Humanity Elevated Future Proofing Your Career - Living Software Systems with Generative & Agentic AI

Episode Date: January 9, 2025

The podcast examine the burgeoning field of AI agents, particularly those powered by Large Language Models (LLMs). One paper critiques current agent benchmarks, highlighting issues with reproducibility, standardization, and cost considerations, advocating for improved evaluation metrics beyond simple accuracy. Another paper champions the potential of generative and agentic AI to create "living software systems," arguing that these AI-driven systems can overcome the limitations of traditional, static software by acting as universal translators between human intent and computational actions. A final paper focuses on the design and implementation of vertical AI agents, proposing a standardized architecture with core modules for memory, reasoning, cognitive skills, and tools, and exploring various agent categories and applications across different industries.

Okay, so imagine software that's like a chameleon, you know, constantly adapting to your needs. Yeah. That's the potential of vertical AI agents. Yeah. And it's not, you know, some distant sci-fi fantasy. We've got a whole stack of research right here in front of us that suggests this is happening now. Right. And it's poised to reshape the entire tech landscape as we know it.

Yeah. It's fascinating because we're not just talking about like incremental improvements to existing software or anything like that. We're talking about a fundamental shift in how software is actually designed and used. Yeah. This isn't just an upgrade.

It's a whole new paradigm. One of the papers actually quotes Alan Watts. Okay. And he said, the more permanent something is, the more likeless it becomes. And doesn't that perfectly capture the frustration we all feel with software today? It really does. It's a great point.

Think about it. Most software is stuck in time, right? Yeah. It's a static translation of someone else's idea of how things should work. Right. And we're constantly struggling to adapt our goals to these rigid tools. It's like trying to fit a square peg into a round hole. You hit cancel 10 times and that online form still won't let you do what you need to do.

Our research calls this dead software. Yeah. And it's everywhere. It is. And that's exactly what these researchers are trying to address. Yeah. The idea is to create living software systems that are powered by AI and can adapt to human needs in real time. So instead of us having to contort ourselves to fit the software's limitations, the AI will adapt the software to us.

Precisely. Imagine interfaces that personalize themselves to your workflow. Okay. Systems that understand the context of your requests and workflows that dynamically adjust on the fly. Okay, but how does this actually work? Is this where all that AI magic comes in? Well, one of the key concepts here is generative AI, which is basically AI that can create things like text code images, you name it, even music. In the context of living software, generative AI acts like a universal translator bridging

the gap between human intent and the specific actions a computer needs to take. So it's like having a digital interpreter that can understand what we want. Yes. And then tell the computer how to make it happen. Exactly. Instead of writing complex code or clicking through endless menus, you could simply tell the AI what you want to achieve in natural language. And it would handle all the technical details behind the scenes.

Can you give us an example of how this might work in a real world scenario? Sure. Imagine you need to submit your travel expenses. Right now, it's a tedious process of manually entering every receipt, categorizing it, double checking everything. Don't remind me. I swear I spend more time hunting for receipts than I do on the actual trips. Right.

But with a vertical AI agent, you could simply tell it, submit my travel expenses for that conference last week, and it would handle everything. It could scan your emails and phone for receipts, categorize them based on your company's policies, even convert currencies if you traveled internationally. Wait, so no more digging through my wallet for faded receipts. Exactly. Now that's a future I can get behind.

But what happens if my company changes its expense policy? Or if there's a new expense category, like say those electric scooters everyone's using these days, do I have to wait for the IT department to update the software? That's the beauty of these living software systems. You simply inform the AI about the policy change or the new category, and it adapts in real time. We're talking about going from software updates that take months or years to updates that happen in minutes. Okay, that's insanely cool. It sounds like we're

talking about more than just some minor tweaks to existing software. Yeah. This feels like a major paradigm shift in how we interact with technology. You're absolutely right. This is about moving beyond static applications to dynamic, adaptable systems that can learn and evolve alongside our needs. We keep using the term vertical AI agents. Yeah. And that vertical part seems important. Is it just a fancy buzzword? Right.

