[ { "speaker": "host", "text": "I feel like agents for consumers are I feel like agents for consumers are" }, { "speaker": "host", "text": "like fairly overh right now Goot um like fairly overh right now Goot um like fairly overh right now Goot um" }, { "speaker": "host", "text": "trying to have a agent like fully book a trying to have a agent like fully book a trying to have a agent like fully book a" }, { "speaker": "guest", "text": "vacation for you almost just as hard as vacation for you almost just as hard as vacation for you almost just as hard as" }, { "speaker": "guest", "text": "just going and booking it yourself take just going and booking it yourself take just going and booking it yourself take" }, { "speaker": "guest", "text": "one today we're going behind the scenes one today we're going behind the scenes one today we're going behind the scenes" }, { "speaker": "host", "text": "on one of our recent blog posts building on one of our recent blog posts building on one of our recent blog posts building" }, { "speaker": "host", "text": "effective agents I'm Alex I lead Claude effective agents I'm Alex I lead Claude effective agents I'm Alex I lead Claude" }, { "speaker": "host", "text": "relations here at anthropic I'm Eric I'm relations here at anthropic I'm Eric I'm relations here at anthropic I'm Eric I'm" }, { "speaker": "guest", "text": "on the research team at anthropic I'm on the research team at anthropic I'm on the research team at anthropic I'm" }, { "speaker": "guest", "text": "Barry I'm on the applied AI team I'm Barry I'm on the applied AI team I'm Barry I'm on the applied AI team I'm" }, { "speaker": "guest", "text": "going to kick us off here for viewers going to kick us off here for viewers going to kick us off here for viewers" }, { "speaker": "host", "text": "just jumping in uh what's the quick just jumping in uh what's the quick just jumping in uh what's the quick" }, { "speaker": "host", "text": "version of what an agent actually is I version of what an agent actually is I version of what an agent actually is I" }, { "speaker": "host", "text": "mean there's a million definitions of it mean there's a million definitions of it mean there's a million definitions of it" }, { "speaker": "guest", "text": "and why should a developer or somebody and why should a developer or somebody and why should a developer or somebody" }, { "speaker": "guest", "text": "that's actually building with AI care that's actually building with AI care that's actually building with AI care" }, { "speaker": "guest", "text": "about these things um Eric maybe we can about these things um Eric maybe we can about these things um Eric maybe we can" }, { "speaker": "host", "text": "start with you sure yeah so I think start with you sure yeah so I think start with you sure yeah so I think" }, { "speaker": "host", "text": "something we explored in the blog post something we explored in the blog post something we explored in the blog post" }, { "speaker": "host", "text": "is that first of all a lot of people is that first of all a lot of people is that first of all a lot of people" }, { "speaker": "guest", "text": "have been saying everything is an agent have been saying everything is an agent have been saying everything is an agent" }, { "speaker": "guest", "text": "referring to almost anything more than referring to almost anything more than referring to almost anything more than" }, { "speaker": "guest", "text": "just a single llm call um one of the just a single llm call um one of the just a single llm call um one of the" }, { "speaker": "host", "text": "things we tried to do in the blog post things we tried to do in the blog post things we tried to do in the blog post" }, { "speaker": "host", "text": "is really kind of separate this out of is really kind of separate this out of is really kind of separate this out of" }, { "speaker": "host", "text": "like hey there's workflows which is like hey there's workflows which is like hey there's workflows which is" }, { "speaker": "guest", "text": "where you have a few llm calls chain where you have a few llm calls chain where you have a few llm calls chain" }, { "speaker": "guest", "text": "together and really what we think an together and really what we think an together and really what we think an" }, { "speaker": "guest", "text": "agent is is where you're letting the llm agent is is where you're letting the llm agent is is where you're letting the llm" }, { "speaker": "host", "text": "decide sort of how many times to run decide sort of how many times to run decide sort of how many times to run" }, { "speaker": "host", "text": "you're having it continuing to Loop you're having it continuing to Loop you're having it continuing to Loop" }, { "speaker": "host", "text": "until it's found a resolution uh and until it's found a resolution uh and until it's found a resolution uh and" }, { "speaker": "guest", "text": "that could be you know talking to a that could be you know talking to a that could be you know talking to a" }, { "speaker": "guest", "text": "customer for customer support that could customer for customer support that could customer for customer support that could" }, { "speaker": "guest", "text": "be iterating on code changes but be iterating on code changes but be iterating on code changes but" }, { "speaker": "host", "text": "something where like you don't know how something where like you don't know how something where like you don't know how" }, { "speaker": "host", "text": "many steps it's going to take to many steps it's going to take to many steps it's going to take to" }, { "speaker": "host", "text": "complete that's really sort of what we complete that's really sort of what we complete that's really sort of what we" }, { "speaker": "guest", "text": "consider an agent interesting so in the consider an agent interesting so in the consider an agent interesting so in the" }, { "speaker": "guest", "text": "definition of an agent we are letting definition of an agent we are letting definition of an agent we are letting" }, { "speaker": "guest", "text": "the llm kind of pick its own fate and the llm kind of pick its own fate and the llm kind of pick its own fate and" }, { "speaker": "host", "text": "decide what it wants to do what actions decide what it wants to do what actions decide what it wants to do what actions" }, { "speaker": "host", "text": "to take instead of us predefining a path to take instead of us predefining a path to take instead of us predefining a path" }, { "speaker": "host", "text": "for it exactly it's more autonomous for it exactly it's more autonomous for it exactly it's more autonomous" }, { "speaker": "guest", "text": "whereas a workflow you can kind of think whereas a workflow you can kind of think whereas a workflow you can kind of think" }, { "speaker": "guest", "text": "of it as like um uh you know a yeah a of it as like um uh you know a yeah a of it as like um uh you know a yeah a" }, { "speaker": "guest", "text": "workflow or sort of like it's on Rails workflow or sort of like it's on Rails workflow or sort of like it's on Rails" }, { "speaker": "host", "text": "through a fixed number of steps I see so through a fixed number of steps I see so through a fixed number of steps I see so" }, { "speaker": "host", "text": "this distinction I assume this was the this distinction I assume this was the this distinction I assume this was the" }, { "speaker": "host", "text": "result of many many conversations with result of many many conversations with result of many many conversations with" }, { "speaker": "guest", "text": "customers and working with different customers and working with different customers and working with different" }, { "speaker": "guest", "text": "teams and even trying things ourself teams and even trying things ourself teams and even trying things ourself" }, { "speaker": "guest", "text": "Barry can you speak more to maybe what Barry can you speak more to maybe what Barry can you speak more to maybe what" }, { "speaker": "host", "text": "that looks like as we got to create this that looks like as we got to create this that looks like as we got to create this" }, { "speaker": "host", "text": "divide between a workflow and and what divide between a workflow and and what divide between a workflow and and what" }, { "speaker": "host", "text": "sort of patterns surprised you the most sort of patterns surprised you the most sort of patterns surprised you the most" }, { "speaker": "guest", "text": "as you were going through this sure um as you were going through this sure um as you were going through this sure um" }, { "speaker": "guest", "text": "honestly I think all of this kind of honestly I think all of this kind of honestly I think all of this kind of" }, { "speaker": "guest", "text": "evolved as like model got better and evolved as like model got better and evolved as like model got better and" }, { "speaker": "host", "text": "like teams got more sophisticated uh we like teams got more sophisticated uh we like teams got more sophisticated uh we" }, { "speaker": "host", "text": "both work with a large number of like both work with a large number of like both work with a large number of like" }, { "speaker": "host", "text": "customers were very sophisticated and we customers were very sophisticated and we customers were very sophisticated and we" }, { "speaker": "guest", "text": "kind of went from like having a single kind of went from like having a single kind of went from like having a single" }, { "speaker": "guest", "text": "LM to having a lot of L and like LM to having a lot of L and like LM to having a lot of L and like" }, { "speaker": "guest", "text": "eventually having like LM orchestrating eventually having like LM orchestrating eventually having like LM orchestrating" }, { "speaker": "host", "text": "themselves um so you know like one of themselves um so you