if the model can't break it, boy, do we have a lot of confidence that the assumption is valid." Eddy: "The labs can say, 'We took our tip-top, totally unrestricted, most cyber-capable model, and we told it to go absolutely insane, and we spent $20 million, $50 million trying to break this scheme.'" "We actually have the prospect to have more confidence in our security than ever before." @eddylazzarin @danboneh

Vídeo

Travis Kalanick on the founder's paradox: the better you get, the harder it is to stay fired up. "The young entrepreneurs that are coming up right now don't totally get it." "When you get into your 3rd and 4th company, I don't know, this may be even my 5th at this point, you get fricking really good at doing stuff." "Things that used to take me a day or three days, a lot of stress, anxiety, like, 'How am I gonna do this?' take me like 45 minutes. And that's a beautiful thing." "I'm so used to adversity at this point that it's kind of normal, so I don't get as pissed off. But being pissed off makes you good as an entrepreneur, so I gotta make sure I'm still fired up." "You get too used to adversity, and that's kind of a problem." Full conversation: t.co/UKMyTXD0zM @travisk

Vídeo

Patrick Collison with Amjad Masad on why AI will be like food: essential to everything, and still not where most of the value goes. Amjad: "There's also this trend of saying that the foundation lab companies are gonna be able to capture a big part of at least new business creation, but perhaps the economy. It's like there's no reason to compete with ChatGPT or Claude because they're just gonna roll over us." Patrick: "Without food, there really would not be much economy or humanity. It is a truly necessary ingredient in production. And yet, most of the economy is not captured by food producers." "The idea that AI is going to be hugely important and abundant in the future does not necessarily mean that all the value is gonna be captured by these AI labs." "The value of any company is in proportion to what its customers can pay it. And so if there's one company in the economy, that's in fact not a very valuable company... Prices and value accrue to that which is scarce." "I think it would be a terrible reason not to start a company because you think that... any given company or sector will gobble everything up. That's not how things have worked through history." Full conversation: t.co/WHFneVZELe @patrickc @stripe @amasad @Replit

Vídeo

Gavin Baker on the AI race as the highest-stakes game of corporate chess ever played: "You've got to give Meta a lot of credit. They were out of the game, and they got back in the game. It's just kind of amazing." "Who could have imagined a year ago, when Gemini was ascendant, that Gemini wouldn't even be in the conversation, and Muse and Meta would be significantly ahead of them from a capability perspective." "Some people have made bad moves, they've made good moves. You've seen some people come out of the game, others come back in." "I don't know if we're going to call it multi-model, a hybrid model. I don't know what terminology the world is going to settle on, but I think that's the future." @GavinSBaker @DavidGeorge83

Vídeo

optimism wins

.@flexport CEO Ryan Petersen on the Charlie Munger advice he passes on to AI founders: "The key to success is dumb competition." "One of my favorite things that Charlie told me at one of these dinners at his house was when I told him about Flexport, he was like, 'Oh, you have a great business because the key to success is dumb competition.'" "I always say that to people who are building AI companies. What a nightmare. Your competitors are literally AI geniuses." "You should find a way to apply AI in some other domain so you're not competing with the smartest people on Earth." "It's like trying to be a boxer. You might beat all the people in your local gym, but at some point you're gonna run into somebody who's gonna kick your ass." @typesfast w/ @shaneparrish

Vídeo

t.co/URbd8O8R0Z CEO Dhaval Adjodah and CTO Owen He on building a self-improving AI company where humans and agents work as one system: "One big limit of [RSI] is the human element. I think the way that a lot of AI companies are thinking is, let us use AI to replace humans." "But what we found is that there's so much that good researchers and good engineers can bring to the table that's gonna be very hard to replace by agents." "And so that's why from the very beginning, we've tried to think of our basically entire company as a multi-agent system, where humans are a kind of agent that's cohabitating with other AI agents." "We have one of our engineers that we named the Viceroy of Variance. That's a real thing. His job is to go and destroy the results we have and find what's wrong." "So far we have not found a replacement for what we call quality time. Literally running hundreds of experiments, looking at it, developing an intuition, trying to figure out what is wrong, what is not wrong... what is just variance, what is not." @makermaker_ai @_dval_

Vídeo

a16z's Martin Casado with t.co/URbd8O8R0Z CTO Owen He on why a lot of RSI pitches are selling something as old as the steam engine: Martin: "I see a lot of pitches come in, and a lot of them use the term RSI. And I'll be like, 'Oh, so what do you do?' And they'll say, 'We use AI to create AI.' 'That sounds very interesting. So what do you do?' 'We use AI to create GPU kernels.'" Owen: "RSI is definitely an overloaded term today... The important thing when you are differentiating them is to ask the question, what is the optimizer and what is the optimizee? When you're self-improving, who is improving?" "There are people who are saying, 'Okay, we're doing RSI. We have humans using AI to develop better AI.' But that to us is just autocatalysis... the ideas and the improvement come from human brains." Martin: "Most technical epochs have had autocatalytic effects. There's no way you can create a computer chip without a computer... You could build better steam engines because you could get coal faster, etc. So this notion of using the thing to build the thing faster is not new." Owen: "That human brain was never the optimizee. Now, for the first time, we could swap the optimizer as the optimizee... The brain of the agents, which is the LLM, the model, can be the thing that you're optimizing, and this leads to this reinvestment of cognitive improvement into itself." @makermaker_ai @martin_casado

Vídeo

t.co/URbd8O8R0Z co-founders Dhaval Adjodah and Owen He with a16z's Martin Casado on recursive self-improvement, and how to tell when it's real: A lot of RSI pitches are really just people using AI to build better AI. Tech has always worked that way. You needed computers to build chips. Steam engines got better once you could get coal faster. Dhaval and Owen's bar is higher. The model has to improve itself: its weights, its harness, even the prompt that tells it what to work on next. And the gains have to hold up when you count what they cost. They track everyone's experiment spend on a "bleeder board," and one engineer, the Viceroy of Variance, has the job of breaking their results. Over the past few months, they've started seeing early signs of their system getting better and cheaper at the same time. 0:55 Why most RSI pitches miss 2:25 Optimizer vs. optimizee 5:10 The RSI equation: performance over cost 6:50 Why a closed loop does not mean takeoff 8:40 Humans are agents inside the system too 9:25 The denominator trap: progress divided by capital raised 11:40 Training GPT-2 for $100 13:40 Most RSI papers fail when replicated 16:40 Why you have to co-optimize 17:40 Why inference is the first problem 20:15 The agent rewrites its own prompt 22:20 "It's not a belief. It's happening." 25:00 Why they haven't published 27:50 Intelligence is a vector and Amdahl's law 29:45 The garden hose theory of RSI YouTube: t.co/4JAW8cu1yp @makermaker_ai @_dval_ @martin_casado

