[
  {
    "startMs": 12639,
    "text": "Hey everyone, thanks for coming. I'm"
  },
  {
    "startMs": 15519,
    "text": "Andrew. I'm the chief of software at"
  },
  {
    "startMs": 17760,
    "text": "Verscell"
  },
  {
    "startMs": 19439,
    "text": "and I'm here to talk to you about how we"
  },
  {
    "startMs": 21680,
    "text": "solved agent building at Verscell. I'm"
  },
  {
    "startMs": 24720,
    "text": "the chief of software. So I work on a"
  },
  {
    "startMs": 26320,
    "text": "mix of internal engineering, external"
  },
  {
    "startMs": 28320,
    "text": "experimentation, and generally being at"
  },
  {
    "startMs": 30400,
    "text": "the frontier and building new libraries,"
  },
  {
    "startMs": 32320,
    "text": "frameworks, and technologies. For those"
  },
  {
    "startMs": 34880,
    "text": "of you that don't know Verscell,"
  },
  {
    "startMs": 37200,
    "text": "Verscell builds a gentic infrastructure"
  },
  {
    "startMs": 39520,
    "text": "so people can build what's next. We get"
  },
  {
    "startMs": 42160,
    "text": "started in the web world helping people"
  },
  {
    "startMs": 43840,
    "text": "ship websites and web apps without"
  },
  {
    "startMs": 46160,
    "text": "having to worry about the infrastructure"
  },
  {
    "startMs": 47680,
    "text": "that doesn't make their app any better."
  },
  {
    "startMs": 49200,
    "text": "It can scale to a million and scale down"
  },
  {
    "startMs": 50800,
    "text": "to zero effortlessly. But we're seeing a"
  },
  {
    "startMs": 53680,
    "text": "change in what people want to build. You"
  },
  {
    "startMs": 55760,
    "text": "know, people started by building pages,"
  },
  {
    "startMs": 57360,
    "text": "but now we see them want to build"
  },
  {
    "startMs": 58800,
    "text": "agents. And we've been embarking on a"
  },
  {
    "startMs": 60879,
    "text": "similar journey to make it easy for"
  },
  {
    "startMs": 62719,
    "text": "people to build agents and agentic"
  },
  {
    "startMs": 65360,
    "text": "applications easier. We built this thing"
  },
  {
    "startMs": 68240,
    "text": "called the AIDK. So instead of needing"
  },
  {
    "startMs": 70080,
    "text": "to switch out 300 400 lines of provider"
  },
  {
    "startMs": 73360,
    "text": "specific code, you're just going to"
  },
  {
    "startMs": 74799,
    "text": "switch out one line of code and we have"
  },
  {
    "startMs": 76560,
    "text": "the same model interface underlying for"
  },
  {
    "startMs": 78799,
    "text": "all these different providers."
  },
  {
    "startMs": 81119,
    "text": "We built a lot of other tools to make it"
  },
  {
    "startMs": 82720,
    "text": "easier to have model fallbacks, secure"
  },
  {
    "startMs": 85119,
    "text": "code execution, better pricing when it's"
  },
  {
    "startMs": 87360,
    "text": "inactive and waiting for responses, as"
  },
  {
    "startMs": 89520,
    "text": "well as for durability and resumability."
  },
  {
    "startMs": 92880,
    "text": "And I'm here to talk to you about how I"
  },
  {
    "startMs": 95360,
    "text": "went on this crazy experiment roughly a"
  },
  {
    "startMs": 97119,
    "text": "year ago that led to a aentic explosion"
  },
  {
    "startMs": 101200,
    "text": "at Verscell and led to a really cool"
  },
  {
    "startMs": 103119,
    "text": "thing that we built recently"
  },
  {
    "startMs": 106159,
    "text": "about uh this is 1980. Uh, Bill Gates"
  },
  {
    "startMs": 109680,
    "text": "before my time had this quote saying he"
  },
  {
    "startMs": 112079,
    "text": "imagined there would be a computer on"
  },
  {
    "startMs": 113920,
    "text": "every desk and in every home. You know,"
  },
  {
    "startMs": 115920,
    "text": "that was probably pretty contrarian then"
  },
  {
    "startMs": 117520,
    "text": "and today it seems like very normal to"
  },
  {
    "startMs": 120479,
    "text": "have that happen. And me and the CTO had"
  },
  {
    "startMs": 123840,
    "text": "this thought, you know, instead of a"
  },
  {
    "startMs": 126079,
    "text": "computer on every desk, could we"
  },
  {
    "startMs": 128080,
    "text": "potentially have an agent on every desk?"
  },
  {
    "startMs": 130879,
    "text": "You know, today we only really use"
  },
  {
    "startMs": 133200,
    "text": "agents for coding and technical"
  },
  {
    "startMs": 135520,
    "text": "workloads, but we're starting to see"
  },
  {
    "startMs": 137120,
    "text": "expansion into things like design,"
  },
  {
    "startMs": 139120,
    "text": "product management, and other verticals."