know like one of themselves um so you know like one of" }, { "speaker": "host", "text": "the reasons why we decided to create the reasons why we decided to create the reasons why we decided to create" }, { "speaker": "host", "text": "this distinction is because we start to this distinction is because we start to this distinction is because we start to" }, { "speaker": "guest", "text": "see these two distinct patterns where see these two distinct patterns where see these two distinct patterns where" }, { "speaker": "guest", "text": "you have workflows that's pre- you have workflows that's pre- you have workflows that's pre-" }, { "speaker": "guest", "text": "orchestrated by code and then you have orchestrated by code and then you have orchestrated by code and then you have" }, { "speaker": "host", "text": "you know agent which is like a simpler you know agent which is like a simpler you know agent which is like a simpler" }, { "speaker": "host", "text": "but you know complex in in other sense U but you know complex in in other sense U but you know complex in in other sense U" }, { "speaker": "host", "text": "like different shape that that that like different shape that that that like different shape that that that" }, { "speaker": "guest", "text": "we're starting to see um really I think we're starting to see um really I think we're starting to see um really I think" }, { "speaker": "guest", "text": "like as the models and all of the tools like as the models and all of the tools like as the models and all of the tools" }, { "speaker": "guest", "text": "start to get better um you know agents start to get better um you know agents start to get better um you know agents" }, { "speaker": "host", "text": "are becoming more and more prevalent and are becoming more and more prevalent and are becoming more and more prevalent and" }, { "speaker": "host", "text": "more and more capable and that's when we more and more capable and that's when we more and more capable and that's when we" }, { "speaker": "host", "text": "decided like hey this is probably a good decided like hey this is probably a good decided like hey this is probably a good" }, { "speaker": "guest", "text": "time for us to to like get give a formal time for us to to like get give a formal time for us to to like get give a formal" }, { "speaker": "guest", "text": "definition so in practice if you're a definition so in practice if you're a definition so in practice if you're a" }, { "speaker": "guest", "text": "developer implementing one of these developer implementing one of these developer implementing one of these" }, { "speaker": "host", "text": "things what would that actually look things what would that actually look things what would that actually look" }, { "speaker": "host", "text": "like in your your code as you're like in your your code as you're like in your your code as you're" }, { "speaker": "host", "text": "starting to build this like the starting to build this like the starting to build this like the" }, { "speaker": "guest", "text": "differences between uh like maybe we differences between uh like maybe we differences between uh like maybe we" }, { "speaker": "guest", "text": "actually go down to like the prompt actually go down to like the prompt actually go down to like the prompt" }, { "speaker": "guest", "text": "level here what what does an agent level here what what does an agent level here what what does an agent" }, { "speaker": "host", "text": "prompt look like or flow and what does prompt look like or flow and what does prompt look like or flow and what does" }, { "speaker": "host", "text": "like a workflow look like yeah so I like a workflow look like yeah so I like a workflow look like yeah so I" }, { "speaker": "host", "text": "think a workflow prompt looks like you think a workflow prompt looks like you think a workflow prompt looks like you" }, { "speaker": "guest", "text": "have one prompt uh you take the output have one prompt uh you take the output have one prompt uh you take the output" }, { "speaker": "guest", "text": "of it you feed it into prompt B take the of it you feed it into prompt B take the of it you feed it into prompt B take the" }, { "speaker": "guest", "text": "output of that feed it into prompt C and output of that feed it into prompt C and output of that feed it into prompt C and" }, { "speaker": "host", "text": "then you're done kind of there's this then you're done kind of there's this then you're done kind of there's this" }, { "speaker": "host", "text": "straight line fixed number of steps you straight line fixed number of steps you straight line fixed number of steps you" }, { "speaker": "host", "text": "know exactly what's going to happen and know exactly what's going to happen and know exactly what's going to happen and" }, { "speaker": "guest", "text": "maybe you have some extra code that sort maybe you have some extra code that sort maybe you have some extra code that sort" }, { "speaker": "guest", "text": "of checks the intermediate results of of checks the intermediate results of of checks the intermediate results of" }, { "speaker": "guest", "text": "these and makes sure okay um but you these and makes sure okay um but you these and makes sure okay um but you" }, { "speaker": "host", "text": "kind of know exactly what's going to kind of know exactly what's going to kind of know exactly what's going to" }, { "speaker": "host", "text": "happen in one of these paths um and each happen in one of these paths um and each happen in one of these paths um and each" }, { "speaker": "host", "text": "of those prompts is um is sort of a very of those prompts is um is sort of a very of those prompts is um is sort of a very" }, { "speaker": "guest", "text": "specific prompt just sort of taking one specific prompt just sort of taking one specific prompt just sort of taking one" }, { "speaker": "guest", "text": "input and transforming it into another input and transforming it into another input and transforming it into another" }, { "speaker": "guest", "text": "output for instance uh maybe one of output for instance uh maybe one of output for instance uh maybe one of" }, { "speaker": "host", "text": "these prompts is taking in the user these prompts is taking in the user these prompts is taking in the user" }, { "speaker": "host", "text": "question and uh categorizing it into one question and uh categorizing it into one question and uh categorizing it into one" }, { "speaker": "host", "text": "of five categories so that then the next of five categories so that then the next of five categories so that then the next" }, { "speaker": "guest", "text": "prompt can be more specific for that prompt can be more specific for that prompt can be more specific for that" }, { "speaker": "guest", "text": "okay in contrast an agent prompt will be okay in contrast an agent prompt will be okay in contrast an agent prompt will be" }, { "speaker": "guest", "text": "sort of much more open-ended sort of much more open-ended sort of much more open-ended" }, { "speaker": "host", "text": "and usually give the model tools or and usually give the model tools or and usually give the model tools or" }, { "speaker": "host", "text": "multiple things to check and say hey multiple things to check and say hey multiple things to check and say hey" }, { "speaker": "host", "text": "here's the question and you can do web here's the question and you can do web here's the question and you can do web" }, { "speaker": "guest", "text": "searches or you can edit these code searches or you can edit these code searches or you can edit these code" }, { "speaker": "guest", "text": "files or run code and keep doing this files or run code and keep doing this files or run code and keep doing this" }, { "speaker": "guest", "text": "until you have the answer I see so until you have the answer I see so until you have the answer I see so" }, { "speaker": "host", "text": "that's a few different use cases there that's a few different use cases there that's a few different use cases there" }, { "speaker": "host", "text": "um that makes sense as we start to um that makes sense as we start to um that makes sense as we start to" }, { "speaker": "host", "text": "arrive at like these different arrive at like these different arrive at like these different" }, { "speaker": "guest", "text": "conclusions I'm curious as we've now conclusions I'm curious as we've now conclusions I'm curious as we've now" }, { "speaker": "guest", "text": "kind of covered at a high level how kind of covered at a high level how kind of covered at a high level how" }, { "speaker": "guest", "text": "we're thinking about these these we're thinking about these these we're thinking about these these" }, { "speaker": "host", "text": "workflows and agents and talking about workflows and agents and talking about workflows and agents and talking about" }, { "speaker": "host", "text": "the blog post the blog post the blog post" }, { "speaker": "host", "text": "I want to dive even further behind the I want to dive even further behind the I want to dive even further behind the" }, { "speaker": "guest", "text": "scenes were there any funny stories scenes were there any funny stories scenes were there any funny stories" }, { "speaker": "guest", "text": "Barry of like wild things that you saw Barry of like wild things that you saw Barry of like wild things that you saw" }, { "speaker": "guest", "text": "from customers that were interesting or from customers that were interesting or from customers that were interesting or" }, { "speaker": "host", "text": "are just like kind of far out there in are just like kind of far out there in are just like kind of far out there in" }, { "speaker": "host", "text": "terms of how people are starting to terms of how people are starting to terms of how people are starting to" }, { "speaker": "host", "text": "actually use