Markdown source

Transcript: Beyond the Harness: A Journey Towards Adaptative Engineering - Rajiv Chandegra, Annicha Labs

Source Video

Local Cache

Transcript

Hello, I'm Rajiv. I'm a practicing Hello, I'm Rajiv. I'm a practicing medical doctor here in London medical doctor here in London medical doctor here in London and been doing some AI engineering for and been doing some AI engineering for and been doing some AI engineering for the past few years. Particularly the past few years. Particularly the past few years. Particularly interested in real world application and interested in real world application and interested in real world application and really the future which is multi-agent really the future which is multi-agent really the future which is multi-agent multi-human multi-institutional multi-human multi-institutional multi-human multi-institutional collaboration through a company called collaboration through a company called collaboration through a company called Anitcha Labs. Anitcha Labs. Anitcha Labs. And really the premise of my talk today And really the premise of my talk today And really the premise of my talk today is to introduce a new design philosophy is to introduce a new design philosophy is to introduce a new design philosophy for the future of AI engineering which for the future of AI engineering which for the future of AI engineering which is beyond a harness or beyond a fixed is beyond a harness or beyond a fixed is beyond a harness or beyond a fixed static harness and towards adaptive static harness and towards adaptive static harness and towards adaptive engineering. This slide basically summarizes my This slide basically summarizes my entire talk entire talk entire talk and really it's about exploring what the and really it's about exploring what the and really it's about exploring what the current AI engineering paradigm is which current AI engineering paradigm is which current AI engineering paradigm is which is to use fixed harnesses is to use fixed harnesses is to use fixed harnesses to steer agents. So we either use or to steer agents. So we either use or to steer agents. So we either use or build an existing harness like Pi, build an existing harness like Pi, build an existing harness like Pi, Claude, Code, Cursor, CodeX, whatever it Claude, Code, Cursor, CodeX, whatever it Claude, Code, Cursor, CodeX, whatever it is ahead of run time that remains fairly is ahead of run time that remains fairly is ahead of run time that remains fairly consistent throughout the engineering consistent throughout the engineering consistent throughout the engineering process and is largely constrained with process and is largely constrained with process and is largely constrained with fixed roles and tools and sequencing fixed roles and tools and sequencing fixed roles and tools and sequencing which again ahead of the process which again ahead of the process which again ahead of the process beginning of the process we can tweak beginning of the process we can tweak beginning of the process we can tweak and it essentially produces very and it essentially produces very and it essentially produces very reliable and orderly work and it's great reliable and orderly work and it's great reliable and orderly work and it's great for most engineering problems and ideas. for most engineering problems and ideas. for most engineering problems and ideas. And so your role as an AI engineering And so your role as an AI engineering And so your role as an AI engineering right now is to use a harness and steer right now is to use a harness and steer right now is to use a harness and steer agents. agents. agents. But the future's going to be different But the future's going to be different But the future's going to be different in at least two ways. in at least two ways. in at least two ways. Number one, models are going to become Number one, models are going to become Number one, models are going to become so powerful that existing fixed so powerful that existing fixed so powerful that existing fixed harnesses are constantly going to become harnesses are constantly going to become harnesses are constantly going to become outdated. outdated. outdated. And number two is there's going to be a And number two is there's going to be a And number two is there's going to be a huge exposure to the real world. Right huge exposure to the real world. Right huge exposure to the real world. Right now a lot of AI engineering is software now a lot of AI engineering is software now a lot of AI engineering is software based. It's within behind a screen. based. It's within behind a screen. based. It's within behind a screen. But actually, as things get more But actually, as things get more But actually, as things get more powerful, as models get more powerful, powerful, as models get more powerful, powerful, as models get more powerful, we're going to have exposure to the real we're going to have exposure to the real we're going to have exposure to the real world where world where world where I'd argue that this is where the real I'd argue that this is where the real I'd argue that this is where the real challenges are. challenges are. challenges are. And the real world is dynamic and messy. And the real world is dynamic and messy. And the real world is dynamic and messy. You know, it's full of multi-agent, You know, it's full of multi-agent, You know, it's full of multi-agent, multi-human, cross-institutional, multi-human, cross-institutional, multi-human, cross-institutional, and really touching the physical world, and really touching the physical world, and really touching the physical world, which means that a fixed harness is which means that a fixed harness is which means that a fixed harness is quite brittle. quite brittle. quite brittle. And so that's where we enter what I coin And so that's where we enter what I coin And so that's where we enter what I coin as adaptive engineering, where you as adaptive engineering, where you as adaptive engineering, where you actually allow the harness to emerge and actually allow the harness to emerge and actually allow the harness to emerge and adapt mid-engineering adapt mid-engineering adapt mid-engineering to find its most optimal position and to find its most optimal position and to find its most optimal position and structure. structure. structure. And essentially, the engineer's role is And essentially, the engineer's role is And essentially, the engineer's role is going to be to design some of the going to be to design some of the going to be to design some of the constraints, which are kind of the rules constraints, which are kind of the rules constraints, which are kind of the rules of play, of play, of play, and really allowing the harness and really allowing the harness and really allowing the harness to emerge, stabilize, change, and to emerge, stabilize, change, and to emerge, stabilize, change, and eventually dissolve eventually dissolve eventually dissolve as you go through the runtime of as you go through the runtime of as you go through the runtime of engineering. So, let's look firstly at the current So, let's look firstly at the current paradigm for AI engineering. And it's paradigm for AI engineering. And it's paradigm for AI engineering. And it's essentially essentially essentially where a harness where a harness where a harness serves as the primary method to guide an serves as the primary method to guide an serves as the primary method to guide an LLM, which is typically stateless, LLM, which is typically stateless, LLM, which is typically stateless, and make it into something quite useful. and make it into something quite useful. and make it into something quite useful. And there's a couple of aspects of a And there's a couple of aspects of a And there's a couple of aspects of a harness, which I won't go to in detail, harness, which I won't go to in detail, harness, which I won't go to in detail, but essentially things like system but essentially things like system but essentially things like system prompts, which is established by the prompts, which is established by the prompts, which is established by the harness vendor, harness vendor, harness vendor, meaning that users can't actually modify meaning that users can't actually modify meaning that users can't actually modify it. And this essentially it. And this essentially it. And this essentially tells the harness what it can and can't tells the harness what it can and can't tells the harness what it can and can't do. do. do. You've also got the agents.md files or You've also got the agents.md files or You've also got the agents.md files or the Claude.md if you're using Claude the Claude.md if you're using Claude the Claude.md if you're using Claude code, which is loaded into every context code, which is loaded into every context code, which is loaded into every context window at the start of a session, window at the start of a session, window at the start of a session, tool calling, tool calling, tool calling, and essentially then you have the and essentially then you have the and essentially then you have the genesis of agents, which are specialized genesis of agents, which are specialized genesis of agents, which are specialized entities that have been harnessed uh, entities that have been harnessed uh, entities that have been harnessed uh, granting them unique capabilities granting them unique capabilities granting them unique capabilities um, and and allowing them to um, and and allowing them to um, and and allowing them to differentiate from other agents. And differentiate from other agents. And differentiate from other agents. And these capabilities are typically these capabilities are typically these capabilities are typically manifest as specific skills when it manifest as specific skills when it manifest as specific skills when it comes to role typology, their comes to role typology, their comes to role typology, their capabilities, the rules um, and that capabilities, the rules um, and that capabilities, the rules um, and that enables them to deliver targeted outputs enables them to deliver targeted outputs enables them to deliver targeted outputs and specified outputs, whether it be and specified outputs, whether it be and specified outputs, whether it be code code code documentation documentation documentation issues, handoffs and and most recently issues, handoffs and and most recently issues, handoffs and and most recently we've seen the genesis of kind of loops, we've seen the genesis of kind of loops, we've seen the genesis of kind of loops, loop engineering has become