  },
  {
    "startMs": 142400,
    "text": "And this was maybe about a year ago, so"
  },
  {
    "startMs": 144000,
    "text": "I would say I'm pretty early to this,"
  },
  {
    "startMs": 145599,
    "text": "but that was when it was like sonnet 4"
  },
  {
    "startMs": 147520,
    "text": "and things weren't as sophisticated as"
  },
  {
    "startMs": 149120,
    "text": "they were today. And I tried to actually"
  },
  {
    "startMs": 152560,
    "text": "explore this out, see what we could do"
  },
  {
    "startMs": 154080,
    "text": "about it. I went around to various job"
  },
  {
    "startMs": 156239,
    "text": "functions at Verscell, marketing, sales,"
  },
  {
    "startMs": 159120,
    "text": "finance, legal, and I asked them, what"
  },
  {
    "startMs": 161680,
    "text": "do you hate most about your job? And the"
  },
  {
    "startMs": 165120,
    "text": "most compelling use case I heard was"
  },
  {
    "startMs": 167280,
    "text": "that the data team, they were growing."
  },
  {
    "startMs": 170080,
    "text": "They were a very lean team, but Versel"
  },
  {
    "startMs": 172400,
    "text": "was growing faster. You know, they had"
  },
  {
    "startMs": 174480,
    "text": "so much more data from customers,"
  },
  {
    "startMs": 176800,
    "text": "analytics, metrics, sales. They just had"
  },
  {
    "startMs": 179840,
    "text": "to keep on aggregating and keep on"
  },
  {
    "startMs": 181599,
    "text": "making available for themselves to use."
  },
  {
    "startMs": 185200,
    "text": "And at this time, if you think about"
  },
  {
    "startMs": 187040,
    "text": "what the data science people ever have"
  },
  {
    "startMs": 188800,
    "text": "to do, whenever someone from marketing"
  },
  {
    "startMs": 190879,
    "text": "or sales has a question about a customer"
  },
  {
    "startMs": 192959,
    "text": "or product, the data science team has to"
  },
  {
    "startMs": 195360,
    "text": "drop everything they're doing, write the"
  },
  {
    "startMs": 197280,
    "text": "query, process it, do an analysis, and"
  },
  {
    "startMs": 199519,
    "text": "come back with some recommendation on"
  },
  {
    "startMs": 201040,
    "text": "what to do. And this was really killer"
  },
  {
    "startMs": 204080,
    "text": "to productivity. You know, the data team"
  },
  {
    "startMs": 206000,
    "text": "did not want to drop everything and just"
  },
  {
    "startMs": 208000,
    "text": "write queries all day. And so I worked"
  },
  {
    "startMs": 209920,
    "text": "with our VP of data to try to build a"
  },
  {
    "startMs": 211920,
    "text": "better way for them to operate this way."
  },
  {
    "startMs": 216000,
    "text": "And so if you think about the very first"
  },
  {
    "startMs": 217920,
    "text": "thing you would ever do if you want to"
  },
  {
    "startMs": 219519,
    "text": "try to use AI to solve a problem, you"
  },
  {
    "startMs": 221519,
    "text": "may just build like a huge mega prompt."
  },
  {
    "startMs": 223840,
    "text": "You know, you just have a question, you"
  },
  {
    "startMs": 226159,
    "text": "pass into an LM, you have it respond,"
  },
  {
    "startMs": 228319,
    "text": "and that's it. You know, this was how"
  },
  {
    "startMs": 230159,
    "text": "the first version really looked."
  },
  {
    "startMs": 231599,
    "text": "Honestly, I asked them for a dump of of"
  },
  {
    "startMs": 234400,
    "text": "the snowflake schema. I pasted it into a"
  },
  {
    "startMs": 237360,
    "text": "system prompt with a question and then"
  },
  {
    "startMs": 239280,
    "text": "when it generated SQL I actually copy"
  },
  {
    "startMs": 241200,
    "text": "and pasted that in and just ran it"
  },
  {
    "startMs": 242799,
    "text": "myself. You know I just want to see are"
  },
  {
    "startMs": 244319,
    "text": "the models good enough today in order to"
  },
  {
    "startMs": 246799,
    "text": "write valid SQL given some decent"
  },
  {
    "startMs": 249519,
    "text": "structure. And I would say this gave us"
  },
  {
    "startMs": 251760,
    "text": "a little bit of confidence that you know"
  },
  {
    "startMs": 253280,
    "text": "models today aren't that good but maybe"
  },
  {
    "startMs": 255120,
    "text": "we can harness engineer or make the"
  },
  {
    "startMs": 257440,
    "text": "context around it a little better and"
  },
  {
    "startMs": 259519,
    "text": "give us some more guardrails to operate"
  },
  {
    "startMs": 261120,
    "text": "a little better."
  },
  {
    "startMs": 262720,
    "text": "And so if you actually think about what"
  },
  {
    "startMs": 264000,
    "text": "a data scientist actually needs to do"
  },
  {
    "startMs": 265520,
    "text": "when they get a question, you know, they"
  },
  {
    "startMs": 267280,
    "text": "have to process the question, they may"
  },
  {
    "startMs": 269360,
    "text": "have to explore the semantic layer and"
  },
  {
    "startMs": 271120,
    "text": "actually figure out what the join"
  },
  {
    "startMs": 272320,
    "text": "patterns are. They will actually go and"
  },
  {
    "startMs": 274720,
    "text": "execute the SQL. They may go back and do"
  },
  {
    "startMs": 277120,
    "text": "that again if the SQL did not execute or"
  },
  {
    "startMs": 279440,
    "text": "was too expensive. And they'll"
  },
  {
    "startMs": 280960,
    "text": "eventually report on it, including"
  },
  {
    "startMs": 282400,
    "text": "visualize the data, maybe write some"
  },
  {
    "startMs": 284320,
    "text": "paragraphs, maybe do a retro, maybe do"
  },
  {
    "startMs": 286240,
    "text": "some other stuff."