these things in production actually use these things in production actually use these things in production" }, { "speaker": "guest", "text": "yeah um this is actually from my own yeah um this is actually from my own yeah um this is actually from my own" }, { "speaker": "guest", "text": "experience like uh buing agents uh I experience like uh buing agents uh I experience like uh buing agents uh I" }, { "speaker": "guest", "text": "when I I joined like about a month when I I joined like about a month when I I joined like about a month" }, { "speaker": "host", "text": "before the uh Sona V2 refresh and one of before the uh Sona V2 refresh and one of before the uh Sona V2 refresh and one of" }, { "speaker": "host", "text": "my onboarding tasks was to like run OS my onboarding tasks was to like run OS my onboarding tasks was to like run OS" }, { "speaker": "host", "text": "World which was a computer us Benchmark World which was a computer us Benchmark World which was a computer us Benchmark" }, { "speaker": "guest", "text": "and uh for like a whole week me and this and uh for like a whole week me and this and uh for like a whole week me and this" }, { "speaker": "guest", "text": "other engineer were just staring at this other engineer were just staring at this other engineer were just staring at this" }, { "speaker": "guest", "text": "like agent trajectories that were like like agent trajectories that were like like agent trajectories that were like" }, { "speaker": "host", "text": "counterintuitive to us uh and then well counterintuitive to us uh and then well counterintuitive to us uh and then well" }, { "speaker": "host", "text": "like you know we we weren't sure like like you know we we weren't sure like like you know we we weren't sure like" }, { "speaker": "host", "text": "why the model was making decision you why the model was making decision you why the model was making decision you" }, { "speaker": "guest", "text": "know it was uh given the instructions know it was uh given the instructions know it was uh given the instructions" }, { "speaker": "guest", "text": "that we give it um so we decided like that we give it um so we decided like that we give it um so we decided like" }, { "speaker": "guest", "text": "we're going to act like cloud and you we're going to act like cloud and you we're going to act like cloud and you" }, { "speaker": "host", "text": "know put ourselves in that environment know put ourselves in that environment know put ourselves in that environment" }, { "speaker": "host", "text": "so we would do this really silly thing so we would do this really silly thing so we would do this really silly thing" }, { "speaker": "host", "text": "where we close our eyes for like a whole where we close our eyes for like a whole where we close our eyes for like a whole" }, { "speaker": "guest", "text": "minute and then we like blink at a minute and then we like blink at a minute and then we like blink at a" }, { "speaker": "guest", "text": "screen for a second and we close our screen for a second and we close our screen for a second and we close our" }, { "speaker": "guest", "text": "eyes again just think like well I have eyes again just think like well I have eyes again just think like well I have" }, { "speaker": "host", "text": "to write python code to operate in this to write python code to operate in this to write python code to operate in this" }, { "speaker": "host", "text": "environment what would I do and suddenly environment what would I do and suddenly environment what would I do and suddenly" }, { "speaker": "host", "text": "it made a lot more sense and I feel like it made a lot more sense and I feel like it made a lot more sense and I feel like" }, { "speaker": "guest", "text": "a lot of like agent design comes down to a lot of like agent design comes down to a lot of like agent design comes down to" }, { "speaker": "guest", "text": "that it's like there's a lot of context that it's like there's a lot of context that it's like there's a lot of context" }, { "speaker": "guest", "text": "and a lot of knowledge that the model and a lot of knowledge that the model and a lot of knowledge that the model" }, { "speaker": "host", "text": "maybe not like does not have and we have maybe not like does not have and we have maybe not like does not have and we have" }, { "speaker": "host", "text": "to be empathetic to like the model and to be empathetic to like the model and to be empathetic to like the model and" }, { "speaker": "host", "text": "we have to like make a lot of that clear we have to like make a lot of that clear we have to like make a lot of that clear" }, { "speaker": "guest", "text": "in the prompt in the two description and in the prompt in the two description and in the prompt in the two description and" }, { "speaker": "guest", "text": "in the environment I see so a tip here in the environment I see so a tip here in the environment I see so a tip here" }, { "speaker": "guest", "text": "for developers is almost like to act as for developers is almost like to act as for developers is almost like to act as" }, { "speaker": "host", "text": "if you are looking through the lens of if you are looking through the lens of if you are looking through the lens of" }, { "speaker": "host", "text": "the model itself in terms of like okay the model itself in terms of like okay the model itself in terms of like okay" }, { "speaker": "host", "text": "what would be the most applicable what would be the most applicable what would be the most applicable" }, { "speaker": "guest", "text": "instructions here how is the model like instructions here how is the model like instructions here how is the model like" }, { "speaker": "guest", "text": "seeing the world which is very different seeing the world which is very different seeing the world which is very different" }, { "speaker": "guest", "text": "than how like we operate as a human I than how like we operate as a human I than how like we operate as a human I" }, { "speaker": "host", "text": "guess with additional context um Eric guess with additional context um Eric guess with additional context um Eric" }, { "speaker": "host", "text": "I'm curious if you have any other I'm curious if you have any other I'm curious if you have any other" }, { "speaker": "host", "text": "stories that you've seen yeah I think stories that you've seen yeah I think stories that you've seen yeah I think" }, { "speaker": "guest", "text": "actually my like uh in a very similar actually my like uh in a very similar actually my like uh in a very similar" }, { "speaker": "guest", "text": "vein I think a lot of people really vein I think a lot of people really vein I think a lot of people really" }, { "speaker": "guest", "text": "forget to uh forget to do this and I forget to uh forget to do this and I forget to uh forget to do this and I" }, { "speaker": "host", "text": "think maybe the funniest things I see is think maybe the funniest things I see is think maybe the funniest things I see is" }, { "speaker": "host", "text": "that people will put a lot of effort that people will put a lot of effort that people will put a lot of effort" }, { "speaker": "host", "text": "into creating these really beautiful into creating these really beautiful into creating these really beautiful" }, { "speaker": "guest", "text": "detailed prompts um and then the tools detailed prompts um and then the tools detailed prompts um and then the tools" }, { "speaker": "guest", "text": "that they make to give the model are that they make to give the model are that they make to give the model are" }, { "speaker": "guest", "text": "sort of these incredibly Bare Bones uh sort of these incredibly Bare Bones uh sort of these incredibly Bare Bones uh" }, { "speaker": "host", "text": "like you know no documentation fun like like you know no documentation fun like like you know no documentation fun like" }, { "speaker": "host", "text": "the parameters were named A and B and the parameters were named A and B and the parameters were named A and B and" }, { "speaker": "host", "text": "it's kind of like oh like an an engineer it's kind of like oh like an an engineer it's kind of like oh like an an engineer" }, { "speaker": "guest", "text": "wouldn't be able to like you know work wouldn't be able to like you know work wouldn't be able to like you know work" }, { "speaker": "guest", "text": "with this as a um you know work with with this as a um you know work with with this as a um you know work with" }, { "speaker": "guest", "text": "this as if this was a function they had this as if this was a function they had this as if this was a function they had" }, { "speaker": "host", "text": "to use because there's no documentation to use because there's no documentation to use because there's no documentation" }, { "speaker": "host", "text": "like how can Claude expect how can you like how can Claude expect how can you like how can Claude expect how can you" }, { "speaker": "host", "text": "expect claw to use this as well so kind expect claw to use this as well so kind expect claw to use this as well so kind" }, { "speaker": "guest", "text": "of it's like that lack of like putting of it's like that lack of like putting of it's like that lack of like putting" }, { "speaker": "guest", "text": "yourself in the model shoes and I think yourself in the model shoes and I think yourself in the model shoes and I think" }, { "speaker": "guest", "text": "I think a lot of people when they start I think a lot of people when they start I think a lot of people when they start" }, { "speaker": "host", "text": "trying to use tool use and function trying to use tool use and function trying to use tool use and function" }, { "speaker": "host", "text": "calling um they kind of forget that they calling um they kind of forget that they calling um they kind of forget that they" }, { "speaker": "host", "text": "have