WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=eBUyTS7SzV4 NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/eBUyTS7SzV4.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:01.309 --> 00:00:03.309 [music] s2 00:00:13.280 --> 00:00:14.440 Okay, great. s3 00:00:14.440 --> 00:00:15.560 Hey everyone. s4 00:00:15.560 --> 00:00:17.640 How's everyone doing? s5 00:00:17.640 --> 00:00:18.600 All right. s6 00:00:18.600 --> 00:00:21.440 Are we ready for the revolution? s7 00:00:21.440 --> 00:00:24.200 Okay, Theo just asked the right question. s8 00:00:24.200 --> 00:00:27.400 What do we build now? s9 00:00:27.400 --> 00:00:30.440 I'm going to answer it from the other side of the table. s10 00:00:30.440 --> 00:00:43.320 Uh I'm a founder, I'm an investor, and I run a 20-year-old institution that is becoming AI-native right now, which is a strange and wonderful thing to do to a 20-year-old institution. s11 00:00:43.360 --> 00:00:57.040 Uh and I'll spend about 20 minutes um talking about, you know, what YC is we're trying to build companies where one person does what it took to two it s12 00:00:57.040 --> 00:01:00.880 One person does what used to take a thousand people. s13 00:01:00.880 --> 00:01:03.000 And I don't mean that as a metaphor. s14 00:01:03.000 --> 00:01:09.560 I mean that me- mechanically this year, the people in this room will do this. s15 00:01:09.560 --> 00:01:18.920 In about an hour, some of you will walk into the startup battlefield, and I want you to walk in knowing what's actually possible right now, because what is possible now s16 00:01:18.920 --> 00:01:24.800 is much much bigger than what people believe. s17 00:01:25.080 --> 00:01:32.400 So, let me start with a number and uh you know, I got torn apart on the internet for this, but I'm going to say it again in front of all of you anyway. s18 00:01:32.400 --> 00:01:38.240 Uh this is the one room in the world that will stress test it, and I'd rather stress test it with you myself. s19 00:01:38.240 --> 00:01:44.640 In 2013, I was a YC partner uh building the internal social network um at at YC. s20 00:01:44.640 --> 00:01:50.400 Uh I was also investing in companies, but you know, I was also uh near full-time engineer. s21 00:01:50.400 --> 00:01:56.200 Um and when I was doing that, I could maybe do about 14 usable logical lines of code a day. s22 00:01:56.200 --> 00:02:02.200 Take out of the comments, take out all the and that's how many lines of code I was writing. s23 00:02:02.200 --> 00:02:07.720 And if you look at the literature from that era, um you know, that's kind of normal. s24 00:02:07.720 --> 00:02:10.520 Like some people write 15, some people write 50. s25 00:02:10.520 --> 00:02:16.959 It was not, you know, the thousands of lines of code that I know a lot of you in this room are actually writing now per day. s26 00:02:16.959 --> 00:02:18.800 Um that's about median 15. s27 00:02:18.800 --> 00:02:20.920 That was me at full effort at that time. s28 00:02:20.920 --> 00:02:23.640 This year I run uh YC full-time. s29 00:02:23.640 --> 00:02:28.440 Uh same person, same hours, actually way less hours weirdly. s30 00:02:28.440 --> 00:02:31.720 Uh you know, but I have a 5:00 p.m. kid pick-up now. s31 00:02:31.720 --> 00:02:35.120 And I did the math on my output, and it's about 400X. s32 00:02:35.120 --> 00:02:40.959 Now, before the skeptic in the third row right there deflates the number from for me, let me deflate it myself. s33 00:02:40.959 --> 00:02:43.240 If you don't trust the raw code, well, fine. s34 00:02:43.240 --> 00:02:50.120 Take the most pathological verbosity plant penalty you can stomach, and assume the agent writes bloated code. s35 00:02:50.120 --> 00:02:51.840 Assume half of it is scaffolding. s36 00:02:51.840 --> 00:02:53.640 Assume I'm flattering myself. s37 00:02:53.640 --> 00:02:57.080 It's still 8X at the floor and 80X in the middle. s38 00:02:57.080 --> 00:03:00.160 That number is large, no matter how you torture it. s39 00:03:00.160 --> 00:03:05.600 And here's the part that matters, the part that I'd tattoo on the inside of