  },
  {
    "startMs": 288080,
    "text": "And so if you think about those"
  },
  {
    "startMs": 289199,
    "text": "different phases, me and the VP of data"
  },
  {
    "startMs": 291040,
    "text": "tried to sit down and map those out into"
  },
  {
    "startMs": 293280,
    "text": "specific agent workloads. And so the"
  },
  {
    "startMs": 295680,
    "text": "second version of this data science"
  },
  {
    "startMs": 297680,
    "text": "agent uh called D0. I'm going to"
  },
  {
    "startMs": 299840,
    "text": "reference D0 from now on is you ask a"
  },
  {
    "startMs": 302479,
    "text": "question. We have a query agent that"
  },
  {
    "startMs": 304479,
    "text": "passes on a query to the planning agent"
  },
  {
    "startMs": 306800,
    "text": "that will then have an execution agent"
  },
  {
    "startMs": 308800,
    "text": "etc. And if you chain all of these"
  },
  {
    "startMs": 311120,
    "text": "together, you actually get something"
  },
  {
    "startMs": 312560,
    "text": "that looks like this where each agent"
  },
  {
    "startMs": 314639,
    "text": "has a very dedicated system prompt"
  },
  {
    "startMs": 316560,
    "text": "focused to what that does with tools"
  },
  {
    "startMs": 318479,
    "text": "scoped to exactly that function. So"
  },
  {
    "startMs": 320880,
    "text": "example here, you can see that for the"
  },
  {
    "startMs": 323039,
    "text": "first one, the planning agent has a read"
  },
  {
    "startMs": 325840,
    "text": "entity YAML and a and a search schemas"
  },
  {
    "startMs": 329199,
    "text": "tool. And so it will only use those"
  },
  {
    "startMs": 332080,
    "text": "capabilities until it has an answer to"
  },
  {
    "startMs": 334000,
    "text": "pass on to the planning agent and then"
  },
  {
    "startMs": 336160,
    "text": "to the SQL agent and then to reporting."
  },
  {
    "startMs": 338800,
    "text": "And this was getting better. You know,"
  },
  {
    "startMs": 340560,
    "text": "we were able to get away from having to"
  },
  {
    "startMs": 342960,
    "text": "copy and paste a SQL and have to come"
  },
  {
    "startMs": 345039,
    "text": "back and report on it. It was now"
  },
  {
    "startMs": 346720,
    "text": "actually doing like the end to end loop"
  },
  {
    "startMs": 348400,
    "text": "from question to answer."
  },
  {
    "startMs": 351600,
    "text": "But we started hitting some walls with"
  },
  {
    "startMs": 354639,
    "text": "this architecture. And around this time"
  },
  {
    "startMs": 356639,
    "text": "we came to the conclusion that you know"
  },
  {
    "startMs": 358479,
    "text": "what you actually need is you need one"
  },
  {
    "startMs": 359919,
    "text": "agent with all the mega context within"
  },
  {
    "startMs": 361840,
    "text": "it and for it to sort of manage its own"
  },
  {
    "startMs": 364319,
    "text": "memory. You know this was around the"
  },
  {
    "startMs": 366400,
    "text": "time when we realized that you want to"
  },
  {
    "startMs": 370000,
    "text": "actually have the agent be able to look"
  },
  {
    "startMs": 371280,
    "text": "back on what it's done sort of reflect"
  },
  {
    "startMs": 373120,
    "text": "and figure out the steps that got to get"
  },
  {
    "startMs": 375199,
    "text": "here. And with the previous model you"
  },
  {
    "startMs": 377039,
    "text": "may have noticed that the only thing"
  },
  {
    "startMs": 378880,
    "text": "that the next agent gets is a summary"
  },
  {
    "startMs": 381280,
    "text": "and a small snippet of the previous"
  },
  {
    "startMs": 382880,
    "text": "thing that was done. Now this way you"
  },
  {
    "startMs": 384960,
    "text": "can imagine that you have one mega agent"
  },
  {
    "startMs": 387360,
    "text": "and internally it manages its own state."
