[
  {
    "startMs": 1309,
    "text": "[music]"
  },
  {
    "startMs": 12480,
    "text": ">> Hi everyone. I'm Varun. I'm one of the"
  },
  {
    "startMs": 14640,
    "text": "co-founders at Long Lake and I'm excited"
  },
  {
    "startMs": 17680,
    "text": "to share a little bit about what we've"
  },
  {
    "startMs": 19280,
    "text": "been up to for the last 2 years."
  },
  {
    "startMs": 21920,
    "text": "It all comes back to a question all of"
  },
  {
    "startMs": 24240,
    "text": "us have asked time and time again."
  },
  {
    "startMs": 27280,
    "text": "The models are getting better,"
  },
  {
    "startMs": 29080,
    "text": "but the real question is how do you"
  },
  {
    "startMs": 31120,
    "text": "actually deploy the AI into the real"
  },
  {
    "startMs": 34120,
    "text": "world?"
  },
  {
    "startMs": 35360,
    "text": "How do you get the models to complete"
  },
  {
    "startMs": 37840,
    "text": "economically relevant tasks?"
  },
  {
    "startMs": 43120,
    "text": "Let me start by saying everyone has seen"
  },
  {
    "startMs": 45640,
    "text": "the demo. Think of the agent"
  },
  {
    "startMs": 47920,
    "text": "automatically booking a flight, the"
  },
  {
    "startMs": 50000,
    "text": "agent automatically completing a ticket"
  },
  {
    "startMs": 53080,
    "text": "in some kind of customer service portal."
  },
  {
    "startMs": 55720,
    "text": "Think of an agent completing a block of"
  },
  {
    "startMs": 57640,
    "text": "code ready to commit and go."
  },
  {
    "startMs": 60560,
    "text": "The reality is we've all seen this and"
  },
  {
    "startMs": 62960,
    "text": "it feels like magic. 2 years ago any of"
  },
  {
    "startMs": 66720,
    "text": "this would have been complete science"
  },
  {
    "startMs": 68480,
    "text": "fiction. The capabilities are real."
  },
  {
    "startMs": 72400,
    "text": "Now, walk with me into a 200-person"
  },
  {
    "startMs": 75200,
    "text": "property management firm."
  },
  {
    "startMs": 76920,
    "text": "Real people, real properties,"
  },
  {
    "startMs": 79640,
    "text": "real dollars, real customers all across"
  },
  {
    "startMs": 82560,
    "text": "the US."
  },
  {
    "startMs": 84760,
    "text": "You would expect AI to show up by now,"
  },
  {
    "startMs": 87840,
    "text": "but the reality is nothing has changed"
  },
  {
    "startMs": 90680,
    "text": "at all."
  },
  {
    "startMs": 94600,
    "text": "Here's the thing."
  },
  {
    "startMs": 96120,
    "text": "This is totally normal and maybe in fact"
  },
  {
    "startMs": 98960,
    "text": "I'd argue this is what we should expect."
  },
  {
    "startMs": 102960,
    "text": "This is true for every general-purpose"
  },
  {
    "startMs": 105320,
    "text": "technology. You know, take electricity"
  },
  {
    "startMs": 107360,
    "text": "for example."
  },
  {
    "startMs": 108560,
    "text": "Electricity was invented in the 1880s"
  },
  {
    "startMs": 111640,
    "text": "and it was first demoed at Edison's"
  },
  {
    "startMs": 113680,
    "text": "Pearl Street Station Dynamo Room over in"
  },
  {
    "startMs": 115840,
    "text": "Manhattan."
  },
  {
    "startMs": 117520,
    "text": "This was the magic demo of its time."
  },
  {
    "startMs": 121560,
    "text": "The reality is it took a long time for"
  },
  {
    "startMs": 124800,
    "text": "electricity to be fully adopted."
  },
  {
    "startMs": 127040,
    "text": "Consider a Ford factory."
  },
  {
    "startMs": 129600,
    "text": "It's not enough to just have"
  },
  {
    "startMs": 131280,
    "text": "electricity. You have to rip out the"
  },
  {
    "startMs": 133520,
    "text": "existing motors and equipment. You have"
  },
  {
    "startMs": 135840,
    "text": "to bring in the new equipment. You have"
  },
  {
    "startMs": 137120,
    "text": "to go and train everybody to use that"
  },
  {
    "startMs": 139120,
    "text": "very same equipment."
  },
  {
    "startMs": 141120,
    "text": "Here's a picture of a Ford electrified"
  },
  {
    "startMs": 143440,
    "text": "moving assembly in 1924."
  },
  {
    "startMs": 146200,
    "text": "These things take time."
  },
  {
    "startMs": 148959,
    "text": "Diffusion of any technology takes a"
  },
  {
    "startMs": 151840,
    "text": "generation. And since everyone here in"
  },
  {
    "startMs": 154040,
    "text": "this room today is talking about AI, I"
  },
  {
    "startMs": 156040,
    "text": "would argue"
  },
  {
    "startMs": 157880,
    "text": "AI diffusion is perhaps the single most"
  },
  {
    "startMs": 160920,
    "text": "important problem for the next 20 years."