to prompt as well and they they have to prompt as well and they they have to prompt as well and they they" }, { "speaker": "guest", "text": "think about the model just as this as a think about the model just as this as a think about the model just as this as a" }, { "speaker": "guest", "text": "you know a more classical programming you know a more classical programming you know a more classical programming" }, { "speaker": "guest", "text": "system but it is still a model and you system but it is still a model and you system but it is still a model and you" }, { "speaker": "host", "text": "need to be prompt Engineering in the need to be prompt Engineering in the need to be prompt Engineering in the" }, { "speaker": "host", "text": "descriptions of your tools themselves descriptions of your tools themselves descriptions of your tools themselves" }, { "speaker": "host", "text": "yeah that's yeah I've noticed that it's yeah that's yeah I've noticed that it's yeah that's yeah I've noticed that it's" }, { "speaker": "guest", "text": "like people forget that it's all part of like people forget that it's all part of like people forget that it's all part of" }, { "speaker": "guest", "text": "the same prompt like it's all getting the same prompt like it's all getting the same prompt like it's all getting" }, { "speaker": "guest", "text": "fed into the same prompt in the context fed into the same prompt in the context fed into the same prompt in the context" }, { "speaker": "host", "text": "window and writing a good tool window and writing a good tool window and writing a good tool" }, { "speaker": "host", "text": "description influences other parts of description influences other parts of description influences other parts of" }, { "speaker": "host", "text": "the prompt as well so that is like one the prompt as well so that is like one the prompt as well so that is like one" }, { "speaker": "guest", "text": "aspect to aspect to aspect to" }, { "speaker": "guest", "text": "consider um agents is this kind of all consider um agents is this kind of all consider um agents is this kind of all" }, { "speaker": "guest", "text": "the hype term right now a lot of lot of the hype term right now a lot of lot of the hype term right now a lot of lot of" }, { "speaker": "host", "text": "people are talking about it and there's people are talking about it and there's people are talking about it and there's" }, { "speaker": "host", "text": "been plenty of Articles written and been plenty of Articles written and been plenty of Articles written and" }, { "speaker": "host", "text": "videos made on the subject what made you videos made on the subject what made you videos made on the subject what made you" }, { "speaker": "guest", "text": "guys think that now is the right time to guys think that now is the right time to guys think that now is the right time to" }, { "speaker": "guest", "text": "write something ourselves and uh talk a write something ourselves and uh talk a write something ourselves and uh talk a" }, { "speaker": "guest", "text": "little bit more about like the details little bit more about like the details little bit more about like the details" }, { "speaker": "host", "text": "Agents sure yeah uh I think one of the Agents sure yeah uh I think one of the Agents sure yeah uh I think one of the" }, { "speaker": "host", "text": "like you know most important things for like you know most important things for like you know most important things for" }, { "speaker": "host", "text": "us is to be able to explain things well us is to be able to explain things well us is to be able to explain things well" }, { "speaker": "guest", "text": "I think that's like a big part of our I think that's like a big part of our I think that's like a big part of our" }, { "speaker": "guest", "text": "like motivation which is like we walk like motivation which is like we walk like motivation which is like we walk" }, { "speaker": "guest", "text": "into customer meetings and everything is into customer meetings and everything is into customer meetings and everything is" }, { "speaker": "host", "text": "referred to as a you know different term referred to as a you know different term referred to as a you know different term" }, { "speaker": "host", "text": "even though they share the same shape so even though they share the same shape so even though they share the same shape so" }, { "speaker": "host", "text": "we thought like you know it would be we thought like you know it would be we thought like you know it would be" }, { "speaker": "guest", "text": "really useful if we can just like have a really useful if we can just like have a really useful if we can just like have a" }, { "speaker": "guest", "text": "set of definition and a set of like set of definition and a set of like set of definition and a set of like" }, { "speaker": "guest", "text": "diagrams and code to explain these these diagrams and code to explain these these diagrams and code to explain these these" }, { "speaker": "host", "text": "things to our customers um and you know things to our customers um and you know things to our customers um and you know" }, { "speaker": "host", "text": "like we are getting to the point where like we are getting to the point where like we are getting to the point where" }, { "speaker": "host", "text": "like the model is capable of doing a lot like the model is capable of doing a lot like the model is capable of doing a lot" }, { "speaker": "guest", "text": "of the like the agentic uh workflows of the like the agentic uh workflows of the like the agentic uh workflows" }, { "speaker": "guest", "text": "that that we're we're seeing and that that that we're we're seeing and that that that we're we're seeing and that" }, { "speaker": "guest", "text": "seems like you know the the right time seems like you know the the right time seems like you know the the right time" }, { "speaker": "host", "text": "for us to you know have some definitions for us to you know have some definitions for us to you know have some definitions" }, { "speaker": "host", "text": "so just to make these conversations so just to make these conversations so just to make these conversations" }, { "speaker": "host", "text": "easier MH yeah I think for me I saw that easier MH yeah I think for me I saw that easier MH yeah I think for me I saw that" }, { "speaker": "guest", "text": "there was a lot of excitement around there was a lot of excitement around there was a lot of excitement around" }, { "speaker": "guest", "text": "agents but also agents but also agents but also" }, { "speaker": "guest", "text": "a lot of people really didn't know what a lot of people really didn't know what a lot of people really didn't know what" }, { "speaker": "host", "text": "it meant in practice and so they were it meant in practice and so they were it meant in practice and so they were" }, { "speaker": "host", "text": "trying to bring agents to sort of any trying to bring agents to sort of any trying to bring agents to sort of any" }, { "speaker": "host", "text": "problem they had even when much simpler problem they had even when much simpler problem they had even when much simpler" }, { "speaker": "guest", "text": "systems would work and so I I saw that systems would work and so I I saw that systems would work and so I I saw that" }, { "speaker": "guest", "text": "as like one of the reasons that we as like one of the reasons that we as like one of the reasons that we" }, { "speaker": "guest", "text": "should write this is like guide people should write this is like guide people should write this is like guide people" }, { "speaker": "host", "text": "about how to do agents but also like about how to do agents but also like about how to do agents but also like" }, { "speaker": "host", "text": "where agents are appropriate uh and that where agents are appropriate uh and that where agents are appropriate uh and that" }, { "speaker": "host", "text": "you shouldn't go after a fly with a you shouldn't go after a fly with a you shouldn't go after a fly with a" }, { "speaker": "guest", "text": "bazooka I see I see that that was a bazooka I see I see that that was a bazooka I see I see that that was a" }, { "speaker": "guest", "text": "perfect part L to my next question here perfect part L to my next question here perfect part L to my next question here" }, { "speaker": "guest", "text": "there's a lot of talk about the there's a lot of talk about the there's a lot of talk about the" }, { "speaker": "host", "text": "potential of agents and every developer potential of agents and every developer potential of agents and every developer" }, { "speaker": "host", "text": "out there and every startup in business out there and every startup in business out there and every startup in business" }, { "speaker": "host", "text": "is trying to think about how they can is trying to think about how they can is trying to think about how they can" }, { "speaker": "guest", "text": "build their own version of an agent for build their own version of an agent for build their own version of an agent for" }, { "speaker": "guest", "text": "their company or product um but you guys their company or product um but you guys their company or product um but you guys" }, { "speaker": "guest", "text": "are starting to see what actually works are starting to see what actually works are starting to see what actually works" }, { "speaker": "host", "text": "in production so we're going to play a in production so we're going to play a in production so we're going to play a" }, { "speaker": "host", "text": "little game here I want to know one little game here I want to know one little game here I want to know one" }, { "speaker": "host", "text": "thing that's overhyped about agents thing that's overhyped about agents thing that's overhyped about agents" }, { "speaker": "guest", "text": "right