  },
  {
    "startMs": 389600,
    "text": "At some points it's planning, some"
  },
  {
    "startMs": 391039,
    "text": "points it's building, some points it's"
  },
  {
    "startMs": 392400,
    "text": "executing and some points it's"
  },
  {
    "startMs": 393840,
    "text": "reporting. And this is sort of what it"
  },
  {
    "startMs": 396400,
    "text": "looked like. You know you have one big"
  },
  {
    "startMs": 398639,
    "text": "AI call maybe max steps 100 and you give"
  },
  {
    "startMs": 402240,
    "text": "it the ability to manage its own state"
  },
  {
    "startMs": 404240,
    "text": "based on where it's at inside of its"
  },
  {
    "startMs": 406800,
    "text": "execution journey. And so you can see"
  },
  {
    "startMs": 408880,
    "text": "similar tools, you can see a similar"
  },
  {
    "startMs": 410160,
    "text": "shape, but the best part about this is"
  },
  {
    "startMs": 411919,
    "text": "if it ever ran to an error when"
  },
  {
    "startMs": 413600,
    "text": "executing or joining, it could go back"
  },
  {
    "startMs": 415600,
    "text": "and explore more or it could go and read"
  },
  {
    "startMs": 417680,
    "text": "more and figure out what it was doing"
  },
  {
    "startMs": 419120,
    "text": "wrong. And it was very good at this"
  },
  {
    "startMs": 421840,
    "text": "point. We were pretty confident in the"
  },
  {
    "startMs": 424479,
    "text": "actual system at hand and we actually"
  },
  {
    "startMs": 426400,
    "text": "spread it to a few trusted members ever."
  },
  {
    "startMs": 428800,
    "text": "You know, this is a very powerful tool"
  },
  {
    "startMs": 430080,
    "text": "and we didn't really want to put in the"
  },
  {
    "startMs": 431440,
    "text": "hands of the wrong people or people that"
  },
  {
    "startMs": 432880,
    "text": "were using very critical workloads. So,"
  },
  {
    "startMs": 435280,
    "text": "we got to a few people's hands and the"
  },
  {
    "startMs": 437120,
    "text": "immediate response was it was awful. You"
  },
  {
    "startMs": 438880,
    "text": "know, we thought we were cooking. We"
  },
  {
    "startMs": 440240,
    "text": "thought this was, you know, nailing 30%"
  },
  {
    "startMs": 442240,
    "text": "of our evals, but we couldn't have"
  },
  {
    "startMs": 444160,
    "text": "anticipated some of the questions that"
  },
  {
    "startMs": 445840,
    "text": "were being asked. And for us to spend"
  },
  {
    "startMs": 447840,
    "text": "more time manually mapping out some of"
  },
  {
    "startMs": 449840,
    "text": "these scenarios, it didn't seem like a"
  },
  {
    "startMs": 452400,
    "text": "very scalable way to do this."
  },
  {
    "startMs": 455199,
    "text": "And then claude code and opus 4.5 came"
  },
  {
    "startMs": 459120,
    "text": "out. Well, more like Opus 4.5 came out"
  },
  {
    "startMs": 461680,
    "text": "and it in tangent with claw code which"
  },
  {
    "startMs": 463759,
    "text": "is so powerful. You know they sort of"
  },
  {
    "startMs": 466240,
    "text": "unlocked the concept of a file system"
  },
  {
    "startMs": 468960,
    "text": "agent and we on the side were like wow"
  },
  {
    "startMs": 472560,
    "text": "clawed code and Opus 4.5 is basically"
  },
  {
    "startMs": 475199,
    "text": "AGI compared to what we had before. You"
  },
  {
    "startMs": 477440,
    "text": "know it would answer most of our"
  },
  {
    "startMs": 478879,
    "text": "questions without even without even"
  },
  {
    "startMs": 482000,
    "text": "missing a beat um compared to the"
  },
  {
    "startMs": 483919,
    "text": "handgrown agent we had. And when we"
  },
  {
    "startMs": 485919,
    "text": "tried to step back and wonder what we"
  },
  {
    "startMs": 487440,
    "text": "were doing wrong and why this was so"
  },
  {
    "startMs": 488879,
    "text": "much better, we realized that the big"
  },
  {
    "startMs": 491599,
    "text": "unlock was that it was just a file"
  },
  {
    "startMs": 493759,
    "text": "system. You know, we it had a very"
  },
  {
    "startMs": 495759,
    "text": "minimal set of tools, list file, read"
  },
  {
    "startMs": 497919,
    "text": "file, run bash, and we gave a few more"
  },
  {
    "startMs": 500960,
    "text": "here for uh our own data agent use case."
  },
  {
    "startMs": 504479,
    "text": "But the biggest thing was it was able to"
  },
  {
    "startMs": 506479,
    "text": "use the tools that agents are well"
  },
  {
    "startMs": 508080,
    "text": "trained on and was able to explore and"
  },
  {
    "startMs": 510479,
    "text": "write work where it needs to. you know,"
  },
  {
    "startMs": 512880,
    "text": "we weren't giving it claw code was not"
  },
  {
    "startMs": 515279,
    "text": "giving it a very prescriptive set of"
  },
  {
    "startMs": 516719,
    "text": "tools. It was sort of just letting it go"
  },
  {
    "startMs": 518399,
    "text": "wild and explore emergent behavior. And"
  },
  {
    "startMs": 521440,
    "text": "so from this, we learned that you can"
  },
  {
    "startMs": 523279,
    "text": "really just use a file system. You know,"
  },
  {
    "startMs": 525200,
    "text": "we we saw the learnings from claw code"
  },
  {
    "startMs": 527200,
    "text": "and how powerful it was given that it"
  },
  {
    "startMs": 529440,
    "text": "just executes locally. And we tried to"
  },
  {
    "startMs": 532320,
    "text": "rebuild it in a way that was very cloud"
  },
  {
    "startMs": 534080,
    "text": "codeesque. You know, it was now going to"
  },
  {
    "startMs": 536399,
    "text": "run in a sandbox. That sandbox would"
  },
  {
    "startMs": 538480,
    "text": "dump the whole semantic layer into it."