  },
  {
    "startMs": 164240,
    "text": "The models are going to keep getting"
  },
  {
    "startMs": 165480,
    "text": "better. The big question is how do we"
  },
  {
    "startMs": 168040,
    "text": "actually get these models to be in the"
  },
  {
    "startMs": 170280,
    "text": "real world, complete real tasks, uh and"
  },
  {
    "startMs": 172800,
    "text": "make people more efficient, happier, and"
  },
  {
    "startMs": 174640,
    "text": "provide better service."
  },
  {
    "startMs": 177400,
    "text": "So taking a quick step step back, who"
  },
  {
    "startMs": 179160,
    "text": "are we? Uh we are Long Lake. Over the"
  },
  {
    "startMs": 181840,
    "text": "last 2 years, we've raised over $3"
  },
  {
    "startMs": 184120,
    "text": "billion from Elad Gil, General Catalyst,"
  },
  {
    "startMs": 187000,
    "text": "and AlphaWave since our founding."
  },
  {
    "startMs": 190239,
    "text": "Here's the strange part. We we don't"
  },
  {
    "startMs": 192360,
    "text": "sell software. We actually go out and"
  },
  {
    "startMs": 194360,
    "text": "acquire and partner with real services"
  },
  {
    "startMs": 196720,
    "text": "businesses in the world. Uh we've"
  },
  {
    "startMs": 198720,
    "text": "acquired 35 businesses across HOA and"
  },
  {
    "startMs": 201440,
    "text": "property management, architecture, HR"
  },
  {
    "startMs": 203480,
    "text": "services, and a lot more."
  },
  {
    "startMs": 206600,
    "text": "To give you a little bit more flavor, we"
  },
  {
    "startMs": 208360,
    "text": "have roughly a 40% team right now split"
  },
  {
    "startMs": 210760,
    "text": "between technology, finance, and"
  },
  {
    "startMs": 212760,
    "text": "operations. More than half our team is"
  },
  {
    "startMs": 215040,
    "text": "part of the technology team focused on"
  },
  {
    "startMs": 217400,
    "text": "uh building products, data, and"
  },
  {
    "startMs": 219600,
    "text": "deploying the core products into the"
  },
  {
    "startMs": 221480,
    "text": "field. Uh we're in a collected group of"
  },
  {
    "startMs": 223880,
    "text": "folks, a bunch of ex-founders who've"
  },
  {
    "startMs": 226040,
    "text": "worked in the services before,"
  },
  {
    "startMs": 227920,
    "text": "ex-military, folks from Palantir, Ramp,"
  },
  {
    "startMs": 230760,
    "text": "Glean, uh and from the finance side,"
  },
  {
    "startMs": 233280,
    "text": "Blackstone, H.I.G., et cetera. We are we"
  },
  {
    "startMs": 236520,
    "text": "are not selling them to these companies"
  },
  {
    "startMs": 238960,
    "text": "above from the outside. We're actually"
  },
  {
    "startMs": 240960,
    "text": "deploying into these companies and"
  },
  {
    "startMs": 242480,
    "text": "figuring out how to get the technology"
  },
  {
    "startMs": 243920,
    "text": "to work."
  },
  {
    "startMs": 245400,
    "text": "And just to show you the scale we're"
  },
  {
    "startMs": 246720,
    "text": "playing at, we announced recently our"
  },
  {
    "startMs": 248560,
    "text": "$6.3 billion take private of American"
  },
  {
    "startMs": 251120,
    "text": "Express Global Business Travel, the"
  },
  {
    "startMs": 252880,
    "text": "world's largest corporate travel"
  },
  {
    "startMs": 254400,
    "text": "platform."
  },
  {
    "startMs": 256959,
    "text": "We own these businesses."
  },
  {
    "startMs": 259480,
    "text": "So, when the AI doesn't work, it's not"
  },
  {
    "startMs": 262120,
    "text": "their problem. We're not the vendor."
  },
  {
    "startMs": 264400,
    "text": "It's our problem."
  },
  {
    "startMs": 267080,
    "text": "Concretely, again, we are not the"
  },
  {
    "startMs": 269200,
    "text": "vendor. We are the operator owners. And"
  },
  {
    "startMs": 271520,
    "text": "we work very closely with our teams"
  },
  {
    "startMs": 273440,
    "text": "within the businesses to drive real"
  },
  {
    "startMs": 275240,
    "text": "outcomes."
  },
  {
    "startMs": 277600,
    "text": "Now, I want to step back and get to the"
  },
  {
    "startMs": 279440,
    "text": "concrete about the how. What are the"
  },
  {
    "startMs": 281400,
    "text": "lessons we've learned over the last 2"
  },
  {
    "startMs": 283120,
    "text": "and 1/2 years? And what we've learned"
  },
  {
    "startMs": 285600,
    "text": "from deploying AI into companies we've"
  },
  {
    "startMs": 287520,
    "text": "owned."
  },
  {
    "startMs": 291600,
    "text": "Three quick lessons. One, how we move"
  },
  {
    "startMs": 294280,
    "text": "agents from co-pilots to co-workers."
  },
  {
    "startMs": 297360,
    "text": "Two, how we leverage real-world data"
  },
  {
    "startMs": 300360,
    "text": "within these businesses."