now and also one thing that's right now and also one thing that's right now and also one thing that's" }, { "speaker": "guest", "text": "underhyped just in terms of underhyped just in terms of underhyped just in terms of" }, { "speaker": "guest", "text": "implementations or actual uh uses in implementations or actual uh uses in implementations or actual uh uses in" }, { "speaker": "host", "text": "production or like potentials here as production or like potentials here as production or like potentials here as" }, { "speaker": "host", "text": "well so Eric let's let's start with you well so Eric let's let's start with you well so Eric let's let's start with you" }, { "speaker": "host", "text": "first I feel like underhyped is like first I feel like underhyped is like first I feel like underhyped is like" }, { "speaker": "guest", "text": "things that save people time even if things that save people time even if things that save people time even if" }, { "speaker": "guest", "text": "it's a very small amount of time it's a very small amount of time it's a very small amount of time" }, { "speaker": "guest", "text": "um I think a lot of times if you just um I think a lot of times if you just um I think a lot of times if you just" }, { "speaker": "host", "text": "look at that on the surface it's like oh look at that on the surface it's like oh look at that on the surface it's like oh" }, { "speaker": "host", "text": "this is something that takes me a minute this is something that takes me a minute this is something that takes me a minute" }, { "speaker": "host", "text": "and even if you can fully automate it and even if you can fully automate it and even if you can fully automate it" }, { "speaker": "guest", "text": "it's only a minute like what what help it's only a minute like what what help it's only a minute like what what help" }, { "speaker": "guest", "text": "is that but really that changes the is that but really that changes the is that but really that changes the" }, { "speaker": "guest", "text": "Dynamics of now you can do that thing a Dynamics of now you can do that thing a Dynamics of now you can do that thing a" }, { "speaker": "host", "text": "hundred times more than you previously hundred times more than you previously hundred times more than you previously" }, { "speaker": "host", "text": "would so I think I'm like most excited would so I think I'm like most excited would so I think I'm like most excited" }, { "speaker": "host", "text": "about things that if they were easier about things that if they were easier about things that if they were easier" }, { "speaker": "guest", "text": "could be really scaled up yeah I I don't could be really scaled up yeah I I don't could be really scaled up yeah I I don't" }, { "speaker": "guest", "text": "know if this is like necessarily related know if this is like necessarily related know if this is like necessarily related" }, { "speaker": "guest", "text": "to Hype but I think it's really to Hype but I think it's really to Hype but I think it's really" }, { "speaker": "host", "text": "difficult to calibrate right now like difficult to calibrate right now like difficult to calibrate right now like" }, { "speaker": "host", "text": "where agents are really needed I think where agents are really needed I think where agents are really needed I think" }, { "speaker": "host", "text": "there's like this intersection that's a there's like this intersection that's a there's like this intersection that's a" }, { "speaker": "guest", "text": "sweet spot for using agent and that's sweet spot for using agent and that's sweet spot for using agent and that's" }, { "speaker": "guest", "text": "like a set of tasks that's valuable and like a set of tasks that's valuable and like a set of tasks that's valuable and" }, { "speaker": "guest", "text": "complex but also like maybe the cost of complex but also like maybe the cost of complex but also like maybe the cost of" }, { "speaker": "host", "text": "error or cost of monitoring error is error or cost of monitoring error is error or cost of monitoring error is" }, { "speaker": "host", "text": "relatively low um that set of tasks is relatively low um that set of tasks is relatively low um that set of tasks is" }, { "speaker": "host", "text": "like not like not not like super clear like not like not not like super clear like not like not not like super clear" }, { "speaker": "guest", "text": "and obvious unless like you know we we and obvious unless like you know we we and obvious unless like you know we we" }, { "speaker": "guest", "text": "actually like look into the like the actually like look into the like the actually like look into the like the" }, { "speaker": "guest", "text": "existing processes I think like coding existing processes I think like coding existing processes I think like coding" }, { "speaker": "host", "text": "and search are like two pretty canonical and search are like two pretty canonical and search are like two pretty canonical" }, { "speaker": "host", "text": "examples where like agents are very examples where like agents are very examples where like agents are very" }, { "speaker": "host", "text": "useful uh like take search as an example useful uh like take search as an example useful uh like take search as an example" }, { "speaker": "guest", "text": "right like you know um it it's it's a right like you know um it it's it's a right like you know um it it's it's a" }, { "speaker": "guest", "text": "really valuable task it's very hard to really valuable task it's very hard to really valuable task it's very hard to" }, { "speaker": "guest", "text": "do like deep iterative search but you do like deep iterative search but you do like deep iterative search but you" }, { "speaker": "host", "text": "can always trade off some like Precision can always trade off some like Precision can always trade off some like Precision" }, { "speaker": "host", "text": "for recall and then just get a little for recall and then just get a little for recall and then just get a little" }, { "speaker": "host", "text": "bit more documents or a little bit more bit more documents or a little bit more bit more documents or a little bit more" }, { "speaker": "guest", "text": "information than than it's needed and information than than it's needed and information than than it's needed and" }, { "speaker": "guest", "text": "filter it down so like we've seen a lot filter it down so like we've seen a lot filter it down so like we've seen a lot" }, { "speaker": "guest", "text": "of success there with agent search what of success there with agent search what of success there with agent search what" }, { "speaker": "host", "text": "does a coding agent look like right now does a coding agent look like right now does a coding agent look like right now" }, { "speaker": "host", "text": "coding agents I think are are super coding agents I think are are super coding agents I think are are super" }, { "speaker": "host", "text": "exciting because they are verifiable at exciting because they are verifiable at exciting because they are verifiable at" }, { "speaker": "guest", "text": "least partially um you know code has least partially um you know code has least partially um you know code has" }, { "speaker": "guest", "text": "this great property that you can write this great property that you can write this great property that you can write" }, { "speaker": "guest", "text": "tests for it and then you edit the code tests for it and then you edit the code tests for it and then you edit the code" }, { "speaker": "host", "text": "and either the tests pass or they don't and either the tests pass or they don't and either the tests pass or they don't" }, { "speaker": "host", "text": "pass uh now that assumes that you have pass uh now that assumes that you have pass uh now that assumes that you have" }, { "speaker": "host", "text": "good unit tests which I think you know good unit tests which I think you know good unit tests which I think you know" }, { "speaker": "guest", "text": "every engineer in the world can say like every engineer in the world can say like every engineer in the world can say like" }, { "speaker": "guest", "text": "we don't yeah um but at least it's we don't yeah um but at least it's we don't yeah um but at least it's" }, { "speaker": "guest", "text": "better than than a lot of things you better than than a lot of things you better than than a lot of things you" }, { "speaker": "host", "text": "know there's no equivalent way to do know there's no equivalent way to do know there's no equivalent way to do" }, { "speaker": "host", "text": "that for many other fields um so this at that for many other fields um so this at that for many other fields um so this at" }, { "speaker": "host", "text": "least gives this gives a coding agent least gives this gives a coding agent least gives this gives a coding agent" }, { "speaker": "guest", "text": "some way that it can get more signal some way that it can get more signal some way that it can get more signal" }, { "speaker": "guest", "text": "every time it goes through a loop so you every time it goes through a loop so you every time it goes through a loop so you" }, { "speaker": "guest", "text": "know if every time it's running the know if every time it's running the know if every time it's running the" }, { "speaker": "host", "text": "tests again it's seeing what the error tests again it's seeing what the error tests again it's seeing what the error" }, { "speaker": "host", "text": "of the output is that makes me think of the output is that makes me think of the output is that makes me think" }, { "speaker": "host", "text": "that you know the model can kind of that you know the model can kind of that you know the model can kind of" }, { "speaker": "guest", "text": "converge on the right answer by getting converge on the right answer by getting