a thing um loop engineering has become a thing um loop engineering has become a thing um so that eventually we reach some sort of so that eventually we reach some sort of so that eventually we reach some sort of outcome outcome outcome uh, whether it's a feature, uh, whether it's a feature, uh, whether it's a feature, uh, a problem solved which is eventually uh, a problem solved which is eventually uh, a problem solved which is eventually reviewed by a human. Um, so what you see reviewed by a human. Um, so what you see reviewed by a human. Um, so what you see uh, from this is that before running the uh, from this is that before running the uh, from this is that before running the engineering process engineering process engineering process the entire harness has been the entire harness has been the entire harness has been predetermined or pre-engineered. And predetermined or pre-engineered. And predetermined or pre-engineered. And that's harness engineering. And uh, it's that's harness engineering. And uh, it's that's harness engineering. And uh, it's been remarkable been remarkable been remarkable at getting the most out of the latest at getting the most out of the latest at getting the most out of the latest models. Um and just sort of stepping a bit back, Um and just sort of stepping a bit back, um, defining exactly what a harness is. um, defining exactly what a harness is. um, defining exactly what a harness is. Essentially the model is the engine and Essentially the model is the engine and Essentially the model is the engine and the harness is everything built around the harness is everything built around the harness is everything built around to make that engine useful. Um, there's to make that engine useful. Um, there's to make that engine useful. Um, there's many different types of harnesses. I many different types of harnesses. I many different types of harnesses. I mean, there's hundreds uh, actually, but mean, there's hundreds uh, actually, but mean, there's hundreds uh, actually, but I guess the most common ones are, I guess the most common ones are, I guess the most common ones are, um, you know, CLI coding, um, such as um, you know, CLI coding, um, such as um, you know, CLI coding, um, such as Claude Code, Codex, Pi. You have IDEs Claude Code, Codex, Pi. You have IDEs Claude Code, Codex, Pi. You have IDEs like Cursor uh, multi-agent like Cursor uh, multi-agent like Cursor uh, multi-agent orchestration through LangChain orchestration through LangChain orchestration through LangChain and and things like Hermes as well, and and things like Hermes as well, and and things like Hermes as well, Klein and Goose. And And these are all Klein and Goose. And And these are all Klein and Goose. And And these are all differentiated based on number one, differentiated based on number one, differentiated based on number one, properties and how they're configured. properties and how they're configured. properties and how they're configured. But actually more importantly, But actually more importantly, But actually more importantly, number two, number two, number two, based on some sort of design and based on some sort of design and based on some sort of design and engineering philosophy that they believe engineering philosophy that they believe engineering philosophy that they believe in. And what makes them so great and so in. And what makes them so great and so in. And what makes them so great and so distinct is that they are opinionated to distinct is that they are opinionated to distinct is that they are opinionated to a lesser or greater degree. Um and they a lesser or greater degree. Um and they a lesser or greater degree. Um and they allow for, I guess different use cases. allow for, I guess different use cases. allow for, I guess different use cases. Um and depending on your temperament as Um and depending on your temperament as Um and depending on your temperament as an engineer, you'd pick one over the an engineer, you'd pick one over the an engineer, you'd pick one over the other. But what we do see in all of these is But what we do see in all of these is that everything's predefined. You know, that everything's predefined. You know, that everything's predefined. You know, you can customize them, of course, like you can customize them, of course, like you can customize them, of course, like for example, the Pi agent. It's for example, the Pi agent. It's for example, the Pi agent. It's minimalist and it's maximally minimalist and it's maximally minimalist and it's maximally extensible. Um extensible. Um extensible. Um but everything is predefined. You know, but everything is predefined. You know, but everything is predefined. You know, the roles, fixed outputs, sequencing the roles, fixed outputs, sequencing the roles, fixed outputs, sequencing protocols, how memory is created. Uh and protocols, how memory is created. Uh and protocols, how memory is created. Uh and so I want to be absolutely clear here. I so I want to be absolutely clear here. I so I want to be absolutely clear here. I don't mean that you can't customize don't mean that you can't customize don't mean that you can't customize harnesses. Of course, you can. But the harnesses. Of course, you can. But the harnesses. Of course, you can. But the customization occurs ahead of the customization occurs ahead of the customization occurs ahead of the engineering runtime. Uh and not engineering runtime. Uh and not engineering runtime. Uh and not mid-engineering. And that's largely mid-engineering. And that's largely mid-engineering. And that's largely directed by us as humans. And for directed by us as humans. And for directed by us as humans. And for someone like me who started AI someone like me who started AI someone like me who started AI engineering about two years ago, engineering about two years ago, engineering about two years ago, it's probably the most powerful or it's probably the most powerful or it's probably the most powerful or predictive uh predictable method. predictive uh predictable method. predictive uh predictable method. And it's got really three payoffs. It's And it's got really three payoffs. It's And it's got really three payoffs. It's reliable. reliable. reliable. Um you know, the same input leads to a Um you know, the same input leads to a Um you know, the same input leads to a similar set of outputs. Obviously, similar set of outputs. Obviously, similar set of outputs. Obviously, there's going to be some variation. there's going to be some variation. there's going to be some variation. It's auditable. Um you can inspect It's auditable. Um you can inspect It's auditable. Um you can inspect exactly what changed and when. And exactly what changed and when. And exactly what changed and when. And there's some sort of linear causality there's some sort of linear causality there's some sort of linear causality where where where uh you know, if something breaks, you uh you know, if something breaks, you uh you know, if something breaks, you can actually trace it back to its can actually trace it back to its can actually trace it back to its source. source. source. Um Um Um this is like a factory line though. Like this is like a factory line though. Like this is like a factory line though. Like think Taylorism for AI. Just like a think Taylorism for AI. Just like a think Taylorism for AI. Just like a factory assembly line, every station is factory assembly line, every station is factory assembly line, every station is engineered in advance. Um, each agent engineered in advance. Um, each agent engineered in advance. Um, each agent has one job, a fixed place, a sequence, has one job, a fixed place, a sequence, has one job, a fixed place, a sequence, and maybe a defined handoff to the next and maybe a defined handoff to the next and maybe a defined handoff to the next agent. I mean, there's clearly some incredible I mean, there's clearly some incredible uses for this and, you know, factory uses for this and, you know, factory uses for this and, you know, factory engineering gives you, you know, engineering gives you, you know, engineering gives you, you know, accuracy, speed, reproducibility, accuracy, speed, reproducibility, accuracy, speed, reproducibility, behavior that you can certify, and it behavior that you can certify, and it behavior that you can certify, and it really shines in these kind of closed really shines in these kind of closed really shines in these kind of closed deterministic systems, where you have a deterministic systems, where you have a deterministic systems, where you have a well-defined product or problem, and you well-defined product or problem, and you well-defined product or problem, and you know the kind of features you want to know the kind of features you want to know the kind of features you want to build. You know the sort of solutions build. You know the sort of solutions build. You know the sort of solutions that you need. Um, that you need. Um, that you need. Um, but here's the catch. Um, and and it's a but here's the catch. Um, and and it's a but here's the catch. Um, and and it's a real trade-off that that reliability real trade-off that that reliability real trade-off that that reliability isn't free. Um, you buy it by isn't free. Um, you buy it by isn't free. Um, you buy it by suppressing, um, suppressing, um, suppressing, um, variance that I'd argue novelty variance that I'd argue novelty variance that I'd argue novelty requires. requires. requires. Um, and determinism and emergence pull Um, and determinism and emergence pull Um, and determinism and emergence pull kind of in opposite directions. kind of in opposite directions. kind of in opposite directions. Um, and Um, and Um, and this is where we have to pivot to the this is where we have to pivot to the this is where we have to pivot to the real world, because in real life real world, because in real life real world, because in real life situations, in when when AI eventually, situations, in when when AI eventually, situations, in when when AI eventually, uh, uh, uh, when it becomes a norm for AI to capture when it becomes a norm for AI to capture when it becomes a norm for AI to capture real-world scenarios, where you've got real-world scenarios, where you've got real-world scenarios, where you've got multi-agents, multi-humans, multi-agents, multi-humans, multi-agents, multi-humans, multi-institutions interacting, and also multi-institutions interacting, and also multi-institutions interacting, and also touching the physical world, touching the physical world, touching the physical world, um, um, um, there's a