  },
  {
    "startMs": 302120,
    "text": "Remember, we're seeing all of the work"
  },
  {
    "startMs": 304320,
    "text": "that's being done in these real services"
  },
  {
    "startMs": 306520,
    "text": "businesses. There's a lot of interesting"
  },
  {
    "startMs": 308760,
    "text": "problems and solutions embedded within"
  },
  {
    "startMs": 311000,
    "text": "that."
  },
  {
    "startMs": 311880,
    "text": "And then finally, perhaps the most"
  },
  {
    "startMs": 313760,
    "text": "interesting and exciting is how do you"
  },
  {
    "startMs": 315120,
    "text": "actually get all of this technology to"
  },
  {
    "startMs": 316960,
    "text": "compound over time by learning loops in"
  },
  {
    "startMs": 319280,
    "text": "the enterprise. We'll get to that at the"
  },
  {
    "startMs": 321240,
    "text": "end over here."
  },
  {
    "startMs": 323680,
    "text": "So, starting off from co-pilots to"
  },
  {
    "startMs": 325240,
    "text": "co-workers."
  },
  {
    "startMs": 327000,
    "text": "There's a spectrum of how much autonomy"
  },
  {
    "startMs": 329320,
    "text": "you can give an agent."
  },
  {
    "startMs": 331000,
    "text": "On the left here, you see a co-pilot."
  },
  {
    "startMs": 332560,
    "text": "This is, you know, your simple rag"
  },
  {
    "startMs": 334120,
    "text": "chatbot from 2 years ago. It's very"
  },
  {
    "startMs": 336400,
    "text": "quick. You can ask a question. Maybe"
  },
  {
    "startMs": 338120,
    "text": "it's integrated with some systems. It"
  },
  {
    "startMs": 339760,
    "text": "can give you information back very, very"
  },
  {
    "startMs": 341440,
    "text": "quickly."
  },
  {
    "startMs": 344080,
    "text": "The second step is a synchronous agent."
  },
  {
    "startMs": 345960,
    "text": "Consider something like Claude code,"
  },
  {
    "startMs": 347920,
    "text": "Codex, Claude co-work. It's real-time."
  },
  {
    "startMs": 350680,
    "text": "There's this two-way interaction. It's a"
  },
  {
    "startMs": 352360,
    "text": "bit more sophisticated than a co-pilot."
  },
  {
    "startMs": 354160,
    "text": "You can go let it run off for 1 to 5"
  },
  {
    "startMs": 356040,
    "text": "minutes. Uh it'll call tools, maybe use"
  },
  {
    "startMs": 358360,
    "text": "its skills. Uh it's still synchronous."
  },
  {
    "startMs": 360320,
    "text": "You still need to step in and ask a"
  },
  {
    "startMs": 361520,
    "text": "query. So, the next obvious rung of the"
  },
  {
    "startMs": 363920,
    "text": "ladder is the asynchronous agent."
  },
  {
    "startMs": 366600,
    "text": "You can come in here, still ask a query."
  },
  {
    "startMs": 368960,
    "text": "The agent will go off into the"
  },
  {
    "startMs": 370080,
    "text": "background, do some work, and then come"
  },
  {
    "startMs": 371920,
    "text": "back. Uh and what's really interesting"
  },
  {
    "startMs": 374200,
    "text": "about asynchronous agents is that the"
  },
  {
    "startMs": 375880,
    "text": "user does not have to be the one that"
  },
  {
    "startMs": 377720,
    "text": "triggers them."
  },
  {
    "startMs": 379040,
    "text": "You can have external triggers as well."
  },
  {
    "startMs": 381240,
    "text": "Maybe someone completes a certain task"
  },
  {
    "startMs": 383360,
    "text": "and there is an async job queue uh that"
  },
  {
    "startMs": 385800,
    "text": "allows the async agent to pull off from"
  },
  {
    "startMs": 387560,
    "text": "and proactively offer advice to the end"
  },
  {
    "startMs": 389960,
    "text": "user."
  },
  {
    "startMs": 391919,
    "text": "Then, I'd argue the next step is a"
  },
  {
    "startMs": 393800,
    "text": "long-running agent."
  },
  {
    "startMs": 395520,
    "text": "How do you get these agents to work for"
  },
  {
    "startMs": 397480,
    "text": "hours, days, weeks, months, etc.? I"
  },
  {
    "startMs": 400840,
    "text": "think this is currently a very core"
  },
  {
    "startMs": 402600,
    "text": "problem that a lot of the labs are"
  },
  {
    "startMs": 404320,
    "text": "focused on, as are we."
  },
  {
    "startMs": 408000,
    "text": "And then finally, at the end, the holy"
  },
  {
    "startMs": 410160,
    "text": "grail, an AI co-worker."
  },
  {
    "startMs": 413080,
    "text": "This is where most people start off."
  },
  {
    "startMs": 415200,
    "text": "You want a proactive partner that gets"
  },
  {
    "startMs": 417240,
    "text": "work done just alongside you."
  },
  {
    "startMs": 420280,
    "text": "This is what everyone wants to sell you,"
  },
  {
    "startMs": 422200,
    "text": "but what we've learned from owning the"
  },
  {
    "startMs": 423919,
    "text": "outcomes in this business is"
  },
  {
    "startMs": 427280,
    "text": "you have to earn the right to do more."