  },
  {
    "startMs": 540640,
    "text": "You could the agent would be able to"
  },
  {
    "startMs": 542320,
    "text": "grab, bash, read file, write file all"
  },
  {
    "startMs": 544480,
    "text": "around to figure out what it needs and"
  },
  {
    "startMs": 546240,
    "text": "we would just sprinkle a few tools on"
  },
  {
    "startMs": 547600,
    "text": "top to make sure it could do everything"
  },
  {
    "startMs": 548959,
    "text": "that is versel specific."
  },
  {
    "startMs": 552399,
    "text": "And this was actually the biggest unlock"
  },
  {
    "startMs": 554480,
    "text": "ever. You know, the leap from single"
  },
  {
    "startMs": 557360,
    "text": "agent to cloud code SDK and then from"
  },
  {
    "startMs": 559760,
    "text": "cloud code SDK to file system agent in"
  },
  {
    "startMs": 562560,
    "text": "general, fine-tuned or purpose-built for"
  },
  {
    "startMs": 565200,
    "text": "our use case was an amazing leap. At"
  },
  {
    "startMs": 567839,
    "text": "this point, we were starting to get"
  },
  {
    "startMs": 568800,
    "text": "ready to give it away to more people at"
  },
  {
    "startMs": 571040,
    "text": "Versell."
  },
  {
    "startMs": 572640,
    "text": "And at this point, the eval score"
  },
  {
    "startMs": 575600,
    "text": "basically doubled. And I wrote this uh"
  },
  {
    "startMs": 577760,
    "text": "this is basically how it looks. Um it's"
  },
  {
    "startMs": 579519,
    "text": "very simple. You just give it a bash"
  },
  {
    "startMs": 581040,
    "text": "tool. We have a nice helper called bash"
  },
  {
    "startMs": 582880,
    "text": "tool on npm and you attach it to a"
  },
  {
    "startMs": 585120,
    "text": "sandbox and you can attach files to the"
  },
  {
    "startMs": 586959,
    "text": "sandbox for it to read, write and"
  },
  {
    "startMs": 588560,
    "text": "execute."
  },
  {
    "startMs": 590480,
    "text": "And after this revelation and after I"
  },
  {
    "startMs": 592320,
    "text": "saw that we were passing so many of the"
  },
  {
    "startMs": 594640,
    "text": "questions that we failed to do before, I"
  },
  {
    "startMs": 596480,
    "text": "wrote this banger blog post. It's uh"
  },
  {
    "startMs": 598880,
    "text": "it's actually up today. And the week"
  },
  {
    "startMs": 601040,
    "text": "that I wrote this, it was responsible"
  },
  {
    "startMs": 602800,
    "text": "for 70% of our versel.com traffic. So"
  },
  {
    "startMs": 605839,
    "text": "you know it's a banger. And after that,"
  },
  {
    "startMs": 609120,
    "text": "the next logical step was that we want"
  },
  {
    "startMs": 612399,
    "text": "to figure out the common use cases we"
  },
  {
    "startMs": 614480,
    "text": "had. So by then we've already sort of"
  },
  {
    "startMs": 616399,
    "text": "let a leash on all of our cell and we"
  },
  {
    "startMs": 619279,
    "text": "were getting thousands of queries a day"
  },
  {
    "startMs": 621279,
    "text": "from people wanting everything from"
  },
  {
    "startMs": 623120,
    "text": "customer metrics sales metrics number"
  },
  {
    "startMs": 625440,
    "text": "metrics npm downloads and it turns out"
  },
  {
    "startMs": 628880,
    "text": "that a lot of these queries are actually"
  },
  {
    "startMs": 630079,
    "text": "the same in shape you know there's only"
  },
  {
    "startMs": 631680,
    "text": "so many ways you can do an aggregation"
  },
  {
    "startMs": 633600,
    "text": "only so many ways you can look up a"
  },
  {
    "startMs": 634959,
    "text": "product only so many ways you can do"
  },
  {
    "startMs": 636240,
    "text": "billing info and so we actually have a"
  },
  {
    "startMs": 638640,
    "text": "recurring job that takes the most recent"
  },
  {
    "startMs": 640480,
    "text": "queries and tries to distill them into a"
  },
  {
    "startMs": 642480,
    "text": "skill and right now we have roughly 100"
  },
  {
    "startMs": 644880,
    "text": "skills that do a mix of aggregation all"
  },
  {
    "startMs": 647440,
    "text": "the way through looking up specific data"
  },
  {
    "startMs": 649120,
    "text": "about certain people. And we found this"
  },
  {
    "startMs": 651839,
    "text": "very effective because if you think"
  },
  {
    "startMs": 653120,
    "text": "about every new agent run, it sort of"
  },
  {
    "startMs": 655600,
    "text": "just starts from nothing. You know,"
  },
  {
    "startMs": 656880,
    "text": "there's really no pre-established"
  },
  {
    "startMs": 658160,
    "text": "context besides, you know, the semantic"
  },
  {
    "startMs": 660399,
    "text": "layer and the system prompt. But with a"
  },
  {
    "startMs": 662320,
    "text": "skill, it already starts off with a lot"
  },
  {
    "startMs": 664160,
    "text": "of contextual knowledge that has"
  },
  {
    "startMs": 665920,
    "text": "otherwise already been done."