hard ceiling on novelty, um, there's a hard ceiling on novelty, um, there's a hard ceiling on novelty, um, because the real world is messy, and it because the real world is messy, and it because the real world is messy, and it keeps changing. And so, fixed harness keeps changing. And so, fixed harness keeps changing. And so, fixed harness starts to break down. Um, so like I starts to break down. Um, so like I starts to break down. Um, so like I mentioned, there's a hard ceiling on mentioned, there's a hard ceiling on mentioned, there's a hard ceiling on novelty, um, novelty, um, novelty, um, because, because, because, uh, essentially, its reliability is uh, essentially, its reliability is uh, essentially, its reliability is predicated on suppressing predicated on suppressing predicated on suppressing variants. Um the models are going to variants. Um the models are going to variants. Um the models are going to keep accelerating. keep accelerating. keep accelerating. You know, you can build a careful You know, you can build a careful You know, you can build a careful harness today or you can use one off the harness today or you can use one off the harness today or you can use one off the shelf. shelf. shelf. But it could be irrelevant in the next But it could be irrelevant in the next But it could be irrelevant in the next month. The model just got so good that month. The model just got so good that month. The model just got so good that it didn't need that kind of scaffolding it didn't need that kind of scaffolding it didn't need that kind of scaffolding anymore. anymore. anymore. Um and that kind of leads to this whole Um and that kind of leads to this whole Um and that kind of leads to this whole dilemma of brutalness where every dilemma of brutalness where every dilemma of brutalness where every situation you didn't anticipated situation you didn't anticipated situation you didn't anticipated you didn't anticipate needs a human to you didn't anticipate needs a human to you didn't anticipate needs a human to patch the harness. patch the harness. patch the harness. The more real world it needs, the more The more real world it needs, the more The more real world it needs, the more rules you bolt on and eventually the rules you bolt on and eventually the rules you bolt on and eventually the harness just becomes ever more harness just becomes ever more harness just becomes ever more complicated than the actual problem that complicated than the actual problem that complicated than the actual problem that you need to solve. you need to solve. you need to solve. And I guess if there's one takeaway here And I guess if there's one takeaway here And I guess if there's one takeaway here is that the factory method is that the factory method is that the factory method is the right answer to a fixed problem is the right answer to a fixed problem is the right answer to a fixed problem and the wrong answer to a moving and the wrong answer to a moving and the wrong answer to a moving problem. And so let's really examine in what the And so let's really examine in what the real world is about and and I'm going to real world is about and and I'm going to real world is about and and I'm going to go down a bit of a philosophical route go down a bit of a philosophical route go down a bit of a philosophical route here. here. here. Um Um as we've said, as we've said, as we've said, you know, agentic systems will you know, agentic systems will you know, agentic systems will eventually collide with the real world eventually collide with the real world eventually collide with the real world and that raises a question. and that raises a question. and that raises a question. What is the real world actually made of? What is the real world actually made of? What is the real world actually made of? Because there's two ways to answer this Because there's two ways to answer this Because there's two ways to answer this actually. So the first paradigm, the actually. So the first paradigm, the actually. So the first paradigm, the first way to see the world is through first way to see the world is through first way to see the world is through what we call a reductionist or what we call a reductionist or what we call a reductionist or analytical view analytical view analytical view which we're all very used to actually. which we're all very used to actually. which we're all very used to actually. It says to understand the whole It says to understand the whole It says to understand the whole a big problem or a system or a big thing a big problem or a system or a big thing a big problem or a system or a big thing like an organization like an organization like an organization we've got to take it apart and study the we've got to take it apart and study the we've got to take it apart and study the individual parts. So essentially the individual parts. So essentially the individual parts. So essentially the world is made up of these individual world is made up of these individual world is made up of these individual parts which are things that are stable parts which are things that are stable parts which are things that are stable and change is just something that and change is just something that and change is just something that happens to those stable things now and happens to those stable things now and happens to those stable things now and then. then. then. The relationships amongst those things The relationships amongst those things The relationships amongst those things are secondary or an afterthought. are secondary or an afterthought. are secondary or an afterthought. And this is really the metaphysics of a And this is really the metaphysics of a And this is really the metaphysics of a factory factory factory and of nearly all software that we and of nearly all software that we and of nearly all software that we build. You know, you make the build. You know, you make the build. You know, you make the components. components. components. Uh Uh Uh you wire them together. you wire them together. you wire them together. And you get a product. And you get a product. And you get a product. The second paradigm, what we call the The second paradigm, what we call the The second paradigm, what we call the systems or relational paradigm, systems or relational paradigm, systems or relational paradigm, um takes the emphasis away from there um takes the emphasis away from there um takes the emphasis away from there being separate things or components being separate things or components being separate things or components and actually puts more what we call and actually puts more what we call and actually puts more what we call ontological ontological ontological emphasis on the relationships amongst emphasis on the relationships amongst emphasis on the relationships amongst those things. So, flips the order. It those things. So, flips the order. It those things. So, flips the order. It says that the real world isn't made up says that the real world isn't made up says that the real world isn't made up of things at all. of things at all. of things at all. It's made up of processes and um It's made up of processes and um It's made up of processes and um relationships which come first. relationships which come first. relationships which come first. So, what we call a stable thing really So, what we call a stable thing really So, what we call a stable thing really is just a slow pattern in an ongoing is just a slow pattern in an ongoing is just a slow pattern in an ongoing flow. Think an organism. It's more like flow. Think an organism. It's more like flow. Think an organism. It's more like a flame than a crystal. A flame looks a flame than a crystal. A flame looks a flame than a crystal. A flame looks like an object, but it isn't. It's a like an object, but it isn't. It's a like an object, but it isn't. It's a pattern held together moment by moment pattern held together moment by moment pattern held together moment by moment by a process. You stop the process and by a process. You stop the process and by a process. You stop the process and the thing doesn't exist. Uh and there's a really Uh and there's a really good payoff in understanding this good payoff in understanding this good payoff in understanding this paradigm. Um and once you see the world paradigm. Um and once you see the world paradigm. Um and once you see the world in this way, you start to notice in this way, you start to notice in this way, you start to notice something remarkable. Where where you something remarkable. Where where you something remarkable. Where where you have simple local rules between, let's have simple local rules between, let's have simple local rules between, let's say, say, say, particles or between people or between particles or between people or between particles or between people or between creatures, creatures, creatures, that local interaction gives rise to a that local interaction gives rise to a that local interaction gives rise to a whole new level whole new level whole new level of order that can't be predicted from of order that can't be predicted from of order that can't be predicted from its constituent parts and that no its constituent parts and that no its constituent parts and that no constituent part designed ahead of time. constituent part designed ahead of time. constituent part designed ahead of time. For example, water. Water has a property For example, water. Water has a property For example, water. Water has a property of being wet. of being wet. of being wet. But it's But it's But it's constituent parts, oxygen and hydrogen, constituent parts, oxygen and hydrogen, constituent parts, oxygen and hydrogen, are not wet. are not wet. are not wet. But their coming together, again putting But their coming together, again putting But their coming together, again putting emphasis on the relationships, emphasis on the relationships, emphasis on the relationships, their coming together is where you had their coming together is where you had their coming together is where you had the emergence of a novel new property, the emergence of a novel new property, the emergence of a novel new property, i.e. wetness. i.e. wetness. i.e. wetness. Another example is a flock of birds. Not Another example is a flock of birds. Not Another example is a flock of birds. Not one bird wakes up every morning one bird wakes up every morning one bird wakes up every morning intending to make a flock. intending to make a flock. intending to make a flock. Each probably just follows three simple Each probably just follows three simple Each probably just follows three