Or is there something more? No, it's definitely more than just jargon. Vertical AI refers to the fact that these AI agents are purpose-built for specific industries and tasks. They're not generic chatbots trying to do a bit of everything. They're more like hyper-focused experts with deep knowledge and skills tailored to a particular domain. So instead of a jack of all trades, it's more like having a team of specialists. Precisely. And the specialization is achieved through what researchers are calling cognitive skills modules. Okay.

These are essentially specialized add-ons that equip these AI agents with capabilities that go beyond general language processing. So these modules give them superpowers in a specific field. Exactly. Can you give us some examples of what these modules might look like in practice? Think of a legal AI agent that uses a specialized module to analyze contracts for loopholes, or a financial AI agent that can assess market risk using complex algorithms. These modules are what allow vertical AI agents

to perform highly specialized tasks that previously required human expertise. It's starting to sound like these vertical AI agents could have a massive impact on the way we work. They could. But the big question on everyone's mind is, will these AI agents replace human jobs? That's the million dollar question. And it's one that's generating a lot of debate. Yeah. But our research suggests that vertical AI is less about replacing jobs and more about augmenting them. Okay. It's about humans and AI working together, forming a collaborative partnership. So instead of humans versus machines, it's more like humans with machines.

Exactly. And this collaboration takes the form of what some researchers are calling human augmented agents. The idea is that the AI handles the routine repetitive tasks, freeing up humans to focus on areas where their expertise and judgment are truly needed. So it's like having the AI take care of the grunt work. Right. While we humans get to focus on the more creative and strategic aspects of our jobs. That's the idea. And this partnership creates a fascinating feedback loop. As the AI does its work, it learns from human input and guidance becoming more effective and sophisticated over time.

It's a continuous learning process that benefits both sides of the equation. It's exciting to think about the potential here. But this shift to a world of living software and AI-powered agents is going to require some serious adjustments. Right. Absolutely. And those adjustments aren't just technological. They're about skills, mindsets, and the way we approach work itself. Yeah. In the next part of our deep dive, we'll explore what this shift means for your job and the skills you'll need to thrive in this new world of work. Welcome back to our exploration of vertical AI agents. Before we kind of jump into the implications for jobs and skills,

let's clarify something we touched on earlier, the distinction between vertical AI and agentic AI. Sure. We kind of glossed over that. Can you break down the difference for us? Sure. So think of vertical AI as the overarching concept. It encompasses AI systems that are designed for specific industries or tasks. These are the hyper-focused experts we talked about, remember, powered by those cognitive skills modules. They go deep in their chosen domains, not wide. So vertical is all about specialization. Got it.

Where does agentic AI fit into this? Agentic AI focuses on the behavior of AI systems. It highlights their ability to act autonomously, to make decisions and take actions on their own to achieve a goal. They're not just passive tools. They can actually initiate and interact with their environments. So every vertical AI agent is also agentic, but not all agentic AI is necessarily specialized. Precisely. You could have a general purpose agentic AI that's great at scheduling appointments or managing your inbox, but it wouldn't be considered vertical because it doesn't have that industry specific focus.

That makes sense. OK, back to the big question. How will this shift to a world of living software impact the future of work? Are we all going to be replaced by robots? Well, the robot uprising isn't on the agenda for today, but the truth is the impact on jobs is complex. Some experts are predicting, you know, massive displacement, while others see it as a catalyst for new jobs and transformations. Our sources lean towards the latter, but with important caveats.

So it's not a simple jobs loss versus jobs gain situation. Not at all. It's more about how existing jobs will evolve and what new skills will be needed to thrive in this changing landscape. One of our sources compares it to a train. We're all on this train called the future of work, and it's moving forward whether we like it or not. The question is, are we going to be prepared passengers or the ones left scrambling on the platform? I like that analogy. What can we do to make sure we're in the prepared passenger group?