  },
  {
    "startMs": 429720,
    "text": "It's it's not enough to jump to the"
  },
  {
    "startMs": 431480,
    "text": "co-worker immediately,"
  },
  {
    "startMs": 433120,
    "text": "right? For for a bunch of reasons. One,"
  },
  {
    "startMs": 435480,
    "text": "for certain tasks, the models might not"
  },
  {
    "startMs": 437280,
    "text": "quite be there yet. And two, you"
  },
  {
    "startMs": 439320,
    "text": "actually have to work with these"
  },
  {
    "startMs": 440840,
    "text": "companies in the field, interact and"
  },
  {
    "startMs": 443560,
    "text": "iterate very, very closely, so that they"
  },
  {
    "startMs": 446200,
    "text": "understand that this is the beginning of"
  },
  {
    "startMs": 448200,
    "text": "AI, and you can work up the rungs over"
  },
  {
    "startMs": 450040,
    "text": "time."
  },
  {
    "startMs": 453240,
    "text": "I think a really unique lens to look at"
  },
  {
    "startMs": 455000,
    "text": "this problem through is the that of the"
  },
  {
    "startMs": 457400,
    "text": "jagged frontier. We all know that agents"
  },
  {
    "startMs": 460200,
    "text": "are incredibly good at writing code. So,"
  },
  {
    "startMs": 462760,
    "text": "what does the, for example, synchronous"
  },
  {
    "startMs": 464680,
    "text": "agent for code generation look like?"
  },
  {
    "startMs": 467000,
    "text": "This is super simple. This is just your"
  },
  {
    "startMs": 468680,
    "text": "coding agent, maybe it's Codex, Cloud"
  },
  {
    "startMs": 470520,
    "text": "Code, just running on your desktop. It"
  },
  {
    "startMs": 472720,
    "text": "has access to a file system. You"
  },
  {
    "startMs": 474320,
    "text": "collaborate within real time. You get"
  },
  {
    "startMs": 476080,
    "text": "instant feedback and you iterate."
  },
  {
    "startMs": 479440,
    "text": "The next step is, you know, if you look"
  },
  {
    "startMs": 481120,
    "text": "at code code generation, what is the"
  },
  {
    "startMs": 482520,
    "text": "async agent? This is also fairly"
  },
  {
    "startMs": 484800,
    "text": "straightforward and largely solved. You"
  },
  {
    "startMs": 486320,
    "text": "take the exact same coding agent, you"
  },
  {
    "startMs": 488240,
    "text": "wrap it in a sandbox, and you just let"
  },
  {
    "startMs": 490320,
    "text": "it go run. It can build, it can test,"
  },
  {
    "startMs": 492560,
    "text": "and once it's done with its work, it can"
  },
  {
    "startMs": 494200,
    "text": "provide the code in the form of a PR."
  },
  {
    "startMs": 496880,
    "text": "One thing that's really unique about"
  },
  {
    "startMs": 498160,
    "text": "engineers is folks are incredibly good"
  },
  {
    "startMs": 501200,
    "text": "at already paralyzing their work."
  },
  {
    "startMs": 503760,
    "text": "It's very commonplace to launch 10 jobs"
  },
  {
    "startMs": 506560,
    "text": "and be comfortable with the fact that"
  },
  {
    "startMs": 508080,
    "text": "job seven might finish before job three."
  },
  {
    "startMs": 510880,
    "text": "So, engineers are incredibly good at"
  },
  {
    "startMs": 512800,
    "text": "using these async agents."
  },
  {
    "startMs": 515719,
    "text": "Now, when we come to services, the"
  },
  {
    "startMs": 517039,
    "text": "equivalent of a synchronous agent, what"
  },
  {
    "startMs": 518719,
    "text": "we talked about a little bit earlier,"
  },
  {
    "startMs": 520280,
    "text": "it's a co-working agent. It's an agent"
  },
  {
    "startMs": 522159,
    "text": "that has deep context about your"
  },
  {
    "startMs": 523560,
    "text": "enterprise. It interacts potentially"
  },
  {
    "startMs": 525480,
    "text": "with MCPs, custom tools, custom"
  },
  {
    "startMs": 527200,
    "text": "integrations, uh and you can chat with"
  },
  {
    "startMs": 529520,
    "text": "it synchronously just like any of these"
  },
  {
    "startMs": 531080,
    "text": "other products."
  },
  {
    "startMs": 533160,
    "text": "I think this is a frontier here in the"
  },
  {
    "startMs": 534720,
    "text": "bottom right."
  },
  {
    "startMs": 536160,
    "text": "What does it mean to build an"
  },
  {
    "startMs": 538160,
    "text": "asynchronous agent for the services?"
  },
  {
    "startMs": 540520,
    "text": "What does it mean to paralyze work in"
  },
  {
    "startMs": 543000,
    "text": "industries where work is traditionally"
  },
  {
    "startMs": 545200,
    "text": "done in a very, very serial manner?"