converge on the right answer by getting" }, { "speaker": "guest", "text": "this feedback and if you don't have some this feedback and if you don't have some this feedback and if you don't have some" }, { "speaker": "guest", "text": "mechanism to get feedback as you're mechanism to get feedback as you're mechanism to get feedback as you're" }, { "speaker": "host", "text": "iterating you're not injecting any more iterating you're not injecting any more iterating you're not injecting any more" }, { "speaker": "host", "text": "signal you're just going to have noise signal you're just going to have noise signal you're just going to have noise" }, { "speaker": "host", "text": "uh and so there's there's no reason uh and so there's there's no reason uh and so there's there's no reason" }, { "speaker": "guest", "text": "without something like this that an without something like this that an without something like this that an" }, { "speaker": "guest", "text": "agent will converge to the right answer agent will converge to the right answer agent will converge to the right answer" }, { "speaker": "guest", "text": "I see so what's what's the biggest I see so what's what's the biggest I see so what's what's the biggest" }, { "speaker": "host", "text": "blockers then in terms of improving blockers then in terms of improving blockers then in terms of improving" }, { "speaker": "host", "text": "agentic performance on like coding at agentic performance on like coding at agentic performance on like coding at" }, { "speaker": "host", "text": "the moment yeah so I think for for the moment yeah so I think for for the moment yeah so I think for for" }, { "speaker": "guest", "text": "coding um you know we've seen over the coding um you know we've seen over the coding um you know we've seen over the" }, { "speaker": "guest", "text": "last year like on S bench like results last year like on S bench like results last year like on S bench like results" }, { "speaker": "guest", "text": "have gone really from like very very low have gone really from like very very low have gone really from like very very low" }, { "speaker": "host", "text": "to like uh I think you know over 50% now to like uh I think you know over 50% now to like uh I think you know over 50% now" }, { "speaker": "host", "text": "which is really incredible so the models which is really incredible so the models which is really incredible so the models" }, { "speaker": "host", "text": "are getting really good at writing code are getting really good at writing code are getting really good at writing code" }, { "speaker": "guest", "text": "to solve these issues um I feel like I to solve these issues um I feel like I to solve these issues um I feel like I" }, { "speaker": "guest", "text": "have a slightly controversial take here have a slightly controversial take here have a slightly controversial take here" }, { "speaker": "guest", "text": "that I think the next limiting factor is that I think the next limiting factor is that I think the next limiting factor is" }, { "speaker": "host", "text": "going to come back to that verification going to come back to that verification going to come back to that verification" }, { "speaker": "host", "text": "that like it's it's great for these that like it's it's great for these that like it's it's great for these" }, { "speaker": "host", "text": "cases where we do have perfect unit cases where we do have perfect unit cases where we do have perfect unit" }, { "speaker": "guest", "text": "tests and that's starting to work but tests and that's starting to work but tests and that's starting to work but" }, { "speaker": "guest", "text": "for the real world cases we usually for the real world cases we usually for the real world cases we usually" }, { "speaker": "guest", "text": "don't have perfect unit tests for them don't have perfect unit tests for them don't have perfect unit tests for them" }, { "speaker": "host", "text": "and so that's I'm thinking now like and so that's I'm thinking now like and so that's I'm thinking now like" }, { "speaker": "host", "text": "finding ways that we can verify and we finding ways that we can verify and we finding ways that we can verify and we" }, { "speaker": "host", "text": "can add tests for the things that you can add tests for the things that you can add tests for the things that you" }, { "speaker": "guest", "text": "really care about so that the model really care about so that the model really care about so that the model" }, { "speaker": "guest", "text": "itself can test this and know whether itself can test this and know whether itself can test this and know whether" }, { "speaker": "guest", "text": "it's right or wrong before it goes back it's right or wrong before it goes back it's right or wrong before it goes back" }, { "speaker": "host", "text": "to the human I see making sure that we to the human I see making sure that we to the human I see making sure that we" }, { "speaker": "host", "text": "can embed some sort of feedback loop can embed some sort of feedback loop can embed some sort of feedback loop" }, { "speaker": "host", "text": "into the processes it's exactly the into the processes it's exactly the into the processes it's exactly the" }, { "speaker": "guest", "text": "right or right or right or" }, { "speaker": "guest", "text": "wrong okay what's the future of Agents wrong okay what's the future of Agents wrong okay what's the future of Agents" }, { "speaker": "guest", "text": "look like in 2025 Barry we're gonna look like in 2025 Barry we're gonna look like in 2025 Barry we're gonna" }, { "speaker": "host", "text": "start with you yeah uh that I think start with you yeah uh that I think start with you yeah uh that I think" }, { "speaker": "host", "text": "that's a really difficult question um that's a really difficult question um that's a really difficult question um" }, { "speaker": "host", "text": "this is like probably not like a this is like probably not like a this is like probably not like a" }, { "speaker": "guest", "text": "practical thing um but one thing i' I've practical thing um but one thing i' I've practical thing um but one thing i' I've" }, { "speaker": "guest", "text": "been really interested in just like how been really interested in just like how been really interested in just like how" }, { "speaker": "guest", "text": "like a multi-agent environment will look like a multi-agent environment will look like a multi-agent environment will look" }, { "speaker": "host", "text": "like I think I've already shown Eric like I think I've already shown Eric like I think I've already shown Eric" }, { "speaker": "host", "text": "this like I buil an environment where this like I buil an environment where this like I buil an environment where" }, { "speaker": "host", "text": "like a bunch of cloud can like spin up like a bunch of cloud can like spin up like a bunch of cloud can like spin up" }, { "speaker": "guest", "text": "other clouds and play like werewolf other clouds and play like werewolf other clouds and play like werewolf" }, { "speaker": "guest", "text": "together and it's like a completely what together and it's like a completely what together and it's like a completely what" }, { "speaker": "guest", "text": "is werewolf werewolf is a social is werewolf werewolf is a social is werewolf werewolf is a social" }, { "speaker": "host", "text": "deduction game where um all all of the deduction game where um all all of the deduction game where um all all of the" }, { "speaker": "host", "text": "players are trying to figure out what players are trying to figure out what players are trying to figure out what" }, { "speaker": "host", "text": "each other's row is uh it's very similar each other's row is uh it's very similar each other's row is uh it's very similar" }, { "speaker": "guest", "text": "to Mafia it's entirely Tech space which to Mafia it's entirely Tech space which to Mafia it's entirely Tech space which" }, { "speaker": "guest", "text": "is great for for cloud to play in I see is great for for cloud to play in I see is great for for cloud to play in I see" }, { "speaker": "guest", "text": "so we have clouds multiple different so we have clouds multiple different so we have clouds multiple different" }, { "speaker": "host", "text": "clouds playing different roles within clouds playing different roles within clouds playing different roles within" }, { "speaker": "host", "text": "this game all communicating with each this game all communicating with each this game all communicating with each" }, { "speaker": "host", "text": "other yeah exactly and then you see a other yeah exactly and then you see a other yeah exactly and then you see a" }, { "speaker": "guest", "text": "lot of like interesting interaction in lot of like interesting interaction in lot of like interesting interaction in" }, { "speaker": "guest", "text": "there that you just haven't like seen there that you just haven't like seen there that you just haven't like seen" }, { "speaker": "guest", "text": "before and that's something I'm really before and that's something I'm really before and that's something I'm really" }, { "speaker": "host", "text": "excited about as like you know uh very excited about as like you know uh very excited about as like you know uh very" }, { "speaker": "host", "text": "similar to how we went from like single similar to how we went from like single similar to how we went from like single" }, { "speaker": "host", "text": "LM to multi LM I think by the end of the LM to multi LM I think by the end of the LM to multi LM I think by the end of the" }, { "speaker": "guest", "text": "year we can potentially see us going year we can potentially see us going year we can potentially see us going" }, { "speaker": "guest", "text": "from like agent to multi-agent and there from like agent to multi-agent and there from like agent to