Vídeo

John Collison on why @Stripe calls January 1st, 2026 "day one of the singularity," and the record high of new companies being created: "The Unix epoch is January 1st, 1970. We decided that the singularity epoch is January 1st, 2026... We are now in the singularity." "It's not a total coincidence that we picked that date. It was the end of last year that I feel like for many of us it was the 'holy crap' moment when it came to coding models and how they're getting so good." "All the AI folks talk about this idea of recursive self-improvement where the AI uses the AI to make itself better... People refer to the AI as this monolithic thing, but we are clearly in a world where AI capabilities are helping with the rate of improvement in AI." "As a result of intelligence on tap getting much cheaper, new firm creation is way up. We're seeing the highest rates of new firm creation that we've seen in recent memory on Stripe, right now." "Because when you have all these tools, it's easier to start a company, it's easier to actually make it successful." "I think this technology has a lot to be optimistic about. And so we don't see the singularity in any kind of a pessimistic or worried manner. We are in this moment of AI takeoff right now." @collision w/ @mollyoshea

Vídeo

.@heronpower CEO Drew Baglino on the 40-year streak that kept US electricity flat while the economy grew, and why it just ended: "We benefited from energy efficiency from the '70s... until like the 2010s. You saw efficiency everywhere... And all of that energy efficiency more or less negated the growth in energy services." "The energy services were growing over time in the United States, but the net electricity delivered to accomplish those energy services stayed basically flat. And the same thing even happened in data centers." "If you look at the efficiency of compute, even with all the AI things that are going on, in terms of flops per watt, it's really gone up an immense amount from the first data centers to today." "And that has allowed more and more compute capability for the same percentage of GDP electricity production." "Now it's changing... We are seeing electricity growth higher than it's ever been before. And that's not just because the benefits of energy efficiency are tailing off, it's because we are seeing broad based demand for more electricity: compute, transportation, industrial electrification." "Electricity as a carrier is incredibly flexible... whether it's for intelligence or for heating, I see it as a great thing and wanna accelerate it." @baglino @heronpower

Vídeo

Dmitri Dolgov of @Waymo on how to tell if you’re one year away from shipping or ten: "One pattern that I've seen over and over again is do you have convergence of discovery? The problems that you are seeing today, how different are they from the problems you saw six months ago?" "As you're pushing on the system, does it feel like you have stability, and you're just iteratively improving your metrics in the same areas?" "Or does it feel like every six months the landscape is shifting from under you and the problems of today are very different from what you saw a while ago? "If you don't have that stability, then that's often a good indicator that you got some ways to go." "If you think you're six months or a year away from deployment, do you have a scorecard that if it were green today would allow you to deploy with confidence?" "If you don't have one, you're saving that for later, often that turns out to be more difficult than you expected, and there's a long distance to go." "But if you already have one and you're seeing stability in your discovery and stability in your metrics, and it's just a matter of closing the gap, that is a really good indicator that you're getting pretty close." @dmitri_dolgov w/ South Park Commons

Vídeo

Airbnb's Brian Chesky on the classic management philosophy, and why he rejects it: "Go through the list of great CEOs, almost none of them that I'm aware of do recurring one-on-ones... Jensen Huang does not do recurring one-on-ones, famously." "I do think a CEO should meet one-on-one with your people. I don't think it should be recurring... If you wanna get time with me, get a meeting with me. If you wanna call me, call me. Bump into me, talk to me. But most of my recurring meetings are group meetings." "If there's lots of one-way kinda sensitive topics, there's probably something wrong in the company... If we're having all these sensitive conversations, well, why are they so sensitive? Why can't other people hear about them?" "If I have a one-on-one with you and you tell me there's a problem and we debate it and we say, 'Here's the solution,' we haven't consulted anyone else. So now we just made a decision and no one else helped, consulted, or knew how we made that decision." "'Hire great people, trust them, let them do their job, stay out of the details, do recurring one-on-ones.' I don't believe any of that, and I wanna acknowledge that many great leaders do." @bchesky w/ @sampjacobs

Vídeo

Figma's Dylan Field on why design is the new code, and where the value goes as models get better: "If you believe that coding models will continue to get better, then you also believe that it's more outer loop architecture, the decisions you make about how to go build a system that matter, not the code itself." "The code is material, and then the value rises to the top and that becomes brand, it becomes product design, your marketing, your point of view in the world, and I think overall your judgment calls and strategy." "Product design is really at the center of a lot of that." "For designers, they will have a bigger role to play if they step up. Not everyone will step up. Some people will be like, 'But I liked what I was doing.'" "Overall the way you'll shape products is through design." @zoink w/ @sophiadew on @MTSLive

Vídeo

Call center job losses started in wealthy countries, now they’re everywhere High-income markets went negative in 2022, and lower-middle-income countries flipped in 2025 Charts of the Week: a16z.news/p/this-week-in…

China generated half as much electricity as the US in 2005 Today it generates more than 2x

Figma's Dylan Field says our definition of "slop" keeps silently moving: "I said, 'AI will lower the floor and raise the ceiling.' And I had to correct myself, because actually that's not what we're seeing yet." "We're seeing it lower the floor. It's more democratized to do something. But the ceiling doesn't seem to have changed." "Anything you can verify, AI's great at. Math being number one, maybe cybersecurity number two... You get five designers in a room, they can't agree all the time on exactly what they should do. And that's what's interesting about design. There is not one solution." "A lot of the stuff that we call slop wouldn't be stuff we'd call slop a month ago or two months ago. As the models change, our definition of slop changes." "The craft that you imbue into something is reflective of your thinking. It's reflective of your care. People pick up on it. Whether they can articulate it or not, they can tell." "It creates a lower bound that you have to exceed, but it doesn't define the upper bound. You define the ceiling." @zoink w/ @sophiadew on @MTSLive

Vídeo

Since ChatGPT launched, call center jobs slowed or fell in every country listed below, except China Other white-collar roles kept growing Charts of the Week: a16z.news/p/this-week-in…

Energy capacity = economic capacity, in one chart There are no low electricity, rich countries

Call center jobs grew ~4% annually for 15 years Now they’re shrinking 4% annually Charts of the Week: a16z.news/p/this-week-in…

Token prices keep falling and GPU rental prices keep rising Intelligence gets cheaper, usage goes up, and a model ends up in every piece of software

AI deflated as much in 3 years as PCs did in 15 Cheap intelligence is what lets a model move out of the chatbot and into the software where the work happens

TypeSafe AI's Diogo Almeida on why the boring SaaS companies are best positioned to win AI: Ben: "When the coding agents came out, it was the SaaSpocalypse, and all their values dropped through the floor. Then when Jev came out, every SaaS company is like, 'This is the greatest thing ever.'" Diogo: "In the SaaSpocalypse story, the story I feel has panned out really poorly is that software is very cheap and perhaps easy to replicate. I could believe the former. I could not believe the latter, because a lot of the stuff happens beneath the hood." "I think SaaS will be one of the largest winners of the whole AI game." "I want to work really, really well with the biggest, most boring SaaS companies because I think they're the best positioned to know what workflows to automate, what people need. That's their bread and butter." "I'm not going to forecast anything about the financial markets. But as far as capabilities go, I think it's going to be like an inverse SaaSpocalypse. I'm so jazzed about it." @CompleteSkeptic @typesafeai @bhorowitz @martin_casado