  },
  {
    "startMs": 548120,
    "text": "This is where we spend a lot of time and"
  },
  {
    "startMs": 550000,
    "text": "this is what I wake up every morning"
  },
  {
    "startMs": 551320,
    "text": "really excited thinking about, you know,"
  },
  {
    "startMs": 552800,
    "text": "we've we've figured out what the async"
  },
  {
    "startMs": 555240,
    "text": "and forking mechanism for code is. You"
  },
  {
    "startMs": 557640,
    "text": "just spin up a bunch of sandboxes and do"
  },
  {
    "startMs": 559200,
    "text": "work. What does that look like for the"
  },
  {
    "startMs": 561520,
    "text": "rest of the world?"
  },
  {
    "startMs": 564839,
    "text": "So, here's a couple questions we think"
  },
  {
    "startMs": 566080,
    "text": "about pretty seriously. One, you know,"
  },
  {
    "startMs": 568720,
    "text": "the models are trained on code, they"
  },
  {
    "startMs": 570000,
    "text": "want to write code, they're incredibly"
  },
  {
    "startMs": 571280,
    "text": "good at writing code. How do we leverage"
  },
  {
    "startMs": 573120,
    "text": "these coding agents for actual knowledge"
  },
  {
    "startMs": 574680,
    "text": "work? You You rather than wait for the"
  },
  {
    "startMs": 576360,
    "text": "models to catch up on doing services"
  },
  {
    "startMs": 578040,
    "text": "knowledge work, what if we just use that"
  },
  {
    "startMs": 580560,
    "text": "code knowledge and represent knowledge"
  },
  {
    "startMs": 582560,
    "text": "work as code?"
  },
  {
    "startMs": 584560,
    "text": "Two, as I mentioned, engineers are used"
  },
  {
    "startMs": 586480,
    "text": "to paralyzing work. How do you paralyze"
  },
  {
    "startMs": 588280,
    "text": "work that's traditionally serial? You"
  },
  {
    "startMs": 589440,
    "text": "know, people clean out their inbox one"
  },
  {
    "startMs": 591160,
    "text": "email by one email, not 10 emails at"
  },
  {
    "startMs": 592920,
    "text": "once."
  },
  {
    "startMs": 594280,
    "text": "And finally, how do you move up the"
  },
  {
    "startMs": 595800,
    "text": "ladder here both in terms of product and"
  },
  {
    "startMs": 597760,
    "text": "user enablement?"
  },
  {
    "startMs": 599480,
    "text": "What are the right form factors? And I'd"
  },
  {
    "startMs": 601480,
    "text": "argue this varies dramatically from"
  },
  {
    "startMs": 604280,
    "text": "industry to industry. Just because you"
  },
  {
    "startMs": 606160,
    "text": "have one way of launching an async agent"
  },
  {
    "startMs": 607600,
    "text": "for code, doesn't mean that same way is"
  },
  {
    "startMs": 609200,
    "text": "going to work for architecture or"
  },
  {
    "startMs": 610720,
    "text": "property management."
  },
  {
    "startMs": 613440,
    "text": "The second point I want to cover today"
  },
  {
    "startMs": 615040,
    "text": "is leveraging real-world data."
  },
  {
    "startMs": 617560,
    "text": "We all know this. Frontier models have"
  },
  {
    "startMs": 619440,
    "text": "learned from everything humanity has"
  },
  {
    "startMs": 621040,
    "text": "written down,"
  },
  {
    "startMs": 622200,
    "text": "but the most valuable tasks are not on"
  },
  {
    "startMs": 624720,
    "text": "the internet."
  },
  {
    "startMs": 626040,
    "text": "How do you actually close the books when"
  },
  {
    "startMs": 627960,
    "text": "you're missing receipts?"
  },
  {
    "startMs": 629462,
    "text": ">> [snorts]"
  },
  {
    "startMs": 629560,
    "text": ">> How do you scope a building for"
  },
  {
    "startMs": 631200,
    "text": "construction in a blueprint, potentially"
  },
  {
    "startMs": 633160,
    "text": "collaboratively?"
  },
  {
    "startMs": 635040,
    "text": "How do you coordinate vendors for fixing"
  },
  {
    "startMs": 636960,
    "text": "a broken roof?"
  },
  {
    "startMs": 639120,
    "text": "All of this knowledge lives in people's"
  },
  {
    "startMs": 640960,
    "text": "heads, in 20-year-old software, uh in"
  },
  {
    "startMs": 643560,
    "text": "the way that one senior person on one of"
  },
  {
    "startMs": 645640,
    "text": "these teams just knows how to do it. How"
  },
  {
    "startMs": 648000,
    "text": "do you make this information explicit"
  },
  {
    "startMs": 649839,
    "text": "and create tasks that you can actually"
  },
  {
    "startMs": 651480,
    "text": "learn from?"
  },
  {
    "startMs": 652640,
    "text": "So, we've constructed a little bit of a"
  },
  {
    "startMs": 653839,
    "text": "flywheel. We get our agents to"
  },
  {
    "startMs": 656000,
    "text": "collaborate with our employees to do"
  },
  {
    "startMs": 658120,
    "text": "real work. And this allows us to"
  },
  {
    "startMs": 660000,
    "text": "generate rich traces of data and"
  },
  {
    "startMs": 661920,
    "text": "information. Tool calls, the hiccups,"
  },
  {
    "startMs": 664280,
    "text": "the papercuts, everything that goes"
  },
  {
    "startMs": 666240,
    "text": "wrong with doing real work."