everyone's eyelids if I could. s40 00:03:05.600 --> 00:03:07.880 It's not the model. s41 00:03:07.880 --> 00:03:11.600 The 2X people and the 100X people are using the exact same Claude. s42 00:03:11.600 --> 00:03:16.480 Same weights, same context window, same API. s43 00:03:17.480 --> 00:03:20.040 So, the leverage is not in the weights. s44 00:03:20.040 --> 00:03:22.720 It's in how you wire the work. s45 00:03:22.720 --> 00:03:23.640 And it's not just me. s46 00:03:23.640 --> 00:03:25.400 At YC, we see this all the time. s47 00:03:25.400 --> 00:03:32.320 In the winter 25 batch, a quarter of the companies had code base code bases that were 95% AI-generated, and that was a year ago. s48 00:03:32.320 --> 00:03:38.560 That batch has become the fastest-growing, most profitable batch in the history of YC. s49 00:03:38.560 --> 00:03:45.320 94 companies total have now crossed a hundred million in dollars in revenue from a seed check in the history of YC. s50 00:03:45.320 --> 00:03:48.360 So, uh I think we know what we're talking about here. s51 00:03:48.360 --> 00:03:58.480 And I can't prove that the AI-generated code caused the growth, but what I can tell you is the fastest growing founders we fund are not treating AI as auto-complete. s52 00:03:58.480 --> 00:04:00.840 They're treating it as a workforce. s53 00:04:00.840 --> 00:04:06.080 The companies that wired the work differently are the ones that are bending the curve. s54 00:04:06.080 --> 00:04:09.840 So, what does wiring the work really mean? s55 00:04:11.560 --> 00:04:13.480 This is the heart of the talk. s56 00:04:13.480 --> 00:04:16.200 And this is what I most want you to steal. s57 00:04:16.200 --> 00:04:20.959 Everything we've learned building with agents maps to an organization. s58 00:04:23.000 --> 00:04:24.623 Sorry, there's no slides. s59 00:04:24.623 --> 00:04:26.040 [laughter] s60 00:04:26.040 --> 00:04:26.840 I have no slides. s61 00:04:26.840 --> 00:04:29.080 I'm so sorry. s62 00:04:29.208 --> 00:04:31.208 [applause] s63 00:04:31.760 --> 00:04:35.280 A skill file is an employee. s64 00:04:35.280 --> 00:04:41.440 It has one capability, one job, written down clearly enough that someone can execute it. s65 00:04:41.440 --> 00:04:52.000 Uh a resolver table, uh the thing that many of you, you know, when you run into Claude code and it says your context is too big in Claude.md, you run off and create a resolver table. s66 00:04:52.000 --> 00:05:02.600 Well, you know, it and if you don't know what that is, it's literally like whenever you need to uh alter a test, load tests.md, you have a whole table of these things. s67 00:05:02.600 --> 00:05:05.000 Um that's an org chart. s68 00:05:05.000 --> 00:05:10.680 A task comes in and the resolver decides who handles it and where it goes. s69 00:05:10.680 --> 00:05:13.560 Uh filing rules are your internal process. s70 00:05:13.560 --> 00:05:21.520 So, this can be um you know, whether or not the resolver is actually working and uh is you know, is there a is it actually in compliance? s71 00:05:21.520 --> 00:05:23.400 And uh trigger e-vals. s72 00:05:23.400 --> 00:05:32.480 So, going in and actually having a test that says, "When I need to alter a test file, does test.md actually get loaded?" s73 00:05:32.480 --> 00:05:34.880 Those are performance reviews. s74 00:05:34.880 --> 00:05:36.600 So, you know, what have we done? s75 00:05:36.600 --> 00:05:45.600 Like literally every part of an organization, the organization that you used to have to hire a thousand people for, I just told you what those things are. s76 00:05:45.600 --> 00:05:48.880 They're markdown files and other types of markdown files. s77 00:05:48.880 --> 00:05:52.200 And maybe there's some TypeScript in there, too. s78 00:05:52.200 --> 00:05:58.800 We've been building organizations this whole time, but we didn't have a management layer, but now that's what we have. s79 00:05:58.800 --> 00:06:09.240 When you sit down with Claude Coder Codex, you're not writing software, you're hiring, training, and managing a workforce made of markdown. s80 00:06:10.480 --> 00:06:14.560 