What specific skills will be in high demand as we hurdle towards this AI-powered future? Well, one of the most crucial skills will be what's called prompt engineering. Now, this isn't just about, you know, knowing how to write a clever prompt for a chatbot. It's about deeply understanding how to communicate effectively with AI systems. So it's more than just typing in a request. What else is involved? It's about articulating your goals, constraints, and desired outcomes in a way that the AI can understand and act upon. You need to be able to anticipate how the AI might interpret your

instructions, guide it towards the right approach, and refine its output based on your feedback. Almost sounds like learning a new language, the language of AI. That's a great way to put it. You need to understand the AI's logic, its capabilities, and its limitations to really get the most out of AI. That's a great way to put it. You need to understand the AI's logic, its capabilities, and its limitations to really get the most out of it. Remember our analogy of the AI intern? It's not enough to just bark orders. You need to coach them, give them clear instructions and feedback so they can learn and grow. Right. So it's about treating the AI as a collaborative partner, not just a mindless tool. But how does this prompt engineering skill actually play out in different jobs? What might it look like for someone in, say, marketing, finance, or healthcare?

Let's take marketing as an example. Imagine you're working on a new campaign. With an AI marketing agent, you could input your target audience, budget, and goals. The AI could then generate various campaign ideas, analyze data to predict their effectiveness, and even help you create targeted content for different platforms. Your role shifts from being the sole creator to more of a strategist, a curator, and a guide for the AI. So the AI takes on a lot of the heavy lifting, freeing up the marketer to focus

on the higher level thinking. That's pretty cool. What about finance? How might these skills impact someone working in that world? In finance, an AI agent could be invaluable for things like risk assessment, fraud detection, and portfolio management. It could analyze vast amounts of data to identify patterns and trends, generate insights into market movements, and even help automate trading strategies based on your parameters. The human role then becomes one of oversight interpretation

and making those crucial decisions that still require human judgment. It sounds like a lot of the repetitive data-heavy tests could be automated, allowing financial professionals to focus on the more strategic aspects of their work. Okay, last one, healthcare. We touched on this before, but I'm curious to hear how prompt engineering might play out in a hospital setting. We talked about AI assisting with diagnosis and treatment planning, but think about the administrative side of health care, which is often bogged down by paperwork and complex processes. An AI agent could handle tasks like scheduling appointments, managing patient records, and even interacting with insurance companies. This would free up doctors and nurses to spend more time with patients, which is where their skills are most valuable. It feels like across all these

fields, the key is figuring out how to leverage AI to enhance our own skills and expertise, not replace them entirely. Exactly. And that's where the concept of human oversight becomes incredibly important. We need to be actively involved in shaping how these AI agents are developed, trained, and deployed. It's not about blindly trusting the technology. It's about understanding its limitations and making sure it's used responsibly and ethically. That makes sense. We've talked about the what and the how of vertical AI, but I'm also curious about the who. Who's actually building these AI agents?

Is it just a matter of writing some code or is there more to it? That's a great question. And it's one we'll explore in the final part of our deep dive. We'll delve into the different approaches to building these agents from using pre-trained models to developing custom solutions tailored to specific needs. We'll also discuss the key considerations for deployment and integration to make these agents work seamlessly in real world settings. Welcome back to the final part of our deep dive into vertical AI agents. We've talked about what they are, how they work, and their potential impact on the future of work. Right. But I'm still curious about the nuts and bolts.

Yeah. Like how are these AI agents actually built? Yeah. Is it something only a team of super coders can do? Right. Or can regular folks like us understand the basics? Well, building a vertical AI agent definitely involves some technical know-how, but it's not as mystical as some might think.

There are established steps and considerations that kind of guide the process, making it more accessible than you might imagine. Okay. So walk us through it. What are the main stages of bringing one of these AI agents to life? So the first and perhaps most crucial step is really defining the problem you're trying to solve with your AI agent. What are the specific tasks you want it to handle? This might seem obvious, but having a crystal clear understanding of the agent's purpose

is absolutely paramount before even writing a single line of code. So it's like crafting a detailed job description for your AI employee. You need to outline their responsibilities, their goals, and the boundaries within which they'll operate. Yeah, that's a great way to put it. The more precise the instructions, the more effective the AI agent will be. Once you've defined the problem and the AI's role, the next step is gathering and preparing the data that AI will use to learn and make decisions.