multi-agent and there" }, { "speaker": "guest", "text": "are some like I think interesting are some like I think interesting are some like I think interesting" }, { "speaker": "host", "text": "research questions to like figure out in research questions to like figure out in research questions to like figure out in" }, { "speaker": "host", "text": "that domain in terms of how the agents that domain in terms of how the agents that domain in terms of how the agents" }, { "speaker": "host", "text": "interact with each other what does like interact with each other what does like interact with each other what does like" }, { "speaker": "guest", "text": "this kind of emergent Behavior look like this kind of emergent Behavior look like this kind of emergent Behavior look like" }, { "speaker": "guest", "text": "in that front as you coordinate between in that front as you coordinate between in that front as you coordinate between" }, { "speaker": "guest", "text": "agents doing different things exactly agents doing different things exactly agents doing different things exactly" }, { "speaker": "host", "text": "and just like whether this is actually and just like whether this is actually and just like whether this is actually" }, { "speaker": "host", "text": "like going to be useful or better than like going to be useful or better than like going to be useful or better than" }, { "speaker": "host", "text": "like a single agent with access to a lot like a single agent with access to a lot like a single agent with access to a lot" }, { "speaker": "guest", "text": "more resources do we see any more resources do we see any more resources do we see any" }, { "speaker": "guest", "text": "multi-agent approaches right now that multi-agent approaches right now that multi-agent approaches right now that" }, { "speaker": "guest", "text": "are actually working in are actually working in are actually working in" }, { "speaker": "host", "text": "production um I I feel like in production um I I feel like in production um I I feel like in" }, { "speaker": "host", "text": "production we we haven't even seen a lot production we we haven't even seen a lot production we we haven't even seen a lot" }, { "speaker": "host", "text": "of like successful single agent okay of like successful single agent okay of like successful single agent okay" }, { "speaker": "guest", "text": "interesting um but like you know this is interesting um but like you know this is interesting um but like you know this is" }, { "speaker": "guest", "text": "kind of like a potential extension of kind of like a potential extension of kind of like a potential extension of" }, { "speaker": "guest", "text": "like successful agents with the um I like successful agents with the um I like successful agents with the um I" }, { "speaker": "host", "text": "guess like uh improved capabilities of guess like uh improved capabilities of guess like uh improved capabilities of" }, { "speaker": "host", "text": "the next couple of generations models MH the next couple of generations models MH the next couple of generations models MH" }, { "speaker": "host", "text": "um yeah so this is not advice that um yeah so this is not advice that um yeah so this is not advice that" }, { "speaker": "guest", "text": "everyone should go explore multiagent everyone should go explore multiagent everyone should go explore multiagent" }, { "speaker": "guest", "text": "environment uh it's just I think like environment uh it's just I think like environment uh it's just I think like" }, { "speaker": "guest", "text": "you know to understand the models behav you know to understand the models behav you know to understand the models behav" }, { "speaker": "host", "text": "like this provides us with a better way like this provides us with a better way like this provides us with a better way" }, { "speaker": "host", "text": "to understand model behaviors I see okay to understand model behaviors I see okay to understand model behaviors I see okay" }, { "speaker": "host", "text": "Eric what's what's the future of agents Eric what's what's the future of agents Eric what's what's the future of agents" }, { "speaker": "guest", "text": "in 2025 I feel like in 2025 we're going in 2025 I feel like in 2025 we're going in 2025 I feel like in 2025 we're going" }, { "speaker": "guest", "text": "to see a lot of business adoption of to see a lot of business adoption of to see a lot of business adoption of" }, { "speaker": "guest", "text": "Agents um starting to automate a lot of Agents um starting to automate a lot of Agents um starting to automate a lot of" }, { "speaker": "host", "text": "repetitive tasks and really like scale repetitive tasks and really like scale repetitive tasks and really like scale" }, { "speaker": "host", "text": "up a lot of things that people wanted to up a lot of things that people wanted to up a lot of things that people wanted to" }, { "speaker": "host", "text": "do more of before but were too expensive do more of before but were too expensive do more of before but were too expensive" }, { "speaker": "guest", "text": "you could now have 10x or 100x how much you could now have 10x or 100x how much you could now have 10x or 100x how much" }, { "speaker": "guest", "text": "you do of these things um I'm imagining you do of these things um I'm imagining you do of these things um I'm imagining" }, { "speaker": "guest", "text": "things like you know every single pull things like you know every single pull things like you know every single pull" }, { "speaker": "host", "text": "request in triggers a coding agent to request in triggers a coding agent to request in triggers a coding agent to" }, { "speaker": "host", "text": "come and update all of your come and update all of your come and update all of your" }, { "speaker": "host", "text": "documentation things like that would be documentation things like that would be documentation things like that would be" }, { "speaker": "guest", "text": "cost prohibitive to do before but once cost prohibitive to do before but once cost prohibitive to do before but once" }, { "speaker": "guest", "text": "you think of Agents is sort of almost you think of Agents is sort of almost you think of Agents is sort of almost" }, { "speaker": "guest", "text": "free you can start doing these you know free you can start doing these you know free you can start doing these you know" }, { "speaker": "host", "text": "adding these bells and whistles adding these bells and whistles adding these bells and whistles" }, { "speaker": "host", "text": "everywhere I think maybe something everywhere I think maybe something everywhere I think maybe something" }, { "speaker": "host", "text": "that's not going to happen yet going that's not going to happen yet going that's not going to happen yet going" }, { "speaker": "guest", "text": "back to like what's overhyped yeah I back to like what's overhyped yeah I back to like what's overhyped yeah I" }, { "speaker": "guest", "text": "feel like agents for consumers are like feel like agents for consumers are like feel like agents for consumers are like" }, { "speaker": "guest", "text": "fairly overhyped right now we Goot um fairly overhyped right now we Goot um fairly overhyped right now we Goot um" }, { "speaker": "host", "text": "because I think that like like we talked because I think that like like we talked because I think that like like we talked" }, { "speaker": "host", "text": "about like uh verifiability I think that about like uh verifiability I think that about like uh verifiability I think that" }, { "speaker": "host", "text": "for a lot of consumer tasks um it's for a lot of consumer tasks um it's for a lot of consumer tasks um it's" }, { "speaker": "guest", "text": "almost as much work to sort of fully almost as much work to sort of fully almost as much work to sort of fully" }, { "speaker": "guest", "text": "specify your preferences and what the specify your preferences and what the specify your preferences and what the" }, { "speaker": "guest", "text": "task is as to just do it yourself and task is as to just do it yourself and task is as to just do it yourself and" }, { "speaker": "host", "text": "it's very expensive to verify so like it's very expensive to verify so like it's very expensive to verify so like" }, { "speaker": "host", "text": "trying to have a agent like fully book a trying to have a agent like fully book a" }, { "speaker": "host", "text": "vacation for you describing exactly what vacation for you describing exactly what vacation for you describing exactly what" }, { "speaker": "guest", "text": "you want your vacation to be and your you want your vacation to be and your you want your vacation to be and your" }, { "speaker": "guest", "text": "preferences is like almost just as hard preferences is like almost just as hard preferences is like almost just as hard" }, { "speaker": "guest", "text": "as just going and booking it yourself as just going and booking it yourself as just going and booking it yourself" }, { "speaker": "host", "text": "interesting and you it's like very high interesting and you it's like very high interesting and you it's like very high" }, { "speaker": "host", "text": "risk you don't want the agent to go like risk you don't want the agent to go like risk you don't want the agent to go like" }, { "speaker": "host", "text": "actually go book a plane flight interes actually go book a plane flight interes actually go book a plane flight interes" }, { "speaker": "guest", "text": "without you first accepting it is there without you first accepting it is there without you first accepting it is there" }, { "speaker": "guest", "text": "a matter of maybe context that we're a matter of maybe context that we're a matter of maybe context that we're" }, { "speaker": "guest", "text": "missing here too from like the models missing here too from like the models