Uh and you know, that's there are tons of companies at YC that are doing this. s81 00:06:14.560 --> 00:06:21.919 Uh Emergence, an AI app builder out of Summer 24, they went from public launch to nine figures of ARR in eight months. s82 00:06:21.919 --> 00:06:26.760 Uh when they crossed $15 million ARR, uh they were only 15 people. s83 00:06:26.760 --> 00:06:31.080 Retail out of Winter 24, it's at $60 million with about 40 people. s84 00:06:31.080 --> 00:06:34.720 The kind of that kind of revenue per head did not exist before. s85 00:06:34.720 --> 00:06:38.680 Not in software, not in oil, not in railroads, never. s86 00:06:38.680 --> 00:06:40.520 These are not freaks of nature. s87 00:06:40.520 --> 00:06:45.160 They're just the first companies built natively on the new physics. s88 00:06:45.160 --> 00:06:47.400 And so, how do companies like that actually run? s89 00:06:47.400 --> 00:06:51.880 Not by hiring hundreds of people for sales, support, ops, and finance. s90 00:06:51.880 --> 00:07:04.640 The AI-native companies that I see inside YC encode all of that as skills, written procedures that their agents execute, and they hire they hire engineers whose job it is to maintain those skills, s91 00:07:04.640 --> 00:07:07.000 to do the work the skills can't do yet. s92 00:07:07.000 --> 00:07:12.400 That is an AI-native company, and it's not a thought experiment. s93 00:07:12.400 --> 00:07:15.440 It'll actually file your taxes if you have a skill file for it. s94 00:07:15.440 --> 00:07:18.280 Now, picture YC's batch room. s95 00:07:18.280 --> 00:07:20.360 Like it actually kind of looks like this. s96 00:07:20.360 --> 00:07:29.080 Um 400 companies or 400 founders at long tables, and uh you know, I can imagine every single one of you each with a laptop every single day. s97 00:07:29.080 --> 00:07:35.520 You're doing uh a former person's entire year worth of work in a single day. s98 00:07:35.520 --> 00:07:36.400 That's not the future. s99 00:07:36.400 --> 00:07:37.880 That's actually the bar right now. s100 00:07:37.880 --> 00:07:43.080 And if you're not doing it, your competitor is, and they will eat your lunch politely and thank you for it. s101 00:07:43.080 --> 00:07:46.120 Here's the extension uh most engineering stocks miss. s102 00:07:46.120 --> 00:07:48.240 It's not just the engineers. s103 00:07:48.240 --> 00:07:54.840 At YC, as we make our transformation, it's our media people, our event staff, our finance team. s104 00:07:54.840 --> 00:08:00.680 People have never opened a terminal in their lives are building skill files and cron jobs. s105 00:08:00.680 --> 00:08:11.000 And one of our finance folks just collapsed about a hundred Excel workbooks into a single app she built with our internal open claw and company brain. s106 00:08:11.000 --> 00:08:12.400 She's not a programmer. s107 00:08:12.400 --> 00:08:17.440 She's a manager of agents now and everyone at YC now is. s108 00:08:17.440 --> 00:08:23.840 So, that's why YC can run at the scale it does with a staff that would look like a rounding error at any other comparable firm. s109 00:08:23.840 --> 00:08:28.280 And that's not because we work harder, because it's because we have a different type of org. s110 00:08:28.280 --> 00:08:29.400 And that's whole the whole game. s111 00:08:29.400 --> 00:08:31.800 It's not just 400X engineers. s112 00:08:31.800 --> 00:08:36.880 It's one company that operates at the level of 400X everyone else. s113 00:08:36.880 --> 00:08:43.320 And so, you know, if you remember only one thing about this, I mean, this is one of the things I had to discover along the way. s114 00:08:43.320 --> 00:08:49.040 Like, you actually have to be really, really careful about where the compute computation is actually happening. s115 00:08:49.040 --> 00:09:03.080 It's happening almost always in two different places and all of the bugs, all of the AI engineering that we run into that's a problem, it's usually because something is happening in one side of the equation that should be in the other. s116 00:09:03.080 --> 00:09:06.960 Uh the first area I would