  },
  {
    "startMs": 669200,
    "text": "And this is roughly how it looks. It's"
  },
  {
    "startMs": 671120,
    "text": "very similar to the previous one, but"
  },
  {
    "startMs": 672560,
    "text": "the inclusion of a skills folder is"
  },
  {
    "startMs": 674079,
    "text": "actually very powerful. Um, we also"
  },
  {
    "startMs": 676160,
    "text": "built this tool at Verscell called"
  },
  {
    "startMs": 677440,
    "text": "Skillsh. It's the most popular way to"
  },
  {
    "startMs": 679440,
    "text": "find agent skills and run them yourself."
  },
  {
    "startMs": 683360,
    "text": "And I I'm saying all this because this"
  },
  {
    "startMs": 685600,
    "text": "journey is something that most of you"
  },
  {
    "startMs": 687440,
    "text": "may hit once in a while where you start"
  },
  {
    "startMs": 689200,
    "text": "from something simple and you gradually"
  },
  {
    "startMs": 690640,
    "text": "add complexity and you eventually hit a"
  },
  {
    "startMs": 692480,
    "text": "system in which you can ship to prod."
  },
  {
    "startMs": 695279,
    "text": "And I'm telling you this because at"
  },
  {
    "startMs": 697680,
    "text": "every step along building this agent,"
  },
  {
    "startMs": 699920,
    "text": "someone Everell was agent curious and"
  },
  {
    "startMs": 702640,
    "text": "they tried to fork off of my DZero agent"
  },
  {
    "startMs": 705519,
    "text": "and build their own. And at every step,"
  },
  {
    "startMs": 707680,
    "text": "we sort of had a better way to do"
  },
  {
    "startMs": 709600,
    "text": "something that was not previously known."
  },
  {
    "startMs": 711760,
    "text": "And we were wondering like what if"
  },
  {
    "startMs": 713360,
    "text": "people today could start from the very"
  },
  {
    "startMs": 715600,
    "text": "last insight and not have to ever start"
  },
  {
    "startMs": 717839,
    "text": "from just a simple prompt or from"
  },
  {
    "startMs": 720079,
    "text": "reinventing best principles from first"
  },
  {
    "startMs": 723120,
    "text": "principles."
  },
  {
    "startMs": 724640,
    "text": "And so we actually thought what if we"
  },
  {
    "startMs": 726639,
    "text": "built the Nex.js for agents. For those"
  },
  {
    "startMs": 729600,
    "text": "that don't know, Nex.js is a popular web"
  },
  {
    "startMs": 731839,
    "text": "framework that Verscell built that"
  },
  {
    "startMs": 733760,
    "text": "invented this thing of file system uh"
  },
  {
    "startMs": 736639,
    "text": "framework defined infrastructure. You"
  },
  {
    "startMs": 738480,
    "text": "don't have to worry about where things"
  },
  {
    "startMs": 739920,
    "text": "go. You just have to write files in the"
  },
  {
    "startMs": 742000,
    "text": "right conventions and it automatically"
  },
  {
    "startMs": 744399,
    "text": "declares where they should go. Your"
  },
  {
    "startMs": 746160,
    "text": "pages go to the CDN. Your serverless"
  },
  {
    "startMs": 748320,
    "text": "functions go there. Your caching goes in"
  },
  {
    "startMs": 750399,
    "text": "the middle. And we thought, you know,"
  },
  {
    "startMs": 752800,
    "text": "building agents should be this simple."
  },
  {
    "startMs": 754560,
    "text": "You should only have to create a skills"
  },
  {
    "startMs": 755920,
    "text": "folder, a tools folder, a channels"
  },
  {
    "startMs": 757839,
    "text": "folder, and you should be able to just"
  },
  {
    "startMs": 759120,
    "text": "declare these very easily. And the"
  },
  {
    "startMs": 760880,
    "text": "framework should know exactly how to"
  },
  {
    "startMs": 762399,
    "text": "make an agent out of it."
  },
  {
    "startMs": 765040,
    "text": "And that's why two weeks ago we released"
  },
  {
    "startMs": 766880,
    "text": "Eve. Eve is a agent framework like the"
  },
  {
    "startMs": 769600,
    "text": "next.js GS for agents where it's very"
  },
  {
    "startMs": 771519,
    "text": "easy from just starting with a sample"
  },
  {
    "startMs": 773600,
    "text": "template to having a fully agent ready"
  },
  {
    "startMs": 776079,
    "text": "and being able to add in your own custom"
  },
  {
    "startMs": 777839,
    "text": "knowledge, your own custom tools and"
  },
  {
    "startMs": 779760,
    "text": "even integrated into the channels that"
  },
  {
    "startMs": 781440,
    "text": "you are familiar with."