Vídeo

Ben Horowitz and Diogo Almeida on TypeSafe AI's "build prod, not God" approach: Ben: "You said something there that is so unusual in today's world, which is, 'AI is really, really fun.' The company has such a different demeanor and view of AI than everybody else." "If we had any other kind of big lab leader, they'd be trying to, even if they had joy, cover that. Your view is so different. You're like, 'No, we're going to create a way better world, and it's going to be awesome.'" Diogo: "I think [most people in the AI world] don't get it. It's just a classifier complaint. It's an ML-level concern while everyone else is having a Jev party." "We can do all the things that we wanted to do, and I think if you don't get developers, it'll be hard to understand what's really going on." "I do think there's a pretty negative world painted that I obviously disagree with... The fact that AI has so much potential is what made it really tragic for me that we had not released this. So now it's a little bit of a party for me." @CompleteSkeptic @typesafeai @bhorowitz

Vídeo

TypeSafe AI's Diogo Almeida on why Jev is called Jev, and what cheap intelligence finally changes about software: "It's wild that AI is so cool and software has been unchanged in 10 years... The most we can do is add a little chatbot in the side sometimes that can take actions, but not all actions." "TypeSafe is making AI for software. We want to make AI powerful, not just for humans in the loop, but to actually build real software." "It's about the laziness to spend 10 hours to do the five-minute task instantly and to never have to do it again. It should be an ROI decision for people who automate stuff. I would just like it to be automatable." "I think that people will just make new kinds of work, hence the Jev in Jevons. New kinds of work once that stuff is doable." "As we have lots of intelligence at lots of different cost and speed trade-offs, the super systems-y types will be making trade-offs. Jev's gonna be 1000x too smart for them. They just want an approximate guess to optimistically route here and there." Martin Casado: "The good news is we get to rebuild systems again. We have a new primitive. It's kind of a new way of thinking about doing software." @CompleteSkeptic @typesafeai @bhorowitz @martin_casado

Vídeo

We’re so excited to announce that we’re leading TypeSafe AI's Series A: "Like @TypeSafeAI, we believe the future of SaaS and software is bright. And AI should make software more powerful, and not just automate the recreation of existing software. This spiritual alignment goes back many years. We’ve long been fans of Diogo, the founder and CEO of TypeSafe, and have resonated strongly with his vision that the true bitter lesson is not necessarily about building a new being, but returning the power to builders by giving them the primitives of intelligence then letting them unleash their imagination. The entire a16z team is excited to partner with TypeSafe. Software continues to eat the world." By @JenniferHli, @sarahdingwang, @martin_casado, @pmarca, and @bhorowitz

a16z.news/p/investing-in…

"Energy Capacity = Economic Capacity" Former Intel CEO Pat Gelsinger on why America's power grid is the ceiling on growth: "We went through 10 to 15 years where essentially I was taking coal offline at the rate I was adding renewables, and essentially the nation was flatlined in terms of overall energy capacity." "So essentially our economic capacity as a nation, flat for fifteen years." "Even with that, we've gone to maybe increasing our national energy capacity 4% per year. Going from essentially zero to one." "Nuclear as base load is a fabulous thing that we want to go build more of. Unfortunately all of our renewables have deep dependencies in China." "Energy capacity somewhat dampens how euphoric we can get on AI. Why build a new data center and buy the million GPUs if I can't power them?" @PGelsinger @RaghuRaghuram @appenz

Vídeo

Former Intel CEO Pat Gelsinger on getting hired at 18, designing Intel's chips while finishing his PhD, and arguing with his Stanford professors: "I'm 16 years old. I accidentally win a scholarship. I go to tech school, skip my last year and a half of high school, and Intel comes recruiting when I'm 18 years old." "The interviewer, he writes on his page, and I was number 12 that he interviewed that day... He says, 'Smart, aggressive, arrogant. He'll fit right in.'" "I was engineer number four on the 386. Architect and design manager for the 486. All of that while doing my bachelor's, my master's and PhD work... Learning by day, and putting it to practice at night." "One of the professors postulated a new carry lookahead adder design in one of my classes at Stanford. So I said, 'Well, I'm working on the 386 carry lookahead adder, so let me go try it.'" "I come back two days later to the professor and I said, 'It doesn't work.' The next two class periods were just me and the professor arguing over why his design didn't work." @PGelsinger @RaghuRaghuram @appenz

Vídeo

Former Intel CEO Pat Gelsinger with a16z's Raghu Raghuram and Guido Appenzeller on the next wave of semiconductor innovation and the physical constraints shaping the AI buildout: Drawing on his experience designing Intel's 386 and 486 processors, Pat explains how AI could transform chip design, but also why faster design alone won't solve the industry's biggest problems. They examine the bottlenecks in manufacturing, memory bandwidth, advanced packaging, and power, and why today's explosion of specialized AI chips may eventually consolidate around a smaller number of architectures. They also discuss the potential for new memory technologies, the shift from copper to optical networking, and why energy capacity could become a major constraint on AI growth. Finally, they revisit Pat's VMware years to ask what virtualization might look like when infrastructure is built for agents rather than humans. YouTube: t.co/ZJFvOky68H 1:05 Hired by Intel at 18 3:50 Arguing with Stanford professors 7:25 Will AI design the next gen chips? 8:45 Designing faster than fabs can build 11:55 Why GPUs are stuck with bad memory 13:40 Why 100 AI chips will collapse to a few 19:15 Agents make new chips cheap to run 22:10 No new memory chip since the 90s 27:20 Why 16-layer chip stacks are a fantasy 32:15 The physics that keeps memory next to compute 34:55 "I declared the death of copper" 38:05 AI traffic breaks the internet's design 42:40 "Energy capacity equals economic capacity" 46:25 The 130-year-old power war AI reopened 48:50 VMware for agents @PGelsinger @RaghuRaghuram @appenz

Vídeo

Your next users are machines Agents burn nearly 5x the tokens people do, up 14x in six months

AI agents don't use the cloud like humans do. AWS CEO Matt Garman on rewiring AWS for its newest "users": "Increasingly we're seeing teams that want a cloud that is great to work with agents and not just with people." " We've doubled down on things like starting a database in three seconds and some capabilities that agents are excited about." "Many agents wanna create a database, do a little bit of work, and then have the database go away. You don't really need that database to have five nines of durability." "There are new building blocks that agents are gonna want that we just didn't really need before: compute sandboxes, gateways, agent permissions versus people or service role permissions." " A lot of those are things that we have built and are building and thinking actively about because they are just brand-new building blocks. It's not using the existing building blocks differently." @mattsgarman @RaghuRaghuram