  },
  {
    "startMs": 668560,
    "text": "This in turn allows us to build"
  },
  {
    "startMs": 670360,
    "text": "real-world evals."
  },
  {
    "startMs": 672320,
    "text": "There is a ground truth here. In the"
  },
  {
    "startMs": 674160,
    "text": "case of the roofing example, the"
  },
  {
    "startMs": 676320,
    "text": "question is, did the roof get repaired?"
  },
  {
    "startMs": 679560,
    "text": "Did the books get closed?"
  },
  {
    "startMs": 681680,
    "text": "And this allows us to hill climb and"
  },
  {
    "startMs": 683160,
    "text": "build better agents, which leads to more"
  },
  {
    "startMs": 685600,
    "text": "and more impact. And what's really"
  },
  {
    "startMs": 687280,
    "text": "exciting is it ratchets up. Every week"
  },
  {
    "startMs": 690320,
    "text": "our hill climbing benchmarks"
  },
  {
    "startMs": 692600,
    "text": "become a regression test. So, our agents"
  },
  {
    "startMs": 694760,
    "text": "get better and better over time."
  },
  {
    "startMs": 697680,
    "text": "Just to drive a little bit deeper here"
  },
  {
    "startMs": 699440,
    "text": "on the traces, there's three upshots of"
  },
  {
    "startMs": 702280,
    "text": "being able to collect these rich traces."
  },
  {
    "startMs": 704480,
    "text": "One, we get to generate amazing evals"
  },
  {
    "startMs": 706600,
    "text": "that are built and scored automatically."
  },
  {
    "startMs": 709120,
    "text": "Uh and we're able to gather both"
  },
  {
    "startMs": 711280,
    "text": "implicit and explicit feedback. Explicit"
  },
  {
    "startMs": 713360,
    "text": "feedback in the sense of thumbs ups and"
  },
  {
    "startMs": 714960,
    "text": "thumbs down, maybe people provide a note"
  },
  {
    "startMs": 717080,
    "text": "telling us whether this response was"
  },
  {
    "startMs": 718440,
    "text": "good or not. Uh and also implicit"
  },
  {
    "startMs": 720600,
    "text": "feedback. Right? Again, we have the"
  },
  {
    "startMs": 722240,
    "text": "ground truth. Maybe there's some data"
  },
  {
    "startMs": 723800,
    "text": "that the AI generated and there's a real"
  },
  {
    "startMs": 725880,
    "text": "diff between the data that the AI"
  },
  {
    "startMs": 727680,
    "text": "generated and what was ultimately"
  },
  {
    "startMs": 729400,
    "text": "submitted. That's rich information that"
  },
  {
    "startMs": 732160,
    "text": "almost no one else has."
  },
  {
    "startMs": 734520,
    "text": "Two, we've started post training models"
  },
  {
    "startMs": 736360,
    "text": "internally on"
  },
  {
    "startMs": 738360,
    "text": "all of the data that these businesses"
  },
  {
    "startMs": 739880,
    "text": "operate on and produce, generally"
  },
  {
    "startMs": 742160,
    "text": "speaking."
  },
  {
    "startMs": 743240,
    "text": "This is all data that is completely out"
  },
  {
    "startMs": 745800,
    "text": "of distribution for most frontier labs."
  },
  {
    "startMs": 747640,
    "text": "Think of the task I showed at the"
  },
  {
    "startMs": 748880,
    "text": "beginning. A lot of the models A lot of"
  },
  {
    "startMs": 751160,
    "text": "the frontier models today just can't do"
  },
  {
    "startMs": 753000,
    "text": "these tasks yet and we're trying to post"
  },
  {
    "startMs": 755520,
    "text": "train our own models internally to be"
  },
  {
    "startMs": 757320,
    "text": "able to do that on the rich source of"
  },
  {
    "startMs": 758800,
    "text": "data that we own."
  },
  {
    "startMs": 760280,
    "text": "And then finally, the actual agents"
  },
  {
    "startMs": 762480,
    "text": "themselves."
  },
  {
    "startMs": 763839,
    "text": "The real world is incredibly hairy and"
  },
  {
    "startMs": 765920,
    "text": "messy and you want customization per"
  },
  {
    "startMs": 767920,
    "text": "company. Every company does things very"
  },
  {
    "startMs": 770240,
    "text": "differently. Customization per user. The"
  },
  {
    "startMs": 772720,
    "text": "way each user does their work is very"
  },
  {
    "startMs": 774960,
    "text": "unique. And customization per client."
  },
  {
    "startMs": 777480,
    "text": "The way you work with every client is"
  },
  {
    "startMs": 779400,
    "text": "different. It's a services business and"
  },
  {
    "startMs": 781040,
    "text": "you want to uphold those standards."
  },
  {
    "startMs": 784440,
    "text": "I love this picture"
  },
  {
    "startMs": 786200,
    "text": "because it's the whole thing in a single"
  },
  {
    "startMs": 788000,
    "text": "image. Um the the way we usually talk"
  },
  {
    "startMs": 790800,
    "text": "about LLM tasks is the top panel. Right?"