  },
  {
    "startMs": 784160,
    "text": "This is roughly what we think an agent"
  },
  {
    "startMs": 785920,
    "text": "actually looks like. You know, an agent"
  },
  {
    "startMs": 787440,
    "text": "has a runtime and it has channels. And"
  },
  {
    "startMs": 789519,
    "text": "in that runtime, you're going to have"
  },
  {
    "startMs": 790880,
    "text": "durability. You're going to want to run"
  },
  {
    "startMs": 792639,
    "text": "things in an isolate environment. You're"
  },
  {
    "startMs": 794160,
    "text": "going to want to call into different"
  },
  {
    "startMs": 795200,
    "text": "models. And you're going to want to have"
  },
  {
    "startMs": 796720,
    "text": "connections. And we built this with open"
  },
  {
    "startMs": 799200,
    "text": "source in mind. You know, we built Eve"
  },
  {
    "startMs": 801040,
    "text": "so you can plug in your own open source"
  },
  {
    "startMs": 802720,
    "text": "adapters for Postgress, OpenAI's uh"
  },
  {
    "startMs": 805200,
    "text": "responses API, Docker, other connectors."
  },
  {
    "startMs": 808639,
    "text": "But we also made it incredibly easy to"
  },
  {
    "startMs": 810240,
    "text": "deploy in Verscell. The only thing here"
  },
  {
    "startMs": 812000,
    "text": "you see different is that everything"
  },
  {
    "startMs": 813519,
    "text": "here is using a Verscell product that"
  },
  {
    "startMs": 815200,
    "text": "we've been building over the years in"
  },
  {
    "startMs": 816880,
    "text": "order to make it easy to build these"
  },
  {
    "startMs": 818399,
    "text": "experiences. Versell workflows for"
  },
  {
    "startMs": 820399,
    "text": "durability, sandbox for secure"
  },
  {
    "startMs": 822320,
    "text": "execution, and Verscell connect,"
  },
  {
    "startMs": 824000,
    "text": "something we just released to make it"
  },
  {
    "startMs": 825680,
    "text": "easy to generate short-lived ODC tokens"
  },
  {
    "startMs": 828240,
    "text": "for connections."
  },
  {
    "startMs": 831040,
    "text": "And we actually rewrote the whole D0ero"
  },
  {
    "startMs": 833279,
    "text": "agent in Eve as we were building Eve and"
  },
  {
    "startMs": 836000,
    "text": "from the convoluted structures behind"
  },
  {
    "startMs": 837519,
    "text": "the scenes that you did not see from the"
  },
  {
    "startMs": 839120,
    "text": "code. Um, this is roughly how the file"
  },
  {
    "startMs": 841040,
    "text": "system looks. It's very simple. You have"
  },
  {
    "startMs": 842880,
    "text": "a bunch of system instructions, a couple"
  },
  {
    "startMs": 845040,
    "text": "skills, a couple tools, and it's very"
  },
  {
    "startMs": 847760,
    "text": "easy to compose this into a real agent,"
  },
  {
    "startMs": 849600,
    "text": "and it's very easy to iterate on. We"
  },
  {
    "startMs": 852320,
    "text": "actually gave this out to a few beta"
  },
  {
    "startMs": 853680,
    "text": "customers before we actually fully"
  },
  {
    "startMs": 855279,
    "text": "released it two weeks ago at our London"
  },
  {
    "startMs": 856959,
    "text": "event. And this one company that"
  },
  {
    "startMs": 858880,
    "text": "partners closely with us, Aura. They've"
  },
  {
    "startMs": 860800,
    "text": "rebuilt their agent that's sort of like"
  },
  {
    "startMs": 862320,
    "text": "a mini claw to go and test people's"
  },
  {
    "startMs": 865360,
    "text": "services. It goes to websites, installs"
  },
  {
    "startMs": 867839,
    "text": "them, it tries to use them. And they've"
  },
  {
    "startMs": 870800,
    "text": "seen incredible success on building"
  },
  {
    "startMs": 873199,
    "text": "their own agent from the ground up using"
  },
  {
    "startMs": 875040,
    "text": "Eve compared to using an off-the-shelf"
  },
  {
    "startMs": 877279,
    "text": "cloud code. Fewer steps, better"
  },
  {
    "startMs": 880320,
    "text": "successes, as well as better insights."
  },
  {
    "startMs": 885040,
    "text": "And when you deploy Eve to Verscell, you"
  },
  {
    "startMs": 887360,
    "text": "get observability observability out of"
  },
  {
    "startMs": 889279,
    "text": "the box. You can see here that you get"
  },
  {
    "startMs": 890800,
    "text": "all the agent runs, you see all the tool"
  },
  {
    "startMs": 892399,
    "text": "calls, you see each step it takes as"
  },
  {
    "startMs": 895120,
    "text": "well as maybe some estimated costs and"
  },
  {
    "startMs": 896800,
    "text": "some optimizations you could potentially"
  },
  {
    "startMs": 898320,
    "text": "take."
  },
  {
    "startMs": 900399,
    "text": "And you can get start today at eve.dev."