Vídeo

AWS CEO Matt Garman on why he isn't worried about an AI bubble: no single customer dominates AWS capacity, and enterprises are already seeing positive ROI. "We feel really good about the spend we're making now. I get lots of questions about how you feel about that spend, and 'Are you nervous about a bubble?'" "We take a diversified approach, so not all of our capacity is bundled up in one customer." "Also, AWS is where people come to launch their production workloads. The majority of our usage today is core compute, storage, and inference... Those are the workloads that just aren't gonna go away." "Go talk to customers and ask: at today's capability and today's cost, are you seeing positive returns to your business? They'll say, 'Yeah." "It's like the VC model. Is every billion-dollar startup gonna make it? No, they won't. But that's been the game for 50 years. You bet on 10, and one makes it and pays for the others." "You saw that with the internet. There was a bubble, a bunch of internet companies didn't make it, and the internet's still a thing." @mattsgarman @RaghuRaghuram

Vídeo

The CFO job has gone from accountant to banker to builder, and the whole finance function is being rebuilt around it: 1. Data stopped being the bottleneck. AI pulls and reconciles data across systems, so teams spend more time acting on numbers instead of chasing them. 2. The finance stack is becoming part of the finance team. AI-native tools stand up in days, pull their own data, and turn a monthly close into a daily one. 3. The rise of the finance engineer. Finance people proficient in Claude Code or Codex are building their own tools. 4. The finance team gets smaller and higher leverage. The old heuristic of finance at 5% of headcount is heading to 2% for the leanest teams. 5. Planning becomes continuous. Usage-based revenue (tokens, API calls) moves too fast for a monthly forecast, so planning shifts from annual budgets to a live view. 6. Controls become a critical feature of the product. Every number an agent touches keeps an audit trail back to its source, and finance still owns sign-off. 7. Finance gets a bigger mandate. Compute spend, people versus software, and build versus buy are now finance decisions. "The CFO is becoming a builder and architect of the company's operating system... The job is still finance, but more of the job now is designing how finance gets done." a16z's @seemaamble, @IvanoMakSF, and David Borecky on why you need a new CFO: t.co/rODlx5uUy8

AWS CEO Matt Garman says they could sell every GPU to the frontier labs, but deliberately hold capacity back for startups to support the full ecosystem: "We really think intentionally about our allocation strategy. We're great partners with the large frontier labs, the Anthropics and OpenAIs and Metas, and other large customers. We want to make sure we invest in them." "And we're very intentional about making sure we have capacity for startups." "We could sell every single GPU or AI accelerator we have to just the big frontier labs and call it a day. We choose not to, because we want to keep growing the full ecosystem." "The whole ecosystem is more healthy for us, there's some diversification, but it's also that we know these are gonna be big companies over time." "We say yes, in some way, shape, or form, to something like 60% of the requests we get." @mattsgarman @RaghuRaghuram

Vídeo

AWS CEO Matt Garman with a16z's Raghu Raghuram on the biggest cloud provider in the world being rebuilt for the agentic era: $220B in CapEx for 2026, two million NVIDIA GPUs on order, AWS's AI chips sold out through next year, and a new 30-second AWS account with no credit card required so agents can get rolling. 0:55 $170B in revenue, growing 37% 2:40 30-40% of revenue started as startups 4:55 Startups want a cloud for agents 9:10 An AWS account in 30 seconds 12:25 Agents don't need bulletproof databases 16:30 Why AWS won't sell every GPU to the frontier labs 22:00 Why Matt isn't worried about a bubble 24:45 Why AWS builds its own power plants 27:45 Planning for the next supply shortage 30:25 Saving one county $5K per resident 33:10 How AWS built its own chips 40:05 What's holding enterprise agents back 45:35 Why Bedrock never sends your data to the model 49:30 AWS's new AI security tool 51:50 Why AWS teams of 10 are now 3-4 YouTube: t.co/KZD2TR5A0r @mattsgarman @awscloud @RaghuRaghuram

Vídeo

a16z's Alex Immerman, David George, and Sarah Wang on why a data center can lower your electricity bill: "It may seem counterintuitive, but a data center can help lower your electricity rate. A recent US study showed that for every 10% increase in data center capacity, residential rates went down by 40 bps." "You should think about it as a power grid is a shared fixed cost base: poles, wires, substations. A large stable customer, like a data center, can help spread those costs across more units of electricity. Simply put, more demand across a shared system is a positive." "Investment in shared infrastructure can also improve the economics for the households and businesses connected to it." "There's real examples of that today. It's not just lip service." Still, 56% of Texas voters say more data centers would raise their power bills. The catch is it depends on who pays. When developers cover their own grid upgrades instead of pushing them onto everyone else's bills, the savings are real. a16z's @rmcentush on how to get that right: t.co/TobsqsZJq9 @aleximm @DavidGeorge83 @sarahdingwang

Vídeo

The US grid plans to add 86 gigawatts this year. Here's where it's coming from: - Solar: 43.4 GW - Batteries: 24.3 GW - Wind: 11.8 GW - Gas: 6.3 GW - New nuclear: 0 GW The AI labs are planning for hundreds of gigawatts. Solar is the one source already produced at that scale, and most of it is made in China. "Scaling firm power that isn't gas is much more complicated. Uprates and restarts can squeeze a little more from the existing nuclear fleet, but the real upside is new reactors." a16z's @rmcentush on how America powers the next wave of growth: t.co/TobsqsZJq9

.@basepowerco has raised $2.5B to bring affordable, reliable energy to people's homes across America. They're building a new nationwide power company by rebuilding every layer from scratch. Three years in: - Built a home battery factory in 8 months - Installing 100 batteries a day - Powering 30,000+ homes And they're hiring! Open roles in Austin, TX: - Head of Firmware - Backend Engineer (BaseOS) - Product Engineer - Electronics Design Engineer - System Integration Engineer - Deployment Finance & BizOps (Bay Area office opening soon, too.) More roles and the full interview with Head of Software @JaredMGreene on a16z Jobs: t.co/mNd0R7yotY

Vídeo

We're excited to invest in Preference Model. Building RL environments that actually work is harder than it looks. Models are relentless at reward hacking, finding shortcuts, and exploiting vulnerabilities. @preferencemodel has focused on the domain that matters most to labs right now: AI research and ML engineering itself. Over the past year, the team has built RL environments for leading labs. Their focus has been on building the infrastructure to make harder, more resistant environments as models improve: tooling that finds where models are weak, generates new tasks to target those gaps, and tests environments against agents actively trying to break it. This week they're open-sourcing Karotte, the framework they've used in production — hardened through more than a million evaluation runs and red-teaming. We're thrilled to partner with @chem_safety and @Ning_Catsnail and the Preference Model team as they build the training grounds for capable and aligned models. By @JenniferHli

Left: Texas's grid only maxes out about 200 hours a year, roughly 2% of the time. Right: New data centers that agree to cut back during peak hours can skip the wait for more capacity. US grids could add 100 GW of them without building a single new plant. a16z's @rmcentush on how flexible data centers get onto the grid sooner: t.co/TobsqsZJq9

Texas just froze new data center permits: - Its queue went from 63 GW to 474 GW in 18 months, more than 5x record peak demand - Only 9.5 GW is approved and about 4.3 GW is actually running - The rest is duplicates, speculation, and developers who have never plugged in a GPU a16z's @rmcentush on why the grid can't build for that many maybes: t.co/TobsqsZJq9