say is latent space. s117 00:09:06.960 --> 00:09:08.280 So, the actual LLM. s118 00:09:08.280 --> 00:09:10.160 It's you know, what do you use it for? s119 00:09:10.160 --> 00:09:22.080 Taste judgment, understanding what a human actually wants when they say something vague, the non-deterministic calls, the computation that lives in the model and you steer it with the markdown file. s120 00:09:22.080 --> 00:09:25.320 And then deterministic space is what engineers know. s121 00:09:25.320 --> 00:09:33.400 Like, your your code agents go off and write TypeScript or, you know, maybe they're writing Erlang if you're using Elixir. s122 00:09:33.400 --> 00:09:35.040 Um yeah. s123 00:09:35.040 --> 00:09:37.880 Deterministic space is the second place. s124 00:09:37.880 --> 00:09:41.200 Let's say, you know, this is a real problem that we have for Startup School coming up. s125 00:09:41.200 --> 00:09:49.400 We have 6,000 people or we're going to try and, you know, one of the experiments we're going to try and do is can we seat 800 people at a time, s126 00:09:49.400 --> 00:09:50.920 um, perfectly clustered? s127 00:09:50.920 --> 00:09:56.320 So, the person sitting to the left and the right of you is the perfect person for you to meet at Startup School. s128 00:09:56.320 --> 00:10:01.840 Um, we have to do that in deterministic space combined with latent space. s129 00:10:01.840 --> 00:10:15.440 Uh, the computation, this computation, this actual storage of like where everyone is inside like, you know, this multi-dimensional array of 800 seats, um, it actually must not live in the context window. s130 00:10:15.440 --> 00:10:19.880 The LLM has to do the human part and seat people. s131 00:10:19.880 --> 00:10:25.120 Um, it's actually exactly what you would do if, um, you know, you were a human tasked with this thing. s132 00:10:25.120 --> 00:10:32.080 You would probably have to physically print out 800 pages and go in a a big room and like say like, well, where does this person go? s133 00:10:32.080 --> 00:10:40.960 Only now it can all happen, um, in your computer and it can instead of taking a month, it might be able to you might be able to do it with, uh, you know, couple hundred s134 00:10:40.960 --> 00:10:44.480 dollars worth of tokens and probably 10 minutes. s135 00:10:44.480 --> 00:10:48.160 Um, and so that's, you know, I would argue that's pretty remarkable. s136 00:10:48.160 --> 00:10:53.160 These are things that you couldn't do even, uh, I don't know, six months ago. s137 00:10:53.160 --> 00:11:05.520 Um, which brings me to working memory and, uh, that's, you know, sort of my favorite way to understand it is, um, you and I, human beings, we only hold about seven things in our head at once. s138 00:11:05.520 --> 00:11:07.040 Uh, seven plus or minus two. s139 00:11:07.040 --> 00:11:17.760 It's one of the most famous papers in cognitive psychology and it's why, uh, local phone numbers are seven digits and why you forget the eighth item on your grocery list. s140 00:11:17.760 --> 00:11:32.960 Uh, that's the entire working memory generally of a human being and every institution humanity has ever built, every checklist, every org chart, every filing cabinet is a prosthetic for that limit. s141 00:11:32.960 --> 00:11:35.640 It's kind of a wild thing to think about. s142 00:11:35.640 --> 00:11:38.760 But an AI agent holds a million tokens. s143 00:11:38.760 --> 00:11:40.560 That's about a thousand pages. s144 00:11:40.560 --> 00:11:43.960 I was trying to explain to my 10-year-old what G brain was recently. s145 00:11:43.960 --> 00:11:51.480 I said, "There's, you know, the AI agent can keep about three Harry Potter books sitting open in its head all at once. s146 00:11:51.480 --> 00:11:57.800 And it can find a needle in any of them and synthesize across all three in seconds. s147 00:11:57.800 --> 00:12:00.360 And that's quite magical, actually." s148 00:12:00.360 --> 00:12:05.840 Three Harry Potter books versus seven digits. s149 00:12:05.840 --> 00:12:07.360 I mean, that's pretty awesome. s150 00:12:07.360 --> 00:12:09.120 I mean, I