simple local rules, which is align with your local rules, which is align with your local rules, which is align with your neighbor, don't crash into them, neighbor, don't crash into them, neighbor, don't crash into them, stay close, and that's it. Local rules, stay close, and that's it. Local rules, stay close, and that's it. Local rules, local interactions, and out of that you local interactions, and out of that you local interactions, and out of that you get a flock, get a flock, get a flock, a new pattern alive and coordinated that a new pattern alive and coordinated that a new pattern alive and coordinated that you could never understand by pulling it you could never understand by pulling it you could never understand by pulling it apart and studying one bird because no apart and studying one bird because no apart and studying one bird because no single bird has a flock in it. The flock single bird has a flock in it. The flock single bird has a flock in it. The flock lives in the relationships, not the lives in the relationships, not the lives in the relationships, not the parts. parts. parts. So this factory method of engineering So this factory method of engineering So this factory method of engineering assumes a world of parts. assumes a world of parts. assumes a world of parts. But the non-machine world, and I'd argue But the non-machine world, and I'd argue But the non-machine world, and I'd argue that AI is sort of on the cusp of that, that AI is sort of on the cusp of that, that AI is sort of on the cusp of that, the non-machine world is a world of the non-machine world is a world of the non-machine world is a world of patterns and moving patterns of patterns and moving patterns of patterns and moving patterns of emergence. And that sort of reflects what sort of And that sort of reflects what sort of problems we face in this moving messy problems we face in this moving messy problems we face in this moving messy world. Um and there's a line from world. Um and there's a line from world. Um and there's a line from Russell Ackoff that I absolutely love. Russell Ackoff that I absolutely love. Russell Ackoff that I absolutely love. He says that managers aren't handed neat He says that managers aren't handed neat He says that managers aren't handed neat separate problems. separate problems. separate problems. They're handed dynamic situations, They're handed dynamic situations, They're handed dynamic situations, tangles of problems that keep changing tangles of problems that keep changing tangles of problems that keep changing and keep bumping to each other. And he and keep bumping to each other. And he and keep bumping to each other. And he And he gave that a name. He called it a And he gave that a name. He called it a And he gave that a name. He called it a mess. A mess is a isn't just a pile of mess. A mess is a isn't just a pile of mess. A mess is a isn't just a pile of separate parts, it's actually a system separate parts, it's actually a system separate parts, it's actually a system of relationships in motion. of relationships in motion. of relationships in motion. Which is exactly why the factory based Which is exactly why the factory based Which is exactly why the factory based on a fixed harness on a fixed harness on a fixed harness way of engineering will struggle when it way of engineering will struggle when it way of engineering will struggle when it collides with real-world live scenarios. collides with real-world live scenarios. collides with real-world live scenarios. Because you can't decompose a mess into Because you can't decompose a mess into Because you can't decompose a mess into tidy boxes and solve each of them tidy boxes and solve each of them tidy boxes and solve each of them because they're not individual parts, because they're not individual parts, because they're not individual parts, they're a moving pattern of they're a moving pattern of they're a moving pattern of relationships. We can then look at problems through We can then look at problems through this this this this distinction between different types this distinction between different types this distinction between different types of problems. So, of problems. So, of problems. So, um engineering is all about solving um engineering is all about solving um engineering is all about solving problems and I'd argue that very simply problems and I'd argue that very simply problems and I'd argue that very simply put there are two kinds of problems. put there are two kinds of problems. put there are two kinds of problems. There are complicated problems There are complicated problems There are complicated problems um um um and there are complex problems. So, a and there are complex problems. So, a and there are complex problems. So, a complicated system or complicated complicated system or complicated complicated system or complicated problem problem problem is like building a jumbo jet or a clock. is like building a jumbo jet or a clock. is like building a jumbo jet or a clock. There are passive parts There are passive parts There are passive parts um but experts can take them apart, um but experts can take them apart, um but experts can take them apart, analyze them, plan, predict, document. analyze them, plan, predict, document. analyze them, plan, predict, document. It's hard, but it's actually knowable It's hard, but it's actually knowable It's hard, but it's actually knowable and it works in a very predictable way. and it works in a very predictable way. and it works in a very predictable way. In contrast, when you have a complex In contrast, when you have a complex In contrast, when you have a complex system or a complex problem like a flock system or a complex problem like a flock system or a complex problem like a flock of birds, a market, a human of birds, a market, a human of birds, a market, a human organization, organization, organization, it's slightly different. It's parts are it's slightly different. It's parts are it's slightly different. It's parts are constantly interacting and adapting to constantly interacting and adapting to constantly interacting and adapting to one another. So, the whole one another. So, the whole one another. So, the whole can't be derived can't be derived can't be derived from analyzing the parts. So, you don't from analyzing the parts. So, you don't from analyzing the parts. So, you don't analyze and plan a complex system. analyze and plan a complex system. analyze and plan a complex system. Instead, you probe and sense and respond Instead, you probe and sense and respond Instead, you probe and sense and respond and and and this is probably one of the most this is probably one of the most this is probably one of the most expensive mistakes we make in the modern expensive mistakes we make in the modern expensive mistakes we make in the modern world of design and engineering, world of design and engineering, world of design and engineering, which is that we treat a complex problem which is that we treat a complex problem which is that we treat a complex problem like a complicated one like a complicated one like a complicated one um and and things fail not due to lack um and and things fail not due to lack um and and things fail not due to lack of execution, but essentially because of execution, but essentially because of execution, but essentially because there's a failure in categorizing the there's a failure in categorizing the there's a failure in categorizing the problem spaces. If we go If we go just one step deeper into what a complex just one step deeper into what a complex just one step deeper into what a complex system is, system is, system is, um essentially, you know, the um essentially, you know, the um essentially, you know, the ingredients of it are first, you need a ingredients of it are first, you need a ingredients of it are first, you need a diverse set of agents or actors if it's diverse set of agents or actors if it's diverse set of agents or actors if it's a human organization. a human organization. a human organization. Um not clones essentially, but different Um not clones essentially, but different Um not clones essentially, but different from one another in some sort of way. from one another in some sort of way. from one another in some sort of way. So, because the diversity is the fuel So, because the diversity is the fuel So, because the diversity is the fuel there. there. there. Um they need to interact locally uh cuz Um they need to interact locally uh cuz Um they need to interact locally uh cuz no agent, no individual actor, no no agent, no individual actor, no no agent, no individual actor, no individual component sees the whole. Um individual component sees the whole. Um individual component sees the whole. Um instead each respond to whoever's next instead each respond to whoever's next instead each respond to whoever's next to it. And this is like the bird to it. And this is like the bird to it. And this is like the bird aligning with the neighbor. aligning with the neighbor. aligning with the neighbor. And they learn. They learn recursively. And they learn. They learn recursively. And they learn. They learn recursively. They adapt, then they adapt to the They adapt, then they adapt to the They adapt, then they adapt to the results of their own adapting, and it results of their own adapting, and it results of their own adapting, and it constantly loops, and constantly loops, and constantly loops, and um um you know, with respect to one another you know, with respect to one another you know, with respect to one another and the environment. and the environment. and the environment. Everything is moving in response to Everything is moving in response to Everything is moving in response to everything else, and nothing holds everything else, and nothing holds everything else, and nothing holds still. And out of all of that local still. And out of all of that local still. And out of all of that local adaptive looping interaction, adaptive looping interaction, adaptive looping interaction, you get something called emergence, you get something called emergence, you get something called emergence, which is a key which is a key which is a key concept in complexity science. concept in complexity science. concept in complexity science. Um basically, it's a new novel pattern Um basically, it's a new novel pattern Um basically, it's a new novel pattern uh or organization or behavior uh or organization or behavior uh or organization or behavior that uh no single part designed on its that uh no single part designed on its that uh no single part