"At the end of 2024, Texas' grid operator had 63 GW of large new customers in its queue. By this June, that figure reached 474 GW, more than five times record peak demand." "Then Texas hit pause." "Developers routinely spam requests across several sites, and many are speculative builds with no customer yet." "Planners can't tell which are real, and they don't want ratepayers covering upgrades built for projects that never show up." "Fully connecting a large data center can take 5 to 10 years... Idle GPUs cost far more than the electricity to run them, so speed matters most... That's why developers are planning to bring their own power on site." "Full reliability for these power islands is very, very hard, but some labs and hyperscalers have shown that, forced to choose between reliability and speed, they'll pick speed." "Every site I'm aware of intends to connect to the grid as soon as it can, chiefly because on-site power almost always costs more." "AI companies have pledged to pay their way, and most would rather connect to the grid than avoid it. It seems ideal for everyone that we let them do so instead of driving that spending into private islands." "Texas created the energy fast lane first; now it has the opportunity to show us a better one, and I expect it will. Power is where America finds out whether it can still build." a16z's @rmcentush on why Texas is making data centers wait: t.co/TobsqsZJq9

🇯🇵

Instagram turns 16 today

.@Valon just raised a $150M Series D at a $2.3B valuation. Within six months of selling software, they signed over $200M in deals. (And they're hiring!) How the seven-year-old company got there: - Mortgage is $13 trillion of consumer debt running on a system built before the internet, and no servicer will trust a new platform. So Valon became one. - Co-founder Andrew Wang read every federal and state regulation, 18 hours a day for six months, and turned it into code. - They ran their own servicer on the software until it hit 3x the industry's efficiency, sold that servicer to a larger mortgage company, and now sell the software to everyone else. One of the biggest servicers in the US is moving 4 million loans onto it, nearly 10% of the market. Valon is hiring deployment strategists in NY and SF. Read more about their open roles: t.co/ZdA0AsffSZ @xlindadu @wangandrewd

Vídeo

Traffic and revenue tell two different stories Replit is a top 10 consumer AI app by spend and a bottom 5 app by traffic AI power users' money shows up where casual users don't Full list of the Top 50 Consumer AI Apps by Revenue, from @omooretweets on Cosign: t.co/yhzThrEwZx

What are the top 1% of AI spenders paying so much money for? Tools to build, automate, and ship. Versus the average spender, they're: - 23x more likely to pay for n8n - 18x for fal and Manus - 15x for Higgsfield Full list of the Top 50 Consumer AI Apps by Revenue, from @omooretweets on Cosign: t.co/yhzThrEwZx

What are the top 1% of AI spenders paying so much money for that everyone else isn't? Tools to build, automate, and ship. Versus the average AI buyer, they're: - 23x more likely to pay for n8n - 18x for fal and Manus - 15x for Higgsfield Full list of the Top 50 Consumer AI Apps by Revenue, from @omooretweets on Cosign: t.co/yhzThrEwZx

Greg Brockman of @OpenAI on the new security loop: point every model release at your own systems, find the vulnerabilities, automate it. "We took 25% of our production engineers and said, 'Sorry, all your projects are on hold. You are now defending. You are now up-leveling our security architecture. You're going to use the models to find all the holes.' And we found a number of serious issues, and we fixed them." "We found some new problems, but eventually it saturated. We basically have found, to our knowledge, all of the P0s, all of the critical problems that Astra is smart enough to find. And of course, there will be a new model, there will be a new round." "You want to be in this tight loop of new cyber capability drops, you deploy it against your systems, you find the new holes, and ideally, you've managed to automate this, what we call defense factory. That's what we're building internally." @gdb @bhorowitz

Vídeo

The window to patch software bugs is collapsing Of the bugs hackers actually exploit, ~87% are now being attacked on or before the day the bug is public knowledge That share was 23% in 2020

Kevin Mandia on why companies need an agent swarm polling their network like a heartbeat, because the window to catch a new security hole keeps shrinking: "We do a thing called a hyperattack. That's just a fancy word for we throw a drone swarm of agents at you, and we map your network. Every service, every route, every system, all assets." "With that metadata, we now just poll you almost like a heartbeat. What's changed? Did an app change? Did a route change? Did a service get updated? So that we can poll cheaply for change and then attack the change." "What you really want in the AI age is the constant pressure of models attacking you... You do it when either the threat changes, new models come out, new intelligence is available, or your network changes." "That's what we had over the weekend. There was a zero-day in a popular product, and immediately we've already got the heartbeat. We just polled who's got the problem." "Our goal at Armadin is to go from a known vulnerability to knowing whether it's actually exploitable before the bad guys can." @ArmadinSecurity @DavidGeorge83

Vídeo

a16z's Angela Strange sits down with @Valon's Andrew Wang and Linda Du to unpack what it takes to rebuild the infrastructure underneath a $13 trillion mortgage market that still relies heavily on systems designed before the internet. Linda and Andrew explain why Valon chose the hardest path: becoming a regulated mortgage servicer, translating decades of federal and state regulation into software, and proving the platform on its own loans before selling it to the industry. That foundation made Valon roughly 3x as efficient as traditional servicing and created the system of record it is now using to bring AI into complex mortgage workflows. They also discuss what AI makes possible on top of that infrastructure, from agents handling long-tail servicing tasks to voice interfaces and more personalized customer experiences. And they explain why deploying the technology into large regulated enterprises is ultimately as much a change-management challenge as a technical one. 01:02 From Soros to mortgage math 03:00 $13 trillion still on 1960s systems 09:38 Why build the whole stack from scratch 11:47 The 3-5 year licensing gauntlet 15:50 Turning 50 states of regulation into code 18:00 Winning the first customers 22:34 What AI unlocks now 29:47 Everything is servicing 32:52 $200M in deals within six months YouTube: t.co/AlQZtYgf7D @wangandrewd @xlindadu @astrange

Vídeo

Greg Brockman of @OpenAI on why AI that finds security holes for attackers to exploit is a blessing in disguise for defenders: "In this case, you saw both an AI that was able to hack out of a secure environment and hack into a company's production environment." "This capability broadly diffused is something that will really empower threat actors in new ways. Defenders need to use this time before that technology's broadly available to secure themselves." "If you can find vulnerabilities, if you're an attacker, you can use it for no good, but if you're a defender, you can patch. If you're a defender, you control the battleground. You control the setup of your systems." "Our belief right now is that there's this window... use these frontier capabilities that you will have differential access to... so that as the frontier capabilities get better, you get pulled along too." @gdb @bhorowitz

Vídeo

Cyber is having a moment Across 21 major software companies, including Apple, AWS, Microsoft, and Google: - Reported critical vulnerabilities never cleared 100 per month in four years - Since spring they've jumped to over 600 per month