don't know. s151 00:12:09.120 --> 00:12:10.040 Is that AGI? s152 00:12:10.040 --> 00:12:15.640 Maybe not, but it's already a very different operating and regime. s153 00:12:15.640 --> 00:12:22.480 Almost every company on the earth is still running an org that's designed for the seven-digit brain. s154 00:12:23.360 --> 00:12:25.640 But notice what that also tells you. s155 00:12:25.640 --> 00:12:28.240 Three books is a lot, but it's also very little. s156 00:12:28.240 --> 00:12:30.000 Your company is not three books. s157 00:12:30.000 --> 00:12:31.800 Your company is a library. s158 00:12:31.800 --> 00:12:39.360 Every email, every meeting, every decision, its reasoning, every customer conversation, every postmortem. s159 00:12:39.760 --> 00:12:51.520 The question that determines whether your agents are geniuses or goldfish is who decides which three books are open on that desk. s160 00:12:51.520 --> 00:12:53.280 That's context engineering. s161 00:12:53.280 --> 00:12:54.880 And this is what a company brain is. s162 00:12:54.880 --> 00:12:57.720 It's the library plus the librarian. s163 00:12:57.720 --> 00:13:01.320 Now, some of you are already thinking, "This is just rag." s164 00:13:01.320 --> 00:13:06.640 And you're right that retrieval is the primitive, the same way Postgres is just B-trees. s165 00:13:06.640 --> 00:13:09.000 The hard part is everything around it. s166 00:13:09.000 --> 00:13:19.200 What gets written down in the first place into the knowledge wiki, how it gets enriched and linked, what gets promoted to hot memory versus filed as cold reference, who arbitrates s167 00:13:19.200 --> 00:13:21.520 when two facts disagree. s168 00:13:21.520 --> 00:13:23.360 Retrieval is easy. s169 00:13:23.360 --> 00:13:28.440 Being worth retrieving from is the product. s170 00:13:28.440 --> 00:13:29.800 So, I've been building mine in the open. s171 00:13:29.800 --> 00:13:33.480 It's called It works with any harness, but it loves open client Hermes agent. s172 00:13:33.480 --> 00:13:44.440 It's effectively Postgres for agents, a retrieval layer whose job is to figure out for uh for any three you know, for any task, what three books should be loaded into the agent's head. s173 00:13:44.440 --> 00:13:47.760 My personal one started as a rooms full of books or so. s174 00:13:47.760 --> 00:13:58.080 Now it's a warehouse, about 220,000 pages written mostly by my agents from my email, meetings, 20 years of notes, uh and the lived experience of me. s175 00:13:58.080 --> 00:13:59.160 And that's the point. s176 00:13:59.160 --> 00:14:00.240 It's my second brain. s177 00:14:00.240 --> 00:14:13.200 And when a founder emails me about a crisis, before I start reading as before I even finish reading that email, my agent has already pulled every prior conversation with that founder, three portfolio companies that hit the same wall, s178 00:14:13.200 --> 00:14:15.640 and what actually worked for those people. s179 00:14:15.640 --> 00:14:21.040 Uh when my agent does anything, it does it does everything knowing what I already know. s180 00:14:21.040 --> 00:14:24.720 And that's the difference between an assistant and a colleague. s181 00:14:24.720 --> 00:14:27.720 So, let me stress test my own pitch because you would anyway. s182 00:14:27.720 --> 00:14:30.440 Company brains do have failure modes. s183 00:14:30.440 --> 00:14:36.760 Uh a brain nobody curates becomes a garbage dump with great search. s184 00:14:36.760 --> 00:14:41.320 Retrieval will surface a stale fact with total confidence. s185 00:14:41.320 --> 00:14:46.000 Um a bad skill file encodes a bad process forever. s186 00:14:46.000 --> 00:14:47.000 Uh that's bad. s187 00:14:47.000 --> 00:14:49.480 So, primitive the primitive is not memory. s188 00:14:49.480 --> 00:15:03.000 It's memory plus hygiene, provenance on every fact, contradiction contradiction checks when new information collides with the old, and a librarian, human plus agent, whose actual job is pruning. s189 00:15:03.000 --> 00:15:06.720 Treat the brain like a production infrastructure, and it compounds. s190 00:15:06.720 --> 00:15:13.240 Treat it like a dumping ground, and you get a very confident agent that is wrong in ways nobody can trace. s191 00:15:13.240 --> 00:15:21.280 And um here's the discipline that I think, you know, personally uh makes our company brain and my personal AI compound. s192 00:15:21.280 --> 00:15:24.440 Um that's my signature move. s193 00:15:24.440 --> 00:15:31.200 And, uh, you know, it's what I say to every YC company and every, uh, everyone inside YC, which is never do one-off work. s194 00:15:31.200 --> 00:15:37.400 You can open open claw, you can open your harness, you do some work, but then when you're happy, you know, and it'll come back. s195 00:15:37.400 --> 00:15:38.720 It's, you know, kind of a bad job. s196 00:15:38.720 --> 00:15:41.120 It's kind of like an intern that's not that good. s197 00:15:41.120 --> 00:15:44.120 But the great thing is you can just say, "Hey, I didn't like that. s198 00:15:44.120 --> 00:15:44.680 Fix it." s199 00:15:44.680 --> 00:15:44.880 Right? s200 00:15:44.880 --> 00:15:46.360 I'm sure all of you do this. s201 00:15:46.360 --> 00:15:47.880 But don't stop there. s202 00:15:47.880 --> 00:15:53.360 You actually need to at the end of that task, uh, skillify it. s203 00:15:53.360 --> 00:15:56.080 And so I have a blog post on X about that. s204 00:15:56.080 --> 00:16:08.000 You can search for skillify it and, you know, go get that skill file and then just load it into your your own harness and it'll just turn whatever you just did into a skill that you can reuse. s205 00:16:08.000 --> 00:16:11.440 Because if you have to ask for something twice, you failed. s206 00:16:11.440 --> 00:16:15.000 Um, so yeah, if you remember only one thing, it's that. s207 00:16:15.000 --> 00:16:22.000 Like when you, you know, use your AI agent and then when you're done with it and you're happy with the output, skillify it. s208 00:16:22.000 --> 00:16:23.600 It's going to be awesome. s209 00:16:23.600 --> 00:16:29.480 Um, the organization that captures what it learns like this gets smarter every single day. s210 00:16:29.480 --> 00:16:35.640 The one that doesn't wakes up every morning with amnesia, no matter how good the model is. s211 00:16:35.640 --> 00:16:41.640 Model quality is rented, but if you build your brain, you you own that brain. s212 00:16:41.640 --> 00:16:43.080 So, Theo's question head-on. s213 00:16:43.080 --> 00:16:44.280 What do we build now? s214 00:16:44.280 --> 00:16:47.760 Build the AI native company, not a company that just uses AI. s215 00:16:47.760 --> 00:16:51.960 A company that is shaped like what I just described from day one. s216 00:16:51.960 --> 00:17:00.480 A thin team, skill files for everything, the founder still in the code, library, this library, this company brain, your personal AI. s217 00:17:00.480 --> 00:17:02.320 Uh, use G brain if you'd like. s218 00:17:02.320 --> 00:17:04.360 Uh, it's open source and free, but you don't have to. s219 00:17:04.360 --> 00:17:05.880 There are a lot of really good ones. s220 00:17:05.880 --> 00:17:15.880 The the library will compound from the first week and your whole org will be wired to run at about 400X. s221 00:17:15.880 --> 00:17:23.079 And if you want the green field, the thing that I'd build if I were 25 and sitting where you're sitting, every company on this earth is about to need a brain. s222 00:17:23.079 --> 00:17:26.600 The memory layer that means that you never have to re-ask what you knew. s223 00:17:26.600 --> 00:17:29.120 Personal AI that actually knows you. s224 00:17:29.120 --> 00:17:32.680 We're building G brain in the open and MIT open source. s225 00:17:32.680 --> 00:17:38.600 Um I'm not trying to make money from this because I think the layer should be open the way Linux is open. s226 00:17:38.600 --> 00:17:45.040 But the layer itself, company brains, personal context, the librarian that picks the three books, that's all wide open territory. s227 00:17:45.040 --> 00:17:48.680 I hope somebody builds the defining company here. s228 00:17:48.680 --> 00:17:52.840 And I'd like to fund you at YC if you do. s229 