  },
  {
    "startMs": 793600,
    "text": "You just It's It's a slope. You got a"
  },
  {
    "startMs": 795960,
    "text": "bike. And but there's clear sight to"
  },
  {
    "startMs": 798040,
    "text": "success."
  },
  {
    "startMs": 799520,
    "text": "The reality is most work is not like"
  },
  {
    "startMs": 802200,
    "text": "that. And And you and I both know that."
  },
  {
    "startMs": 804839,
    "text": "Uh there are hills and ravines. Uh"
  },
  {
    "startMs": 807360,
    "text": "there's death by a thousand paper cuts."
  },
  {
    "startMs": 809760,
    "text": "But But that's what real work looks"
  },
  {
    "startMs": 811680,
    "text": "like. That's the entire job. The"
  },
  {
    "startMs": 813920,
    "text": "exceptions are the job."
  },
  {
    "startMs": 817720,
    "text": "That's That's the demo."
  },
  {
    "startMs": 819640,
    "text": "That's the actual job."
  },
  {
    "startMs": 823040,
    "text": "Now, on to the final thing I want to"
  },
  {
    "startMs": 824640,
    "text": "chat with you guys today is learning"
  },
  {
    "startMs": 827080,
    "text": "loops within the enterprise."
  },
  {
    "startMs": 829080,
    "text": "I'd argue there's two hot trends"
  },
  {
    "startMs": 831360,
    "text": "everyone's talking about in 2026. One,"
  },
  {
    "startMs": 834440,
    "text": "it's continual learning. How do you make"
  },
  {
    "startMs": 836560,
    "text": "an agent better over time with feedback?"
  },
  {
    "startMs": 838680,
    "text": "I think there are plenty of sessions uh"
  },
  {
    "startMs": 840680,
    "text": "this week on how you can use continual"
  },
  {
    "startMs": 842640,
    "text": "learning, whether it's in the prompt or"
  },
  {
    "startMs": 844440,
    "text": "in the weights."
  },
  {
    "startMs": 846120,
    "text": "And two, enablement. How do you get in"
  },
  {
    "startMs": 848560,
    "text": "these enterprises and actually get them"
  },
  {
    "startMs": 850160,
    "text": "to adopt and use AI?"
  },
  {
    "startMs": 852920,
    "text": "Traditionally speaking,"
  },
  {
    "startMs": 855000,
    "text": "these two initiatives are owned by two"
  },
  {
    "startMs": 856400,
    "text": "separate teams. Right? The continual"
  },
  {
    "startMs": 858640,
    "text": "learning is owned by your research team,"
  },
  {
    "startMs": 860160,
    "text": "your platform engineering team."
  },
  {
    "startMs": 861640,
    "text": "Enablement's owned by growth or"
  },
  {
    "startMs": 863200,
    "text": "deployment or customer experience. Uh"
  },
  {
    "startMs": 865360,
    "text": "usually pretty siloed, not much"
  },
  {
    "startMs": 866800,
    "text": "interaction between the two."
  },
  {
    "startMs": 869720,
    "text": "We think these are part of the exact"
  },
  {
    "startMs": 871120,
    "text": "same loop."
  },
  {
    "startMs": 872200,
    "text": "The agent only improves if people"
  },
  {
    "startMs": 874800,
    "text": "actually use it."
  },
  {
    "startMs": 877000,
    "text": "And people only use the agent if it's"
  },
  {
    "startMs": 879120,
    "text": "worth adopting."
  },
  {
    "startMs": 881160,
    "text": "So, here's a little graphic of a"
  },
  {
    "startMs": 882560,
    "text": "snowball. More usage drives continual"
  },
  {
    "startMs": 885040,
    "text": "learning, which drives a better agent,"
  },
  {
    "startMs": 886960,
    "text": "which drives more usage again."
  },
  {
    "startMs": 889160,
    "text": "All this to say, there's still a really"
  },
  {
    "startMs": 891400,
    "text": "big elephant in the room."
  },
  {
    "startMs": 893760,
    "text": "How do you get the initial usage?"
  },
  {
    "startMs": 895839,
    "text": "I think a lot of people, you know, will"
  },
  {
    "startMs": 897960,
    "text": "use Claude Code or or give it to their"
  },
  {
    "startMs": 900080,
    "text": "whole enterprise, expect folks to just"
  },
  {
    "startMs": 902280,
    "text": "start using it."
  },
  {
    "startMs": 904280,
    "text": "Everyone assumes the usage just shows"
  },
  {
    "startMs": 907000,
    "text": "up."
  },
  {
    "startMs": 908360,
    "text": "But as we all know, that's simply not"
  },
  {
    "startMs": 910600,
    "text": "the case. It never does. Right? Getting"
  },
  {
    "startMs": 912960,
    "text": "a 100-year-old firm to change its"
  },
  {
    "startMs": 915080,
    "text": "processes is hard."
  },
  {
    "startMs": 917680,
    "text": "You could have the best AI coworker on"
  },
  {
    "startMs": 920080,
    "text": "the internet or on Earth. And if the"
  },
  {
    "startMs": 922600,
    "text": "people if the person who's closed the"
  },
  {
    "startMs": 924120,
    "text": "books for the last 20 years continues to"
  },
  {
    "startMs": 926400,
    "text": "do things the same way,"
  },
  {
    "startMs": 928440,
    "text": "nothing changes. Nothing happens."