Kevin Mandia on finding 90+ security holes at Fortune 500 companies that nobody knew existed: "When you have an AI-based attack, it'll find logic flaws rather than code flaws in custom applications. It'll exhaust all routes all the time." "Armadin, since January of this year, we have found over 90 zero-days at customer sites, all in production." "We've post-trained all our models with real red teamers, real folks that actually can develop exploits." "When we're scanning networks, we don't have source code to review. We're not finding these zero-days with source code. We're not finding these zero-days because we can log into an app and now we have access, and we can get to other things. We are black box coming from the internet." "Over 90 zero-days in major software companies, and they're thankful. We're coming from the outside, and then we're calling a CISO, usually within 48 hours, 'Hey, we've got remote code execution in your DMZ.' And usually from there we're getting in, and they agree with us." "That's not a pen test. That is like a real adversary coming at you." @ArmadinSecurity @DavidGeorge83

Vídeo

We're thrilled to lead General Medicine’s $120M Series B as they build the general store for healthcare. @GeneralMedicine is built on the belief that facilitating access to care is a core part of delivering it. The company brings patient history, ongoing clinical guidance, transparent pricing, and help accessing care into one experience. Its own medical group combined with a marketplace of external clinicians connects patients almost immediately with services far beyond what a single practice could ever provide. Alongside its direct consumer offering, General Medicine is also building for the enterprise. The same underlying infrastructure for clinical guidance, transparent pricing, and rapid access that General Medicine has already built can be ‘turned on’ for consumers in partnership with any number of enterprise organizations. We’re excited that the PillPack team is back together. Congrats, @tjparker, @elliotcohen, @ashwinkmurali, and the General Medicine team! By @vintweeta, @JorgeCondeBio, @daisydwolf, and @evajsteinman

.@ArmadinSecurity is a new company that uses AI agents to attack your network before criminals do, finding real exploitable zero-days and eventually patching them autonomously. Their agents map every service, route, and system from the outside, then re-attack whenever something changes. Since January that's uncovered 90+ zero-days at Fortune 500 companies. Kevin Mandia spent 30 years responding to some of the worst breaches in the world and built Mandiant, now a core part of Google Cloud's security arm. He had no plans to start another company until AI arrived and changed everything: in his words, "everything I did is dead." He joins a16z's David George to talk about why he came back and what Armadin is building. In this conversation, they cover what AI attacks look like today, why nation states hack like a sniper and AI hacks like a drone swarm, and why the only defense fast enough to keep up is one with no humans in the loop. 1:05 "Everything I did is dead" 3:10 AI on offense, AI on defense 4:20 What an AI attack looks like today 7:20 Sniper rounds vs drone swarms 10:25 Map everything, attack what changes 14:35 The problem with pen tests 16:40 90+ zero-days at Fortune 500s 18:35 A bad patch beats an intrusion 20:55 Why humans are too slow for the SOC 24:30 Why the next year in cyber won't be pretty 26:30 How Mythos changed the threat 28:15 Caging the attacker Armadin built 30:05 Open and closed models tied on offense 34:40 AI is finding zero-days on its own 35:45 Mandiant in 2004 vs Armadin in 2026 41:20 Why Armadin retrains sales every week YouTube: t.co/SHNj95OKTR @ArmadinSecurity @DavidGeorge83

Vídeo

"The top 1% user is spending $903 per month on AI... The median is spending $25." a16z's Olivia Moore breaks down who actually pays for AI, under 5% of US consumers, with the top 10% of them driving half the revenue: "About half of Americans report using AI, and about 25% think they're interacting with AI on close to a daily basis." "The spending data is much more concentrated." "Our Yipit panel shows around 4.5% of US consumers are paying a subscription to an AI product. Some other sources say as low as 2% or 2.5%. It is expanding, that's doubled from a year ago." "Even within that top 4.5% who do spend, the top 10% of those are more than half of the revenue, and the top 1% are 20% of the revenue... The top 10% of users are driving this market." "We could also see what products they're actually spending on. Probably unsurprisingly, it was mostly developer tools, productivity tools, creative tools, things used to build, make, sell." @omooretweets

Vídeo

Which tasks do people actually want AI agents to do? More charts in our Top 100 Consumer AI Apps breakdown: a16z.news/p/top-100-cons…

"Most people aren't looking to save time, they're looking for ways to spend their time." 9 of 15 consumer internet categories have zero AI products in the Top 100. These built some of the biggest companies of the last two eras: - Streaming - Social - Dating - Gaming - Travel - Retail - Finance - Real estate - Jobs More charts in our Top 100 Consumer AI Apps breakdown: t.co/J9edFBQvCN

a16z's Olivia Moore thinks consumer AI has its business model backwards: "Substantially all of the consumer AI revenue thus far has come from direct subscriptions. On our web list, 85% monetize via subscriptions, another 62% via credits or extra usage payments." "Only 13% had ads or other options where you are the product instead of paying for the product." "If you look back at the history of the consumer internet, that's a pretty unnatural inversion. Almost all of the really big consumer technology companies we have now make the vast majority of their revenue from ads or transaction fees, not subscriptions." "The truth is... most people don't have the funds, or don't want to spend their funds, on software." "If you look at the top 20 consumer subscription products globally, ChatGPT is already on there, which is crazy in three and a half years. Almost all the others are media companies or big platforms like Amazon or Uber, where you have very frequent purchase behavior." "So I'm a big believer: we have transcended the need for everyone to buy a subscription to AI, and we need to see these other business models come back." @omooretweets

Vídeo

13 of the 20 biggest consumer subscription bases are media ChatGPT is the first AI app to crack the top 20 More charts in our Top 100 Consumer AI Apps breakdown: a16z.news/p/top-100-cons…

a16z's Olivia Moore on the untapped category in consumer AI: most people aren't looking to save time, they're looking for ways to spend it. "We talked about this in the report, in terms of areas where we have not yet seen AI-native startups pop up." "@ekuyda, one of our portfolio CEOs, runs a company called Wabi. She has this great quote, I'm probably going to botch it, but it's something like: 'Most people aren't looking to save time, they're looking for ways to spend their time.'" "This is why social media, entertainment, Netflix, TikTok, YouTube are the most used consumer products we have." "Much of what we've seen in AI so far is: how do I do this thing a little bit faster? A little bit easier? A little bit more impressively? And that's not an incredibly compelling daily or hourly active value proposition for most people." @omooretweets

Vídeo

The top 1% of AI spenders now outspend the bottom 50% combined They spend an average of $903/month, while the median customer spends just $25 More charts in our Top 100 Consumer AI Apps breakdown: a16z.news/p/top-100-cons…

The seventh edition of our Top 100 Consumer AI Apps is here, including a new dimension this time: what consumers are actually paying for. a16z's Elena Burger sits down with Olivia Moore and Josh Elman to unpack the current state of consumer AI: The data reveals a striking power-user economy. Only a small share of consumers currently pay for AI, but among those who do, spending is heavily concentrated at the top. Olivia and Josh discuss why developers, creators, and other power users dominate spending today, and why subscriptions may not be the business model that ultimately brings consumer AI to everyone. They also dig into the rise of personal agents, the different trajectories of ChatGPT, Claude, and Gemini, how ads could reshape AI economics, and the enormous amount of consumer white space still left to build, from shopping and entertainment to social, dating, and marketplaces. 00:56 Inside the 7th edition 05:55 Muse and Instinct's launch numbers 09:14 The privacy wall for personal agents 14:39 The AI spending power law 20:20 Why subscription is the wrong model 23:51 OpenAI's $1B ad run rate 29:21 ChatGPT vs Claude vs Gemini 32:30 Where creative tools still win 45:12 The white space: dating, shopping, entertainment YouTube: t.co/bxo4NbwAvt @omooretweets @JoshElman @VirtualElena