  },
  {
    "startMs": 903040,
    "text": "You can just clone it and you can just"
  },
  {
    "startMs": 904399,
    "text": "start a template, deploy easily,"
  },
  {
    "startMs": 906160,
    "text": "self-host if you need. And the reason"
  },
  {
    "startMs": 908240,
    "text": "why I bring this up is because I hope"
  },
  {
    "startMs": 910800,
    "text": "that there will be more and more"
  },
  {
    "startMs": 912399,
    "text": "business specific use case agents. You"
  },
  {
    "startMs": 915199,
    "text": "know, before we built Ezero, we actually"
  },
  {
    "startMs": 917199,
    "text": "battle tested a lot of the industry"
  },
  {
    "startMs": 920320,
    "text": "well-funded startups that were doing"
  },
  {
    "startMs": 922480,
    "text": "these vertical agents that were"
  },
  {
    "startMs": 924160,
    "text": "dedicated to taking your Snowflake"
  },
  {
    "startMs": 926160,
    "text": "instance and making it so their agent"
  },
  {
    "startMs": 928639,
    "text": "could run Snowflake queries against it."
  },
  {
    "startMs": 931440,
    "text": "But we found out that what really makes"
  },
  {
    "startMs": 933040,
    "text": "this agent good is it has a lot of very"
  },
  {
    "startMs": 936480,
    "text": "specific uh company knowledge. You know,"
  },
  {
    "startMs": 939360,
    "text": "the way that Versel is a web- based"
  },
  {
    "startMs": 941519,
    "text": "company. We have a lot of customers that"
  },
  {
    "startMs": 943279,
    "text": "have websites and web properties. That"
  },
  {
    "startMs": 945760,
    "text": "goes a lot deeper into when you should"
  },
  {
    "startMs": 948800,
    "text": "query for what and what things link to"
  },
  {
    "startMs": 950639,
    "text": "what. And so a lot of these"
  },
  {
    "startMs": 952800,
    "text": "off-the-shelf agents, they're great."
  },
  {
    "startMs": 954320,
    "text": "They're good to try, but I think if you"
  },
  {
    "startMs": 956160,
    "text": "really want to get the most juice out of"
  },
  {
    "startMs": 957600,
    "text": "a squeeze, you should really try to"
  },
  {
    "startMs": 958880,
    "text": "build your own agent and add in as much"
  },
  {
    "startMs": 960800,
    "text": "company specific knowledge as you can."
  },
  {
    "startMs": 963759,
    "text": "Today, you know, we've had 20 roughly"
  },
  {
    "startMs": 966880,
    "text": "decently PMF agents adversel that range"
  },
  {
    "startMs": 969920,
    "text": "from anything from marketing retros to"
  },
  {
    "startMs": 973120,
    "text": "figure out who to reach out to to the"
  },
  {
    "startMs": 975519,
    "text": "first ever red line of a contract when"
  },
  {
    "startMs": 978399,
    "text": "legal sees a new negotiation all the way"
  },
  {
    "startMs": 980800,
    "text": "to my data science agent helping with"
  },
  {
    "startMs": 982880,
    "text": "with data queries. And that goes to show"
  },
  {
    "startMs": 986000,
    "text": "that we ever have been very"
  },
  {
    "startMs": 987839,
    "text": "agent-filled. You know, all of this"
  },
  {
    "startMs": 990320,
    "text": "stuff is actually saving us a lot of"
  },
  {
    "startMs": 992560,
    "text": "time. The data team has never been more"
  },
  {
    "startMs": 994639,
    "text": "productive. They have more time to go"
  },
  {
    "startMs": 996639,
    "text": "and improve the performance of"
  },
  {
    "startMs": 998240,
    "text": "Snowflake, to add new data sources that"
  },
  {
    "startMs": 1000160,
    "text": "were missing, to fill in the gaps that"
  },
  {
    "startMs": 1002160,
    "text": "they previously did not have time to"
  },
  {
    "startMs": 1003680,
    "text": "because they were so busy writing"
  },
  {
    "startMs": 1005279,
    "text": "queries. And I think it's never been"
  },
  {
    "startMs": 1008000,
    "text": "easier for you at your big, small,"
  },
  {
    "startMs": 1010800,
    "text": "medium-sized company to sort of automate"
  },
  {
    "startMs": 1013279,
    "text": "away some of the things that you do not"
  },
  {
    "startMs": 1015279,
    "text": "want to do or some of the things that"
  },
  {
    "startMs": 1016560,
    "text": "you're spending too much time doing. You"
  },
  {
    "startMs": 1018480,
    "text": "know, I think a lot of HR, finance,"
  },
  {
    "startMs": 1020880,
    "text": "sales can be somewhat automated with"
  },
  {
    "startMs": 1024079,
    "text": "agents. And I think Eve is the best way"
  },
  {
    "startMs": 1026240,
    "text": "to build said agents today."
  },
  {
    "startMs": 1029679,
    "text": "And these are my socials. Thank you all"
  },
  {
    "startMs": 1031679,
    "text": "for coming and listening. I'm Andrew and"
  },
  {
    "startMs": 1033679,
    "text": "I'll be around if you want to chat"
  },
  {
    "startMs": 1035360,
    "text": "outside."
  },
  {
    "startMs": 1051942,
    "text": ">> [music]"
  }
]