designed on its own and can't be explained own and can't be explained own and can't be explained by splitting that whole into its parts by splitting that whole into its parts by splitting that whole into its parts and studying them. Again, think the and studying them. Again, think the and studying them. Again, think the flock of birds. No bird has a flock in flock of birds. No bird has a flock in flock of birds. No bird has a flock in it. Think about the wetness of water. it. Think about the wetness of water. it. Think about the wetness of water. None of the individual oxygen hydrogen None of the individual oxygen hydrogen None of the individual oxygen hydrogen molecules have wetness within them. And molecules have wetness within them. And molecules have wetness within them. And yet, it leads to the emergence of yet, it leads to the emergence of yet, it leads to the emergence of something quite novel. something quite novel. something quite novel. Now, these systems don't fly apart into Now, these systems don't fly apart into Now, these systems don't fly apart into chaos. chaos. chaos. They tend to settle into attractors, They tend to settle into attractors, They tend to settle into attractors, which are essentially stable states. which are essentially stable states. which are essentially stable states. Uh Uh they keep pulling back towards some sort they keep pulling back towards some sort they keep pulling back towards some sort of equilibrium. So, for example, water of equilibrium. So, for example, water of equilibrium. So, for example, water is stable at room temperature. is stable at room temperature. is stable at room temperature. So, you get this sort of beautiful So, you get this sort of beautiful So, you get this sort of beautiful tension. You've got constant change at a tension. You've got constant change at a tension. You've got constant change at a local level, local level, local level, but then you've got recognizable stable but then you've got recognizable stable but then you've got recognizable stable patterns at a whole system level. patterns at a whole system level. patterns at a whole system level. And crucially, nobody's steering here. And crucially, nobody's steering here. And crucially, nobody's steering here. It's self-organizing. And that's the It's self-organizing. And that's the It's self-organizing. And that's the whole mystery. And I'd argue that's the whole mystery. And I'd argue that's the whole mystery. And I'd argue that's the whole opportunity whole opportunity whole opportunity for design and engineering. And this is for design and engineering. And this is for design and engineering. And this is where where really want to introduce this new design really want to introduce this new design really want to introduce this new design and engineering philosophy called and engineering philosophy called and engineering philosophy called adaptive engineering. Before that, let's just take stock a Before that, let's just take stock a bit. So, everything we covered so far bit. So, everything we covered so far bit. So, everything we covered so far lives inside a particular framework, lives inside a particular framework, lives inside a particular framework, which is this kind of fixed or factory which is this kind of fixed or factory which is this kind of fixed or factory harness. Uh the human engineer decides harness. Uh the human engineer decides harness. Uh the human engineer decides it all up front, the sequencing, the it all up front, the sequencing, the it all up front, the sequencing, the roles, capabilities. roles, capabilities. roles, capabilities. That's really the factory model, and it That's really the factory model, and it That's really the factory model, and it works. Um and we've seen that um because works. Um and we've seen that um because works. Um and we've seen that um because problem spaces that are complicated yet problem spaces that are complicated yet problem spaces that are complicated yet predictable predictable predictable is exactly right. And I'd argue for most is exactly right. And I'd argue for most is exactly right. And I'd argue for most engineering problems, that is the engineering problems, that is the engineering problems, that is the perfect method. perfect method. perfect method. But the moment AI becomes exponentially But the moment AI becomes exponentially But the moment AI becomes exponentially better, better, better, and it collides with the real world, and it collides with the real world, and it collides with the real world, which is, as we found, messy and which is, as we found, messy and which is, as we found, messy and complex, not just complicated, complex, not just complicated, complex, not just complicated, we need something genuinely new, a we need something genuinely new, a we need something genuinely new, a different idea of what optimal design different idea of what optimal design different idea of what optimal design and engineering even is. Um and engineering even is. Um and engineering even is. Um And it's obviously it rests on those two And it's obviously it rests on those two And it's obviously it rests on those two assumptions that models have become assumptions that models have become assumptions that models have become become exponentially better. become exponentially better. become exponentially better. Uh and that AI will Uh and that AI will Uh and that AI will uh come outside of a sandbox situation, uh come outside of a sandbox situation, uh come outside of a sandbox situation, and actually um engineering will run in and actually um engineering will run in and actually um engineering will run in real time in contact with the social and real time in contact with the social and real time in contact with the social and physical environment. Um so, essentially physical environment. Um so, essentially physical environment. Um so, essentially what we're saying is that the harness what we're saying is that the harness what we're saying is that the harness must adapt mid-flow. And that's the crux must adapt mid-flow. And that's the crux must adapt mid-flow. And that's the crux of harness engineering. And of harness engineering. And of harness engineering. And I guess if I were to crystallize I guess if I were to crystallize I guess if I were to crystallize um the definition, um um the definition, um um the definition, um adaptive engineering is the discipline adaptive engineering is the discipline adaptive engineering is the discipline of designing constraints of designing constraints of designing constraints to the extent that the harness emerges to the extent that the harness emerges to the extent that the harness emerges on its own, stabilizes, on its own, stabilizes, on its own, stabilizes, and adapts as needed in response to the and adapts as needed in response to the and adapts as needed in response to the changing environment in ways that you changing environment in ways that you changing environment in ways that you could not specify in advance. could not specify in advance. could not specify in advance. Essentially, the harness becomes the Essentially, the harness becomes the Essentially, the harness becomes the ongoing output rather than the input. ongoing output rather than the input. ongoing output rather than the input. And And And you know, what this means in practice, you know, what this means in practice, you know, what this means in practice, uh just to make it a bit more concrete, uh just to make it a bit more concrete, uh just to make it a bit more concrete, is is is given that agents have capabilities um given that agents have capabilities um given that agents have capabilities um to interact with each other, to interact with each other, to interact with each other, organization emerges from those organization emerges from those organization emerges from those interactions in response to the interactions in response to the interactions in response to the environment or a goal, environment or a goal, environment or a goal, such that new levels of order appear, such that new levels of order appear, such that new levels of order appear, ones that you could never have specified ones that you could never have specified ones that you could never have specified up front. up front. up front. Um so, think bird in a flock, think Um so, think bird in a flock, think Um so, think bird in a flock, think agent and harness. So, the agents are agent and harness. So, the agents are agent and harness. So, the agents are actually creating the harness actually creating the harness actually creating the harness in the same way as the birds are in the same way as the birds are in the same way as the birds are relating to one another so that the relating to one another so that the relating to one another so that the flock can emerge. So, you don't build flock can emerge. So, you don't build flock can emerge. So, you don't build the harness anymore. You let the agents the harness anymore. You let the agents the harness anymore. You let the agents form the harness that best fits the form the harness that best fits the form the harness that best fits the environment in that moment, in that environment in that moment, in that environment in that moment, in that context. context. context. >> [snorts] >> [snorts] >> [snorts] >> The harness then becomes a >> The harness then becomes a >> The harness then becomes a self-organizing, self-organizing, self-organizing, constantly evolving multi-agent system. constantly evolving multi-agent system. constantly evolving multi-agent system. And if we're to just simulate what it And if we're to just simulate what it And if we're to just simulate what it may become, because this is when agents may become, because this is when agents may become, because this is when agents become a lot more capable. Um become a lot more capable. Um become a lot more capable. Um essentially, you have agents that are essentially, you have agents that are essentially, you have agents that are isomorphic um but yet undifferentiated. isomorphic um but yet undifferentiated. isomorphic um but yet undifferentiated. Um but fully capable of interacting and Um but fully capable of interacting and Um but fully capable of interacting and um learning and changing. um learning and changing. um learning and changing. Um you let those agents interact. At Um you let those agents interact. At Um you let those agents interact. At first, there's nothing. It's a bit of first, there's nothing. It's a bit of first, there's nothing. It's a bit of few exchanges here and there. few exchanges here and there. few exchanges here and