Vídeo

The seventh edition of our Top 100 Consumer AI Apps is here. New this time: a revenue leaderboard, alongside the usual web and mobile traffic rankings. Three years ago we published the first edition. ChatGPT was #1, Claude was unranked, and the entire category was chatbots, image generators, and not much else. In today's edition: - ChatGPT still holds the throne, now with 1B+ monthly actives on mobile - Claude has climbed to #3 on web with nearly 1B monthly visits - The category has expanded to vibe coding (Lovable, Cursor, Replit), music (Suno), design (Figma), voice (ElevenLabs), video (Higgsfield, Kling), agents (Manus), and even hardware (Plaud) Full breakdown from @omooretweets: t.co/J9edFBQvCN

Ben Horowitz on the rap group he started at 18, and why hip-hop became the source material for how he thinks about business: "I was probably 12 when Rapper's Delight came out... The first one where I was really like, 'Oh, this is gonna be very different,' was when Run-DMC's Hard Times came out. This has a whole other feel to it than anything I've heard." "The thing that made me locked into hip-hop: I went to school in New York. I have this friend who I grew up with who had gotten shot in the face in some kind of gun accident, and he was blind... I'd call home to my mom: 'Is he gonna be okay?' She's like, 'He hasn't spoken to anybody since he's been shot.'" "Being a dumb 18-year-old kid, I was like, 'Hip-hop is so great. You can have nothing and you'll rap about your shoes. He needs that feeling.' So I would record the DJ Red Alert and Chuck Chill Out show. Everybody else would be out Saturday night, and I'd be recording. And I would send him the tapes." "Then I called home to see how he's doing, and they're like, 'Well, he's talking now, but all he'll talk about is hip-hop.'" "When I went out that summer, we started a hip-hop group called The Blind and Def Crew. Blind and D-E-F. We rapped about being blind. I wrote most of the lyrics: 'The Blind Def Crew, you know we're fly. Three of us, but we got four eyes.'" "Hip-hop is very business oriented, particularly in the early days. When I started writing on business, so many of the ideas came from those old hip-hop songs. So I started putting the lyrics at the top of the post to give credit where credit was due." @bhorowitz w/ @MotownIni

Vídeo

Ben Horowitz on the midnight phone call from Kanye that turned his CEO barbecue into a studio session: "I answer and go, 'Ye, what's up?' He goes, 'Hey Ben, I heard you're having a barbecue tomorrow. Can I come?' I was like, 'Sure.' 'Okay, great. And I want to record in your studio.'" "I'm like, 'Ye, I don't have a studio.' And he goes, 'Okay.'" "6:30 in the morning I get a call. 'Hey Ben, I'm Kanye's engineer. I'm just calling about the specs on your studio.' I'm like, 'I don't have a studio.' He's like, 'What do you mean you don't have a studio? Kanye is expecting a studio.'" "I'm like, 'Who do we know who has a studio?' And they're like, 'Howie.' So we call. 'What are you doing, Howie?' 'Watching Netflix.' 'Is that studio you have portable?' 'Yeah.' Long story short, you drove the studio down to my house and we set it up in the room." "A lot more crazy stuff happened. Ye was playing us the Pablo album... That was a legendary day. No doubt." @bhorowitz w/ @MotownIni

Vídeo

Ben Horowitz on what he's evaluating when a founder pitches: the idea is just the lens, the person is the bet. "If you had a perfect rating system and could go, 'Do you want this entrepreneur or this idea?' you'd take the entrepreneur every time." "The problem is it's hard to evaluate entrepreneurs. You don't know that much about them when they come in, and you don't know how they're going to perform under pressure. So you need the idea to help you evaluate the entrepreneur." "What did they go through in their life to get to that idea? How good an idea is it? How much original work did they do to get there? That ends up being the lens at which you see the entrepreneur." "The mistake to make is, there's something you don't like about the idea, and you pass on a great entrepreneur. Or vice versa: 'This is a great entrepreneur,' but you don't realize his idea is no good." "What matters is, are their strengths world-class? Are they the best in the world at something? And then you overlook everything else." "Every entrepreneur we see who gets to us is pretty good. But are they outstanding? Are they better than everybody else at this? That's what makes a company great." @bhorowitz w/ @MotownIni

Vídeo

Replit CEO Amjad Masad on how general models could train smaller, domain-specific models on the fly: "There's a lot of talk of recursive self-improvement, but there's something I don't think is getting a lot of discussion, which is models training their replacements." "You can think of it as a just-in-time compiler. As you're executing dynamic code, the interpreter realizes there's an opportunity to optimize, and it emits machine code on the fly that's a lot more optimized." "You can imagine general models, you're doing something with Operator or Astra, some of the big models, and they realize the use case is limited, or some other agent observing realizes the use case is limited." "General agents have all these flaws, but there's also more potential for them to be harmful, more potential for them to go off the rails." "So the model, on the fly, trains a model that could be its replacement, but is a lot more domain specific. Therefore it's cheaper, less vulnerable to prompt injections, and less harmful for you, because it's less capable." "It's almost like a system that's training machine learning models for specific use cases as it's monitoring the entire system." @amasad

Vídeo

.@OpenRouter co-founder Alex Atallah thinks decision models like Jev could be the alignment layer for agents: "One of the cool potential applications of Jev and other decision models like it is going to be alignment: checking to see if a tool call or an agent-to-agent communication is aligned." "There's just so many tool calls. You really need a cheap, fast model if you're going to block something like that... We have a little prototype running internally at OpenRouter." "Imagine looking at the system prompt and the current tool call, and asking: is this aligned with the original agent's system prompt, and with extra guidelines we didn't tell the agent about?" "Say you have agents instructed to red team a new product. They should not access the internet, and if they ever do, they should stop right away. But you might not want to explain that to them. You want them to act like bad actors." "So having another model check every single tool call, to see if it's aligned with something that wasn't in the system prompt, makes sense. And then structural safeguards too... I think companies are probably going to explore a combination of those." @alexatallah @amasad

Vídeo

.@OpenRouter co-founder Alex Atallah on personal agents: 10 specialized chiefs of staff beat one universal agent. One job each means you can see exactly where it's failing. "The worst part about doing cross-domain joins with your personal agent is that the more work you give it, the more understanding of what's going on you're sacrificing." "And no one is taking responsibility for that sacrifice. Agents don't have any responsibility." "With a universal agent that's doing all things, I can't adjust how much understanding I'm sacrificing." "It points me towards: maybe down the road, the sub-agents people use will be very vertically focused, and we have a chief of staff type agent that coordinates between them." "You need vertically focused agents where this agent is more responsible, psychologically, for these things. I want quality checks that make sure it's doing those things correctly. It doesn't need to focus on anything else, it has one focus area." "I wonder if that's going to help people get at least a loose sense of responsibility on top of agents." @alexatallah