missing here too from like the models" }, { "speaker": "host", "text": "being able to infer this information being able to infer this information being able to infer this information" }, { "speaker": "host", "text": "about somebody without having to about somebody without having to about somebody without having to" }, { "speaker": "host", "text": "explicitly go ask and learn the explicitly go ask and learn the explicitly go ask and learn the" }, { "speaker": "guest", "text": "preference over time yeah so I I think preference over time yeah so I I think preference over time yeah so I I think" }, { "speaker": "guest", "text": "that these things will get there but that these things will get there but that these things will get there but" }, { "speaker": "guest", "text": "first you need to build up this context first you need to build up this context first you need to build up this context" }, { "speaker": "host", "text": "so that the model already knows your so that the model already knows your so that the model already knows your" }, { "speaker": "host", "text": "preferences and things and I think that preferences and things and I think that preferences and things and I think that" }, { "speaker": "host", "text": "takes time I see and we'll need some takes time I see and we'll need some takes time I see and we'll need some" }, { "speaker": "guest", "text": "stepping stones to get to bigger tasks stepping stones to get to bigger tasks stepping stones to get to bigger tasks" }, { "speaker": "guest", "text": "like planning a whole vacation I see like planning a whole vacation I see like planning a whole vacation I see" }, { "speaker": "guest", "text": "okay very okay very okay very" }, { "speaker": "host", "text": "interesting last question any any advice interesting last question any any advice interesting last question any any advice" }, { "speaker": "host", "text": "that you give to a developer that's that you give to a developer that's that you give to a developer that's" }, { "speaker": "host", "text": "exploring this right now in terms of exploring this right now in terms of exploring this right now in terms of" }, { "speaker": "guest", "text": "starting to build this or just thinking starting to build this or just thinking starting to build this or just thinking" }, { "speaker": "guest", "text": "about it from a general future proofing about it from a general future proofing about it from a general future proofing" }, { "speaker": "guest", "text": "perspective that you can give I feel perspective that you can give I feel perspective that you can give I feel" }, { "speaker": "host", "text": "like my best advice is like make sure like my best advice is like make sure like my best advice is like make sure" }, { "speaker": "host", "text": "that you have a way to measure your that you have a way to measure your that you have a way to measure your" }, { "speaker": "host", "text": "results um because I've seen a lot of results um because I've seen a lot of results um because I've seen a lot of" }, { "speaker": "guest", "text": "people will go and and sort of build in people will go and and sort of build in people will go and and sort of build in" }, { "speaker": "guest", "text": "a vacuum without any way to get feedback a vacuum without any way to get feedback a vacuum without any way to get feedback" }, { "speaker": "guest", "text": "about whether their building is working about whether their building is working about whether their building is working" }, { "speaker": "host", "text": "or not um and you can end up building a or not um and you can end up building a or not um and you can end up building a" }, { "speaker": "host", "text": "lot sort of without realizing that it's lot sort of without realizing that it's lot sort of without realizing that it's" }, { "speaker": "host", "text": "uh either it's not working or maybe uh either it's not working or maybe uh either it's not working or maybe" }, { "speaker": "guest", "text": "something much simpler have actually something much simpler have actually something much simpler have actually" }, { "speaker": "guest", "text": "done just as good a job yeah I I think done just as good a job yeah I I think done just as good a job yeah I I think" }, { "speaker": "guest", "text": "very similarly like you know starting as very similarly like you know starting as very similarly like you know starting as" }, { "speaker": "host", "text": "simple as possible um and like having simple as possible um and like having simple as possible um and like having" }, { "speaker": "host", "text": "that measurable result as you are you that measurable result as you are you that measurable result as you are you" }, { "speaker": "host", "text": "know like building more complexity into know like building more complexity into know like building more complexity into" }, { "speaker": "guest", "text": "it uh one thing I've been really it uh one thing I've been really it uh one thing I've been really" }, { "speaker": "guest", "text": "impressed by is like there I work with impressed by is like there I work with impressed by is like there I work with" }, { "speaker": "guest", "text": "some really resourceful startups and some really resourceful startups and some really resourceful startups and" }, { "speaker": "host", "text": "they just like they can do everything they just like they can do everything they just like they can do everything" }, { "speaker": "host", "text": "within one LM call and the orchestration within one LM call and the orchestration within one LM call and the orchestration" }, { "speaker": "host", "text": "around like the the code which will like around like the the code which will like around like the the code which will like" }, { "speaker": "guest", "text": "persist even as the model gets better um persist even as the model gets better um persist even as the model gets better um" }, { "speaker": "guest", "text": "is like kind of their their Niche and uh is like kind of their their Niche and uh is like kind of their their Niche and uh" }, { "speaker": "guest", "text": "I I always like get very happy when I I I always like get very happy when I I I always like get very happy when I" }, { "speaker": "host", "text": "see see one of those because I think see see one of those because I think see see one of those because I think" }, { "speaker": "host", "text": "they can reap the benefit of like future they can reap the benefit of like future they can reap the benefit of like future" }, { "speaker": "host", "text": "capability improvements right um and capability improvements right um and capability improvements right um and" }, { "speaker": "guest", "text": "yeah I think like like realistically you yeah I think like like realistically you yeah I think like like realistically you" }, { "speaker": "guest", "text": "know uh we don't know what use case know uh we don't know what use case know uh we don't know what use case" }, { "speaker": "guest", "text": "would be great for for agents and like would be great for for agents and like would be great for for agents and like" }, { "speaker": "host", "text": "the landscape is going to shift but um the landscape is going to shift but um the landscape is going to shift but um" }, { "speaker": "host", "text": "it's probably a good time to like start it's probably a good time to like start it's probably a good time to like start" }, { "speaker": "host", "text": "building up some of that like muscle to building up some of that like muscle to building up some of that like muscle to" }, { "speaker": "guest", "text": "like think in the agent land um just to like think in the agent land um just to like think in the agent land um just to" }, { "speaker": "guest", "text": "understand that capability a little bit understand that capability a little bit understand that capability a little bit" }, { "speaker": "guest", "text": "better yeah I think I want to double better yeah I think I want to double better yeah I think I want to double" }, { "speaker": "host", "text": "click on something you said of like click on something you said of like click on something you said of like" }, { "speaker": "host", "text": "being excited for the models to get being excited for the models to get being excited for the models to get" }, { "speaker": "host", "text": "better I think that if you if you look better I think that if you if you look better I think that if you if you look" }, { "speaker": "guest", "text": "at your startup or your product and at your startup or your product and at your startup or your product and" }, { "speaker": "guest", "text": "think oh man if the models get smarter think oh man if the models get smarter think oh man if the models get smarter" }, { "speaker": "guest", "text": "all of our Mo's going to disappear that all of our Mo's going to disappear that all of our Mo's going to disappear that" }, { "speaker": "host", "text": "means you're building the wrong thing means you're building the wrong thing means you're building the wrong thing" }, { "speaker": "host", "text": "instead you should be building something instead you should be building something instead you should be building something" }, { "speaker": "host", "text": "so that as the models get smarter your so that as the models get smarter your so that as the models get smarter your" }, { "speaker": "guest", "text": "product it's better and better right product it's better and better right product it's better and better right" }, { "speaker": "guest", "text": "that's great advice um Eric Barry thank that's great advice um Eric Barry thank that's great advice um Eric Barry thank" }, { "speaker": "guest", "text": "you guys uh this is building effective you guys uh this is building effective you guys uh this is building effective" }, { "speaker": "host", "text": "agents thank you thanks agents thank you thanks agents thank you thanks" } ]