And this is where our earlier conversations about potential biases in AI become really important, right? Absolutely. If the data itself is flawed, the AI won't be able to make fair and accurate decisions. Precisely. The quality and diversity of the data directly impact the AI agent's performance. You need to carefully select and curate the data, ensuring it's representative of the real world and doesn't perpetuate harmful biases. This might involve collecting data from multiple sources, cleaning and pre-processing it to removing consistencies,

and even using techniques to balance the data set for fairness. It sounds like a lot of work goes into prepping the data before you even start building the AI agent itself. It's a critical stage that shouldn't be underestimated. But once you have a solid data foundation, you can move on to choosing the right approach for building the AI agent. This is where it can get a bit more technical, but the core concepts are still accessible. OK, let's get into the technical stuff. What are the different options here? Sure. So there are several approaches, and the best one really depends on the complexity of the problem you're trying to solve and the resources you have available. One option is to use a pre-trained model as a starting point.

Think of it like buying a car that's already been built and road tested. It's ready to go, but you might want to make a few modifications to suit your specific needs. So you're basically taking an AI that's already learned a lot and fine-tuning it for your specific tasks. Exactly. There are many powerful language models out there that have been trained on massive amounts of data and can perform a wide range of language-based tasks. You can use techniques like transfer learning

to adapt a pre-trained model to your specific domain and tasks. This can be a much faster and more efficient approach compared to building a model from scratch. That makes sense, especially for smaller teams or those with limited resources. But what if you need something more customized, something that goes beyond the capabilities of a pre-trained model? Right. In that case, you might opt for developing a custom AI solution. This is like building a car from the ground up. You have complete control over the design and functionality, but it requires more time, effort, and expertise. So it's a trade-off between speed and customization. Yeah, in a nutshell, yes. Pre-trained models offer speed and ease of implementation, while custom solutions offer greater precision and control. The best

choice depends on the specific needs of the project. Okay, so you've built your AI agent, whether using a pre-trained model or a custom solution. What happens next? How do you actually get it to work in the real world? That's the deployment and integration stage. You need to decide where the AI agent will live. Will it be integrated into an existing software application? Will it be a standalone tool? Will it run on the cloud or on local servers? So it's not just about the AI itself, it's about how it fits into the existing technological

landscape. Precisely. Seamless integration is crucial. You need to ensure that the AI agent can communicate effectively with other systems and access the data it needs. This might involve developing APIs or using specialized integration platforms. This all sounds pretty technical. Are there tools and resources available to help with these challenges? Absolutely. Thankfully, the technology is evolving rapidly, and there are now many tools and frameworks that simplify the deployment and integration process. Cloud platforms, for instance, offer a range of services specifically designed for deploying

and managing AI applications. That's good to hear. It seems like building and deploying AI agents is becoming more accessible, even for those without a deep technical background. Right. But even after deployment, the work doesn't stop, right? You're absolutely right. Ongoing monitoring and maintenance are essential. You need to track the AI agent's performance, gather feedback from users, and make adjustments as needed to ensure it continues to function effectively and ethically. So it's a

continuous process of improvement and refinement. Exactly. It's about ensuring the AI agent remains aligned with its intended purpose, adapts to changing conditions, and continues to provide value over time. This has been an incredible deep dive into the world of vertical AI agents. We've covered a lot of ground, you know, from the basic concepts to the ethical considerations and the technical details of building and deploying these agents. What are your final thoughts as we wrap up our exploration? I think we're at a pivotal moment in the evolution of technology. Vertical AI agents have the potential to fundamentally reshape industries,

automate tasks, enhance productivity, and create new opportunities for technology, vertical AI agents have the potential to fundamentally reshape industries, automate tasks, enhance productivity, and create new opportunities for human ingenuity and creativity. But as with any powerful technology, it's crucial that we approach this transformation thoughtfully and responsibly, ensuring that these tools are used ethically and for the benefit of all. I couldn't agree more. It's an exciting time to be alive, but it also demands that we engage with these technologies critically and thoughtfully. Thanks for joining us on this journey into the world of vertical AI agents. Until next time, stay curious and keep exploring.