Vídeo

.@OpenRouter co-founder Alex Atallah on the "everyone is building the same thing" take: "This reminds me of this tweet I saw... Everybody's building an agent loop with notifications, third-party connectors, context management, memory, sandboxes, agentic web search, an always-on agent on top." "This product is showing up everywhere. It is showing up everywhere, but these are just the new table-stakes primitives." "A 2005 version of that tweet would be: 'Oh, everybody's building the same thing. A database, a users table, a sign-in page, a sign-up page, a profile page, a logout page. Everything's the same.'" "There's like a lot of differentiation really. There's table-stakes needs for AI just like there's table-stakes needs for the web." @alexatallah @amasad

Vídeo

.@OpenRouter co-founder Alex Atallah on what convinced him to do the Stripe deal: full autonomy, faster execution, same mission. "We were not thinking about [selling] at all. I did really respect the company, and do really respect it today. Of possible acquisition options for us, it was my top choice." "As we fleshed out the reasons why it would make sense for both companies, it got more and more interesting." "It was really clear how aligned they were with us having autonomy over the brand, roadmap, and product, and keeping OpenRouter doing what it's already doing, just much faster, with a much more serious go-to-market plan." "There was a bigger picture alignment too. Both Stripe and OpenRouter really want lots of new companies in the world. We don't want everyone to be part of one giant company." "We want to create really good incentives and streamlined workflows for people to start new companies, on top of reliable and price efficient infrastructure and a marketplace that works." "I really want that future, and Stripe demonstrated they've been wanting it for many, many years. In many ways, payments and inference are going to blend together for companies of the future." @alexatallah @amasad

Vídeo

.@OpenRouter co-founder Alex Atallah, in his first podcast since Stripe acquired the company, joins @Replit co-founder Amjad Masad and a16z's Erik Torenberg on why the future of AI is independence and specialization. In this conversation, Alex walks through how the Stripe deal unfolded, why he wasn't originally looking to sell, and why "payments and inference are going to blend together." Pre-OpenRouter, the typical AI workflow had one model provider to choose from, and little pressure on that provider to lower prices. Now enterprises are diversifying across labs and open-weight models, and every board is asking about AI costs and benchmarks. Amjad argues if your company depends on one AI lab, it can turn into your competitor. So Replit is building the layer that lets enterprises use any model and any cloud, without being locked into either. Alex and Amjad are split on personal agents – Amjad runs one agent across his whole company and loves the cross-domain joins, while Alex says general agents cause you to sacrifice understanding, and argues 10 specialized chiefs of staff beats one superagent. 0:45 How the Stripe deal unfolded 5:05 Why mixing models beats one model 7:25 Forcing the labs to compete on price 8:50 Enterprises want open-weight models 10:30 Every board asks about AI every month 12:25 Why companies must own their intelligence 14:15 Replit as the independence layer 15:10 Everyone is building the same agent 16:35 Why Amjad built bring-your-own-cloud 18:10 Amjad's agent that runs his whole company 19:55 Why 10 specialized agents beat one 23:35 Machines, not humans, should specialize 27:30 Guardrails for agents talking to agents 31:10 Models training their own replacements 33:45 Most tasks don't need a frontier model 40:50 Training small models on Qwen 8B 43:25 The Rust cycle is coming for AI 45:10 Fusion models: frontier quality at half the cost YouTube: t.co/stL1g14F3q @alexatallah @OpenRouter @amasad @eriktorenberg

Vídeo

America has 16,876 machine shops. 83% employ fewer than 20 people. Just nine employ more than 500. New defense companies are moving from prototypes to mass production, and every missile, drone, ship, and rocket they ultimately produce needs to run through one of these small shops. That friction is a national security liability. Connor Love and Collen Larson on the next generation of American manufacturing that gets built by smoothing it: t.co/iqaYhGrK32

Lio co-founder and CEO Vlad Keil on the KPI for the company's forward-deployed engineers: automate your own job, then move to the next one. "When you look at the org structure of Lio, 85% of the people are engineers." "And even the people who don't have an engineering title mostly have an engineering background. The reason for that is we don't want to be a consulting company." "There is a lot of forward-deployed work to do if you go to enterprises because they have different nuances in their processes. But how we work is... building a product in a way where we're reducing this customization, but also where it's a lot of self-service." "The job of our FDEs is on the one hand making it self-service, but internally it's automating their own job. Their KPI is... literally your job is to automate yourself. And then if you automate yourself, you go to the next task." @askvladi

Vídeo

It takes *thousands* of suppliers to build one missile, one aircraft, one ship, or one drone. At the top of the chain is the company that delivers the finished technology. Beneath it, makers of engines and guidance sensors, actuators and battery packs, circuit cards, and the machine shops that mill, coat, and heat-treat every part. Silicon Valley funded the top. @ConnorLoveCA and Collen Larson on the case for the rest: t.co/iqaYhGrK32

This is what happens to a rocket engine when the people who design it and the people who build it sit in the same room. SpaceX builds roughly 80% of Starship in-house because when you're solving a once-in-a-generation problem, design, manufacturing, and testing have to live in one loop. Most companies are not SpaceX. They don’t have suppliers in the room with them. And every new defense company can't rebuild American industry inside its own walls. @ConnorLoveCA and Collen Larson on why the suppliers, and the capital behind them, are the next defense-tech opportunity: t.co/iqaYhGrK32

"When we first backed Sigil Wen, we knew we were backing an unusually resourceful builder. He seeks out exceptional people, finds ways to help them, earns their trust, and keeps choosing harder problems." "Sigil came to San Francisco at 17 without much of a network. For a time, he slept at a WeWork. He skipped college to work on AI and spent nearly a year in a hacker house, building alongside researchers and engineers who helped create DALL·E, Whisper, GPT-3, and Stable Diffusion." "We're thrilled to lead Conway Research’s first round of funding and partner with Sigil as he builds Underdog." t.co/ysYl6YRo7e @0xSigil @ConwayResearch @underdogdotai

a16z's Seema Amble on why AI-native startups still beat incumbents: the incumbent owns the system of record, the startup owns the whole job. "Why do you need an AI-native startup if you've got Claudeforce... You've got all your data, and your employees are used to the product. So why another product?" "I absolutely still think there's a case for the AI-native startup, and it centers around the fact that the legacy incumbent is limited to their system of record, and they're not completing the end-to-end job." "Say a customer calls and says they got charged after they cancelled. Resolving that isn't just the customer going into the chat and saying 'Hey, I got overcharged.'" "The response there has to hit billing, it has to look at all the chat history, it has to look at the contract. That's not one system of record, that's the knowledge around that customer and everything it touched." "The opportunity for the AI-native startup is to say: we're going to own that entire end-to-end arc. That could be legal, owning everything from brief all the way through trial... It's really the concept of owning the end-to-end work." @seema_amble

Vídeo