there. But the coupling continues and slowly But the coupling continues and slowly But the coupling continues and slowly builds until you reach a point where builds until you reach a point where builds until you reach a point where roughly one connection on average um roughly one connection on average um roughly one connection on average um occurs um occurs um occurs um with each agent. And suddenly, you start with each agent. And suddenly, you start with each agent. And suddenly, you start to see this whole emerge. to see this whole emerge. to see this whole emerge. >> [snorts] >> [snorts] >> Um a system, you could say. Uh and I >> Um a system, you could say. Uh and I >> Um a system, you could say. Uh and I guess as an engineer, your new lever guess as an engineer, your new lever guess as an engineer, your new lever uh becomes uh becomes uh becomes uh how the rate of coupling. Like, do uh how the rate of coupling. Like, do uh how the rate of coupling. Like, do you dial it up or do you you dial it up or do you you dial it up or do you dampen it down? Um again, the adaptive dampen it down? Um again, the adaptive dampen it down? Um again, the adaptive path focuses on the interactions. path focuses on the interactions. path focuses on the interactions. Um until, you know, agents start to Um until, you know, agents start to Um until, you know, agents start to actually specialize with respect to one actually specialize with respect to one actually specialize with respect to one another. another. another. So, let's say you got two agents doing So, let's say you got two agents doing So, let's say you got two agents doing the same thing in the same place. That's the same thing in the same place. That's the same thing in the same place. That's just redundant and the environment knows just redundant and the environment knows just redundant and the environment knows that. The pressure from the environment that. The pressure from the environment that. The pressure from the environment knows that. So, it rewards anything that knows that. So, it rewards anything that knows that. So, it rewards anything that essentially breaks the tie. Um essentially breaks the tie. Um essentially breaks the tie. Um so so so I guess a tiny difference um I guess a tiny difference um I guess a tiny difference um you know, who got there first, who is you know, who got there first, who is you know, who got there first, who is slightly better slightly better slightly better gets amplified by the feedback until gets amplified by the feedback until gets amplified by the feedback until those two agents are no longer those two agents are no longer those two agents are no longer interchangeable. And so, the pool stops interchangeable. And so, the pool stops interchangeable. And so, the pool stops becoming something uniform and it becoming something uniform and it becoming something uniform and it actually falls into uh niches. And actually falls into uh niches. And actually falls into uh niches. And really, the key insight here is that the really, the key insight here is that the really, the key insight here is that the agent's identity isn't something you agent's identity isn't something you agent's identity isn't something you gave it. gave it. gave it. It's actually the position, role, or It's actually the position, role, or It's actually the position, role, or capability it took relative to the capability it took relative to the capability it took relative to the others and its environment. So, here we others and its environment. So, here we others and its environment. So, here we see specialization see specialization see specialization emerges from the interaction. emerges from the interaction. emerges from the interaction. And if you go on a bit further, uh they And if you go on a bit further, uh they And if you go on a bit further, uh they stop connecting at random. Uh in fact, stop connecting at random. Uh in fact, stop connecting at random. Uh in fact, they start to work well together they start to work well together they start to work well together um, in terms of sort of clusters. Um, um, in terms of sort of clusters. Um, um, in terms of sort of clusters. Um, and you see the first boundaries appear and you see the first boundaries appear and you see the first boundaries appear and crucially the system and crucially the system and crucially the system drew them, you didn't as an engineer drew them, you didn't as an engineer drew them, you didn't as an engineer ahead of time. Um, so these are kind of ahead of time. Um, so these are kind of ahead of time. Um, so these are kind of emergent clusters. emergent clusters. emergent clusters. Um, until a new convention crystallizes, Um, until a new convention crystallizes, Um, until a new convention crystallizes, you know, a division of norms and you know, a division of norms and you know, a division of norms and protocols, and the system produce some protocols, and the system produce some protocols, and the system produce some level of governance without a governor. level of governance without a governor. level of governance without a governor. Conventions emerge spontaneously from Conventions emerge spontaneously from Conventions emerge spontaneously from the local coordination the local coordination the local coordination with no central authority because as with no central authority because as with no central authority because as we've seen, central authority can lead we've seen, central authority can lead we've seen, central authority can lead to brittleness in a changing to brittleness in a changing to brittleness in a changing environment. environment. environment. Um, so there's just enough repeated Um, so there's just enough repeated Um, so there's just enough repeated interaction to tip the group into some interaction to tip the group into some interaction to tip the group into some sort of shared norm. And this is crucial sort of shared norm. And this is crucial sort of shared norm. And this is crucial really because it means that it can really because it means that it can really because it means that it can continue to adapt in a decentralized way continue to adapt in a decentralized way continue to adapt in a decentralized way based on the changing environment or based on the changing environment or based on the changing environment or problem space. And essentially a new um, And essentially a new um, emergent order um, arises from those emergent order um, arises from those emergent order um, arises from those tiny little interactions amongst those tiny little interactions amongst those tiny little interactions amongst those agents, you have a new level of agents, you have a new level of agents, you have a new level of organization that is fairly stable up organization that is fairly stable up organization that is fairly stable up until the next iteration, up until the until the next iteration, up until the until the next iteration, up until the environment changes environment changes environment changes and um, the order needs to essentially and um, the order needs to essentially and um, the order needs to essentially restructure itself. So the question is that what is the role So the question is that what is the role of an engineer? And and really in of an engineer? And and really in of an engineer? And and really in adaptive engineering, uh, you don't adaptive engineering, uh, you don't adaptive engineering, uh, you don't abolish the engineer, you're just abolish the engineer, you're just abolish the engineer, you're just relocating the emphasis of engineering. relocating the emphasis of engineering. relocating the emphasis of engineering. So if you can exploit the model's So if you can exploit the model's So if you can exploit the model's capabilities, um, so agents can capabilities, um, so agents can capabilities, um, so agents can essentially essentially um, interact, learn, and change. What um, interact, learn, and change. What um, interact, learn, and change. What you're doing as an engineer is you're doing as an engineer is you're doing as an engineer is essentially um, essentially um, essentially um, exploring the constraints, you're exploring the constraints, you're exploring the constraints, you're dictating the constraints, the rules of dictating the constraints, the rules of dictating the constraints, the rules of the game. You're not deciding what the game. You're not deciding what the game. You're not deciding what should happen or what you want to should happen or what you want to should happen or what you want to happen, but you're giving the agent happen, but you're giving the agent happen, but you're giving the agent space to explore that field. space to explore that field. space to explore that field. >> [clears throat] >> [clears throat] >> [clears throat] >> And then, once that emergent harness >> And then, once that emergent harness >> And then, once that emergent harness uh uh uh starts to take shape, you are sensing it starts to take shape, you are sensing it starts to take shape, you are sensing it and responding to that rather than um and responding to that rather than um and responding to that rather than um stopping the engineering process and stopping the engineering process and stopping the engineering process and starting from scratch. starting from scratch. starting from scratch. Um Um And I think this slide sort of um And I think this slide sort of um And I think this slide sort of um portrays that. This is not portrays that. This is not portrays that. This is not binary, this is not either or. Uh in binary, this is not either or. Uh in binary, this is not either or. Uh in fact, it's very much a continuum. fact, it's very much a continuum. fact, it's very much a continuum. Um if you you there's no purely fixed Um if you you there's no purely fixed Um if you you there's no purely fixed harness way, and there's no purely harness way, and there's no purely harness way, and there's no purely uh completely adaptive autonomous way. uh completely adaptive autonomous way. uh completely adaptive autonomous way. Um Um fixed engineering is essentially in one fixed engineering is essentially in one fixed engineering is essentially in one hand um prescribing the structure, which hand um prescribing the structure, which hand um prescribing the structure, which agents run and you rarely change the agents run and you rarely change the agents run and you rarely change the structure mid-engineering. structure mid-engineering. structure mid-engineering. Um whereas on the other end of the Um whereas on the other end of the Um whereas on the other end