00:17:52.840 --> 00:17:55.960 Now, let me be honest in a way that maybe undercuts my own pitch. s230 00:17:55.960 --> 00:17:58.120 You don't need my tools to start. s231 00:17:58.120 --> 00:18:03.560 Open Claw is the Ferrari, I will always recommend it, but Codex is a really good Honda. s232 00:18:03.560 --> 00:18:06.240 It will do 90% of this. s233 00:18:06.240 --> 00:18:09.000 Uh it will not blow your face off, but it will get you there. s234 00:18:09.000 --> 00:18:10.120 Use whatever. s235 00:18:10.120 --> 00:18:13.000 The concepts are the point, not my repos. s236 00:18:13.000 --> 00:18:15.080 You know, think about where the computation is. s237 00:18:15.080 --> 00:18:17.280 Use skill files as employees. s238 00:18:17.280 --> 00:18:18.760 The librarian the librarian. s239 00:18:18.760 --> 00:18:20.960 Never do one-off work. s240 00:18:20.960 --> 00:18:23.560 Those travel with you to any stack. s241 00:18:23.560 --> 00:18:25.640 So, let me land this. s242 00:18:25.640 --> 00:18:32.760 A lot of people in the world right now are terrified about what happens to all the jobs, and I understand the fear. s243 00:18:32.760 --> 00:18:36.680 But I want to say it plainly, that is a failure of imagination. s244 00:18:36.680 --> 00:18:40.040 And the people in this room are the answer to it. s245 00:18:40.040 --> 00:18:43.400 What I just described, you're going to take to your startup. s246 00:18:43.400 --> 00:18:56.120 You will multiply yourself, and every person in your company will multiply themselves, and you will go build the companies that become the beacon for how all of this works in society. s247 00:18:56.120 --> 00:19:00.840 Abundance is not a policy paper, it is shipped software. s248 00:19:00.840 --> 00:19:04.200 I have a friend who has a rare form of epilepsy. s249 00:19:04.200 --> 00:19:15.640 He built a repo of 80,000 markdown files, a company brain for one small boy, and he pushed himself to the absolute edge of human, what humanity knows about his son's exact condition. s250 00:19:15.640 --> 00:19:18.760 No lab, no grant, no permission. s251 00:19:18.760 --> 00:19:22.000 A father, a laptop, and a library. s252 00:19:22.000 --> 00:19:23.520 That's not a side story. s253 00:19:23.520 --> 00:19:28.240 That is the exact architecture I've been describing for the last 20 minutes. s254 00:19:28.240 --> 00:19:37.400 A library of the librarian, the right three books open at the right moment, pointed at the thing this man loves the most in the world. s255 00:19:37.400 --> 00:19:38.800 You can do that now. s256 00:19:38.800 --> 00:19:42.920 Every problem where you thought, I wish I had that person but I can't get them. s257 00:19:42.920 --> 00:19:43.920 You can. s258 00:19:43.920 --> 00:19:49.320 Every code base you thought was too buggy to fix, you can fix all of it. s259 00:19:49.320 --> 00:19:59.280 Every archive too big to read, every data set too gnarly to clean, every ocean you were told not to boil. s260 00:19:59.280 --> 00:20:02.800 We can boil the ocean now. s261 00:20:05.868 --> 00:20:07.868 [applause] s262 00:20:10.453 --> 00:20:11.680 [applause] s263 00:20:11.680 --> 00:20:16.400 And every single one of you can fly, not metaphorically, mechanically. s264 00:20:16.400 --> 00:20:21.320 And you need to to survive, to thrive, to win. s265 00:20:21.320 --> 00:20:24.960 Theo asked, "What we should build now?" s266 00:20:24.960 --> 00:20:26.760 And here's the whole damn whole answer. s267 00:20:26.760 --> 00:20:34.080 Build that AI native company and build it build the thing underneath it, the brain, the memory, the compounding library. s268 00:20:34.080 --> 00:20:37.280 That makes every company after yours easier to build. s269 00:20:37.280 --> 00:20:39.000 Go boil the ocean. s270 00:20:39.000 --> 00:20:40.080 Go write that test. s271 00:20:40.080 --> 00:20:42.680 Go ship that skill. s272 00:20:42.720 --> 00:20:46.600 Some of the companies you're about to watch in the battlefield are already doing this. s273 00:20:46.600 --> 00:20:49.560 Go build the one that does it best. s274 00:20:49.560 --> 00:20:51.760 Thank you.