  },
  {
    "startMs": 932240,
    "text": "So, what can you actually do about it?"
  },
  {
    "startMs": 933880,
    "text": "What you know, this this seems like"
  },
  {
    "startMs": 935840,
    "text": "incredibly hard. What what's the upshot?"
  },
  {
    "startMs": 938240,
    "text": "How do you actually get this stuff to"
  },
  {
    "startMs": 939320,
    "text": "work? Well, I think a lot about Jensen"
  },
  {
    "startMs": 941840,
    "text": "and how he dominated the market in his"
  },
  {
    "startMs": 944040,
    "text": "words with extreme hardware software"
  },
  {
    "startMs": 946720,
    "text": "co-design. Designing the chips and the"
  },
  {
    "startMs": 949440,
    "text": "software together as one system."
  },
  {
    "startMs": 952960,
    "text": "We look at this through the lens of"
  },
  {
    "startMs": 954600,
    "text": "extreme software service co-design. How"
  },
  {
    "startMs": 957360,
    "text": "do you co-design our products with the"
  },
  {
    "startMs": 959800,
    "text": "people and the processes at our"
  },
  {
    "startMs": 961840,
    "text": "businesses? And I'd argue this is only"
  },
  {
    "startMs": 964080,
    "text": "possible from being within under the"
  },
  {
    "startMs": 966200,
    "text": "same roof."
  },
  {
    "startMs": 968480,
    "text": "We need to meet the people within these"
  },
  {
    "startMs": 970680,
    "text": "companies both metaphorically, for"
  },
  {
    "startMs": 973280,
    "text": "example, bringing products to their"
  },
  {
    "startMs": 975120,
    "text": "systems so that the energy required for"
  },
  {
    "startMs": 977480,
    "text": "enablement is kept low, and also"
  },
  {
    "startMs": 979440,
    "text": "physically. Get on a plane, show up, say"
  },
  {
    "startMs": 982120,
    "text": "hi, learn what people actually do."
  },
  {
    "startMs": 985600,
    "text": "You know, maybe you build a product"
  },
  {
    "startMs": 987200,
    "text": "that's natively embedded into Excel or"
  },
  {
    "startMs": 989720,
    "text": "into their ERP system, maybe their 3D"
  },
  {
    "startMs": 992160,
    "text": "design software, or or maybe even their"
  },
  {
    "startMs": 994520,
    "text": "Microsoft products like Outlook, Gmail,"
  },
  {
    "startMs": 996400,
    "text": "etc."
  },
  {
    "startMs": 998040,
    "text": "Or you show up in person. You do a lunch"
  },
  {
    "startMs": 999720,
    "text": "and learn with a bunch of folks at one"
  },
  {
    "startMs": 1000800,
    "text": "of the companies. You go to their"
  },
  {
    "startMs": 1002200,
    "text": "conferences and you create cotton candy"
  },
  {
    "startMs": 1004240,
    "text": "and run a stand for them. You go"
  },
  {
    "startMs": 1005920,
    "text": "mountain biking and ask them about all"
  },
  {
    "startMs": 1007360,
    "text": "the difficulties that they have with"
  },
  {
    "startMs": 1008720,
    "text": "their actual day-to-day jobs."
  },
  {
    "startMs": 1010920,
    "text": "Or you show up in person one-on-one or"
  },
  {
    "startMs": 1013400,
    "text": "sometimes even two-on-one in this case"
  },
  {
    "startMs": 1015360,
    "text": "and just show them how to use the tools"
  },
  {
    "startMs": 1017720,
    "text": "and learn from the feedback because this"
  },
  {
    "startMs": 1020000,
    "text": "is what the rest of the world really"
  },
  {
    "startMs": 1021640,
    "text": "looks like. It's not like the folks in"
  },
  {
    "startMs": 1022960,
    "text": "this room or in San Francisco. It's a"
  },
  {
    "startMs": 1024959,
    "text": "lot more like this. You cannot co-design"
  },
  {
    "startMs": 1027400,
    "text": "software with the services business over"
  },
  {
    "startMs": 1028839,
    "text": "Zoom"
  },
  {
    "startMs": 1029800,
    "text": "or over a support ticket. You you have"
  },
  {
    "startMs": 1032000,
    "text": "to be there. You have to be in person."
  },
  {
    "startMs": 1034040,
    "text": "And I'd argue this is the part that"
  },
  {
    "startMs": 1036000,
    "text": "actually makes it work. In order to get"
  },
  {
    "startMs": 1037839,
    "text": "AI diffusion to work, you have to touch"
  },
  {
    "startMs": 1040160,
    "text": "some grass."
  },
  {
    "startMs": 1042160,
    "text": "Thank you so much. I'll be around for"
  },
  {
    "startMs": 1043880,
    "text": "the rest of day if there's anything I"
  },
  {
    "startMs": 1044920,
    "text": "can help with. My email is up there."
  },
  {
    "startMs": 1047680,
    "text": "And yeah, thank you."
  },
  {
    "startMs": 1063732,
    "text": ">> [music]"
  }
]