of the scale, you've got adaptive engineering scale, you've got adaptive engineering scale, you've got adaptive engineering where you allow the agents to freely where you allow the agents to freely where you allow the agents to freely interact and allow the harness or the interact and allow the harness or the interact and allow the harness or the structure that guides them structure that guides them structure that guides them uh to be created by them and change uh to be created by them and change uh to be created by them and change mid-engineering. mid-engineering. mid-engineering. Uh so it's not like a swarm of agents Uh so it's not like a swarm of agents Uh so it's not like a swarm of agents loose with no roles and this sort of loose with no roles and this sort of loose with no roles and this sort of intelligence that magically appears. intelligence that magically appears. intelligence that magically appears. Um and it's not sort of better one one Um and it's not sort of better one one Um and it's not sort of better one one way is not better than the other. They way is not better than the other. They way is not better than the other. They just have two different use cases. And just have two different use cases. And just have two different use cases. And um if I could just, you know, bring it um if I could just, you know, bring it um if I could just, you know, bring it back to some examples of existing uh back to some examples of existing uh back to some examples of existing uh harnesses, so for example, Hermes AI um harnesses, so for example, Hermes AI um harnesses, so for example, Hermes AI um on on their website, it states that it's on on their website, it states that it's on on their website, it states that it's a self-improving AI agent that creates a self-improving AI agent that creates a self-improving AI agent that creates skills from experience and learns from skills from experience and learns from skills from experience and learns from it. So, obviously, this is a huge step it. So, obviously, this is a huge step it. So, obviously, this is a huge step in terms of adaptive direction. Um but I in terms of adaptive direction. Um but I in terms of adaptive direction. Um but I do want to make a distinction here um do want to make a distinction here um do want to make a distinction here um because um that adaptation is in the because um that adaptation is in the because um that adaptation is in the form of vertical intelligence, which is form of vertical intelligence, which is form of vertical intelligence, which is about making individual agents smarter. about making individual agents smarter. about making individual agents smarter. What I'm talking about here is What I'm talking about here is What I'm talking about here is horizontal intelligence, which is all horizontal intelligence, which is all horizontal intelligence, which is all about how groups about how groups about how groups of agents coordinate. And so, those two of agents coordinate. And so, those two of agents coordinate. And so, those two directions are a bit orthogonal. Um directions are a bit orthogonal. Um directions are a bit orthogonal. Um And the thesis I'm presenting is that And the thesis I'm presenting is that And the thesis I'm presenting is that it's horizontal intelligence uh that is it's horizontal intelligence uh that is it's horizontal intelligence uh that is the most adaptive method, the most agile the most adaptive method, the most agile the most adaptive method, the most agile method. Um and and and arguably the method. Um and and and arguably the method. Um and and and arguably the higher leverage point. higher leverage point. higher leverage point. Uh compared to vertical. Uh compared to vertical. Uh compared to vertical. Um Um Another example is Pi, um Another example is Pi, um Another example is Pi, um you know, a great harness, uh which is you know, a great harness, uh which is you know, a great harness, uh which is uh uh sort of essentially minimal, um yet it's sort of essentially minimal, um yet it's sort of essentially minimal, um yet it's maximally extensible. Um so, it's fully maximally extensible. Um so, it's fully maximally extensible. Um so, it's fully customizable to an extent. Um so, you customizable to an extent. Um so, you customizable to an extent. Um so, you could class that as adaptive, but again, could class that as adaptive, but again, could class that as adaptive, but again, there's a distinction here. There's there's a distinction here. There's there's a distinction here. There's adaptive at the design stage, like when adaptive at the design stage, like when adaptive at the design stage, like when you're preparing for that engineering you're preparing for that engineering you're preparing for that engineering process. Um process. Um process. Um Uh Uh where the engineer can customize and the where the engineer can customize and the where the engineer can customize and the tool's quite malleable. tool's quite malleable. tool's quite malleable. But, when it comes to adaptive But, when it comes to adaptive But, when it comes to adaptive engineering, that's all about being engineering, that's all about being engineering, that's all about being adaptive during the run time, where the adaptive during the run time, where the adaptive during the run time, where the system reorganizes itself whilst running system reorganizes itself whilst running system reorganizes itself whilst running in response to the changing problem in response to the changing problem in response to the changing problem space of the environment. And you know, space of the environment. And you know, space of the environment. And you know, Pi is a strong example of the first Pi is a strong example of the first Pi is a strong example of the first kind, kind, kind, um but it doesn't make any claims um to um but it doesn't make any claims um to um but it doesn't make any claims um to be anything other. Uh it's not a be anything other. Uh it's not a be anything other. Uh it's not a self-organizing multi-agent system yet. I'm just going to skip over some of the I'm just going to skip over some of the uh nitty-gritties of what that could uh nitty-gritties of what that could uh nitty-gritties of what that could mean. Um so, for example, mean. Um so, for example, mean. Um so, for example, um um uh when it comes to agent capabilities, uh when it comes to agent capabilities, uh when it comes to agent capabilities, um um you know, obviously, they need to you know, obviously, they need to you know, obviously, they need to interact, learn, and change. That's kind interact, learn, and change. That's kind interact, learn, and change. That's kind of a prerequisite of a prerequisite of a prerequisite for adaptive engineering and I don't for adaptive engineering and I don't for adaptive engineering and I don't doubt that that will doubt that that will doubt that that will become the case as models get become the case as models get become the case as models get exponentially more powerful. exponentially more powerful. exponentially more powerful. There's a few things that the designer There's a few things that the designer There's a few things that the designer can actually do in the form of can actually do in the form of can actually do in the form of constraints and I'm not going to list constraints and I'm not going to list constraints and I'm not going to list out all the different types of out all the different types of out all the different types of constraints cuz they're sort of largely constraints cuz they're sort of largely constraints cuz they're sort of largely dictated by some of the properties like dictated by some of the properties like dictated by some of the properties like the roles and the sequencing and the the roles and the sequencing and the the roles and the sequencing and the memory. These are all sort of categories memory. These are all sort of categories memory. These are all sort of categories or properties of harness. or properties of harness. or properties of harness. But essentially there are three But essentially there are three But essentially there are three questions that you could ask yourself as questions that you could ask yourself as questions that you could ask yourself as an engineer which is an engineer which is an engineer which is do you want to do you want to do you want to enable enable enable the agents more or do you want to close the agents more or do you want to close the agents more or do you want to close them in and govern them more them in and govern them more them in and govern them more and produce more and produce more and produce more give them more guardrails. give them more guardrails. give them more guardrails. Are you rewarding them to cohere towards Are you rewarding them to cohere towards Are you rewarding them to cohere towards a particular goal or are you costing a particular goal or are you costing a particular goal or are you costing them them them if they fall outside a particular if they fall outside a particular if they fall outside a particular container? container? container? And how fast or slow do you want this um And how fast or slow do you want this um And how fast or slow do you want this um to happen? Some of the properties, you to happen? Some of the properties, you to happen? Some of the properties, you know, what's the speed for example the know, what's the speed for example the know, what's the speed for example the rate of coupling. So these are some of rate of coupling. So these are some of rate of coupling. So these are some of the engineering tweaks that you can make the engineering tweaks that you can make the engineering tweaks that you can make when it comes to adaptive engineering. when it comes to adaptive engineering. when it comes to adaptive engineering. And then finally like this emergent And then finally like this emergent And then finally like this emergent harness that is constantly shaping and harness that is constantly shaping and harness that is constantly shaping and reshaping itself based on the problem reshaping itself based on the problem reshaping itself based on the problem space. space. space. You can only really sense and respond to You can only really sense and respond to You can only really sense and respond to that emergent harness. You're not going that emergent harness. You're not going that emergent harness. You're not going to to to completely be able to be able to I guess I guess I guess edit edit edit it in some sort of hard engineering it in some sort of hard engineering it in some sort of hard engineering fashion cuz that's not in the best fashion cuz that's not in the best fashion cuz that's not in the best interest of you as an adaptive engineer. This slide is just a more detailed This slide is just a more detailed comparison of both approaches and again comparison of both approaches and again comparison of both approaches and again I want to emphasize that one is not I want to emphasize that one is not I want to emphasize that one is not better than the other. It's just two better than the other. It's just two better than the other. It's just two different design engineering different design engineering different design engineering philosophies. philosophies. philosophies. Where on one hand in the factory model, Where on one hand in the factory model, Where on one hand in the factory model, um you know, the structure is imposed up um you know, the structure is imposed up um you know, the structure is imposed up front by the engineer, and so the front by the engineer, and so the front by the engineer, and so the harness becomes the input. Um whereas in harness becomes the input. Um whereas in harness becomes the input. Um whereas in the adaptive path, the harness emerges the adaptive path, the harness emerges the adaptive path, the harness emerges and is essentially the output. Um and is essentially the output. Um and is essentially the output. Um And And one of the crucial things here And And one of the crucial things here And And one of the crucial things here is that control is decentralized is that control is decentralized is that control is decentralized in the [clears throat] adaptive path. in the [clears throat] adaptive path. in the [clears throat] adaptive path. Um in the field continuously Um in the field continuously Um in the field continuously um because the problem space eventually um because the problem space eventually um because the problem space eventually when AI um comes or AI engineering comes when AI um comes or AI engineering comes when AI um comes or AI engineering comes off a screen into the real world, um the off a screen into the real world, um the off a screen into the real world, um the problem space is constantly changing. Um Now, it's important to understand that, Now, it's important to understand that, you know, there are particular use cases you know, there are particular use cases you know, there are particular use cases for this. Um but there are a huge number for this. Um but there are a huge number for this. Um but there are a huge number of failure modes um in this as well. Um of failure modes um in this as well. Um of failure modes um in this as well. Um you know, just to list a few of those you know, just to list a few of those you know, just to list a few of those failure modes, you know, emergence leans failure modes, you know, emergence leans failure modes, you know, emergence leans towards uh stability, what we call an towards uh stability, what we call an towards uh stability, what we call an attractor, uh where you find a structure attractor, uh where you find a structure attractor, uh where you find a structure that feels stable and optimal, but that that feels stable and optimal, but that that feels stable and optimal, but that doesn't necessarily mean it's the best. doesn't necessarily mean it's the best. doesn't necessarily mean it's the best. Um Um you need some sort of genuine selection you need some sort of genuine selection you need some sort of genuine selection pressure. So, if you imagine in the pressure. So, if you imagine in the pressure. So, if you imagine in the social world or the physical or the social world or the physical or the social world or the physical or the natural world, the environment is um natural world, the environment is um natural world, the environment is um giving us certain uh pressures. You giving us certain uh pressures. You giving us certain uh pressures. You know, that's what encourages us to know, that's what encourages us to know, that's what encourages us to adapt. That is evolution. You know, adapt. That is evolution. You know, adapt. That is evolution. You know, there is some sort of selection pressure there is some sort of selection pressure there is some sort of selection pressure here. here. And you kind of need to find that um And you kind of need to find that um And you kind of need to find that um in or we as a community need to find in or we as a community need to find in or we as a community need to find that in adaptive engineering, otherwise that in adaptive engineering, otherwise that in adaptive engineering, otherwise you just get drift. you just get drift. you just get drift. There is a risk of monoculture, where There is a risk of monoculture, where There is a risk of monoculture, where you don't get genuine diversity amongst you don't get genuine diversity amongst you don't get genuine diversity amongst agents because they're all trained on agents because they're all trained on agents because they're all trained on the same data. the same data. the same data. Um legibility does collapse. Um you Um legibility does collapse. Um you Um legibility does collapse. Um you know, you as adaptability increases, know, you as adaptability increases, know, you as adaptability increases, um um you can't really pin down or explain you can't really pin down or explain you can't really pin down or explain things that well. And again, that's a things that well. And again, that's a things that well. And again, that's a feature of complex systems. You can't feature of complex systems. You can't feature of complex systems. You can't really pin them down. really pin them down. really pin them down. And and and obviously goes without And and and obviously goes without And and and obviously goes without saying that there's no more saying that there's no more saying that there's no more predictability ahead of runtime because predictability ahead of runtime because predictability ahead of runtime because things are constantly moving. things are constantly moving. things are constantly moving. Um Um And and and these are the real failure And and and these are the real failure And and and these are the real failure modes that we must apprehend because I modes that we must apprehend because I modes that we must apprehend because I do feel like adaptive engineering is do feel like adaptive engineering is do feel like adaptive engineering is something that we're going to be more something that we're going to be more something that we're going to be more inclined towards and it's going to race inclined towards and it's going to race inclined towards and it's going to race forward um from the future. forward um from the future. forward um from the future. >> [clears throat and cough] >> I guess a bit of a takeaway then. >> I guess a bit of a takeaway then. Um you know, as models improve Um you know, as models improve Um you know, as models improve exponentially exponentially exponentially um um AI engineering will move into the real AI engineering will move into the real AI engineering will move into the real world scenarios. world scenarios. world scenarios. You know, the real world where it You know, the real world where it You know, the real world where it touches um the physical space touches um the physical space touches um the physical space uh on an ongoing basis. It's uh in uh on an ongoing basis. It's uh in uh on an ongoing basis. It's uh in contact with multiple agents across contact with multiple agents across contact with multiple agents across different institutions. It's in contact different institutions. It's in contact different institutions. It's in contact with multiple human beings. So with multiple human beings. So with multiple human beings. So So there's a real cost peer where AI is So there's a real cost peer where AI is So there's a real cost peer where AI is heading towards. And so the discipline heading towards. And so the discipline heading towards. And so the discipline of design and engineering is going to be of design and engineering is going to be of design and engineering is going to be forced to rethink itself around forced to rethink itself around forced to rethink itself around continual production. continual production. continual production. >> [snorts] >> [snorts] >> And the limiting factor which is the >> And the limiting factor which is the >> And the limiting factor which is the case now, but probably more so in the case now, but probably more so in the case now, but probably more so in the future, is not going to be the strength future, is not going to be the strength future, is not going to be the strength of the model. of the model. of the model. It's going to be the adaptability of the It's going to be the adaptability of the It's going to be the adaptability of the harness. Um and adaptability really here harness. Um and adaptability really here harness. Um and adaptability really here means multi-agent means multi-agent means multi-agent um um which is essentially about coordinating which is essentially about coordinating which is essentially about coordinating amongst agents. A multi-agent amongst agents. A multi-agent amongst agents. A multi-agent orchestration orchestration orchestration which is decentralized. which is decentralized. which is decentralized. Um and it's not Um and it's not Um and it's not adaptable ahead of runtime. adaptable ahead of runtime. adaptable ahead of runtime. In fact, it starts becoming adaptable In fact, it starts becoming adaptable In fact, it starts becoming adaptable mid run-time or during an entire mid run-time or during an entire mid run-time or during an entire engineering a process. And I think engineering a process. And I think engineering a process. And I think that's real intelligence. That's Well, that's real intelligence. That's Well, that's real intelligence. That's Well, when I say real intelligence, I mean when I say real intelligence, I mean when I say real intelligence, I mean that's the intelligence when it comes to that's the intelligence when it comes to that's the intelligence when it comes to adaptability of the harness. Uh, Uh, I'd be really interested in um hearing I'd be really interested in um hearing I'd be really interested in um hearing your thoughts on this um your thoughts on this um your thoughts on this um and whether this is something that uh and whether this is something that uh and whether this is something that uh other people are also exploring. Um other people are also exploring. Um other people are also exploring. Um please do reach out to me. please do reach out to me. please do reach out to me. Thank you.