Really happy to have you all here today. I see a number of, folks in the room. Can we just confirm you can all hear me okay? Alright.Well, with that, we're gonna get started. Right on time. So welcome everybody to the session called live learning is dead, long live living learning. It's a tongue twister for sure. My name is Heather Zindel, and I'm the CEO of AIMonster. And AIMonster is an ethically sourced data company, and our core platform is an AI center of excellence. And it has really been created out of this frustration, that I've experienced as well as so many of my peers on what it takes to actually learn about AI, especially if you're not technical. And so, this is a passion area of mine. I created a company to solve this particular problem that I have been living every day for the last few years.

And in the process of doing that, I have actually learned quite a bit about the frustrations that myself and other business leaders share. And the basic frustration is that the old ways of learning, these legacy models are, are broken. And if you take a look at the way we learn, you know, the the course mills through LinkedIn Learning and Udemy, and other big learning platforms, you know, they're great, but the feedback is they're useless in a way because they don't actually help us understand how to use AI within our specific role.

And that's really what this session is about. It's how do we learn about AI? Of course, you can go take the expensive, you know, eight week course at Ivy League schools, to learn about AI, but it's one and done. And, you know, it is very expensive. And, again, it doesn't help ground us on how to leverage, you know, AI, in the roles we play. And then AI itself is great. Of course, it's it's a settled matter that it's a capability that's here to stay. But right now, you know, it's an intern at best in so many of the things that we do, and we'll talk a little bit more about that. And then, of course, you know, there is the knowledge that is hoarded by all the consultancies. I came from the consulting world.

I was the CEO of a consulting company for thirteen years. And, you know, knowledge is hoarded behind these very high paywalls, and it's difficult to get access. And so, you know, we we live in this world where it's quite difficult to actually learn about AI. So I guess, you know, I'm here because I'm so passionate about this idea that we want to figure out how to solve a business problem with AI one time and then share that with our peers globally as fast and cost effectively as possible. But we can't just be thinking about how to do that with humans. We have to think about the role AI agents are playing, you know, in this learning, environment. So with the rise of AI agents, and agent workforces sitting alongside of all of us humans, we have to think about what's the problem statement that that we're witnessing here. And, you know, an interesting fact, you know, if you look at where the investment has gone into AI, $250,000,000,000 spent on AI related corporate investment, investment, but less than 10% of that is being focused on people.

And, it's not really surprising. Any of us that have been in the tech world for a long time know that we keep getting thrown tools and are expected to use it, but with not a lot of support to figure out how to do that. And so as I look forward, you know, in the industry and talking to, you know, my peers on a daily basis, the next eighteen months is really gonna be defined by how we land the last mile and solve for the last mile. And what I mean by that is, you know, there is a 5% adoption rate right now of AI. There is less than a 20% ROI. Very few companies even know how to measure the value and demonstrate the business case and ROI for AI. And so it's really gonna be how we solve that last mile and help resistant workers and users actually figure out how to use it to their advantage and get beyond the hype and the doom scrolling about how AI is gonna actually take over our jobs and figure out how to harness it and use it for career advantage.

So we gotta dig a little deeper and understand sort of the reasons AI is failing, and I'm not gonna be able to unpack that all in this, session today. I was just speaking on this topic, at a venture conference last night, And I'm happy to reach out to anybody who would like to get, more information on all of these particular reasons. But the ones we'll focus on today, are in two areas. One is most of these AI tools sit alongside of us as we work. And so it's not as meaningful if we don't have AI that understands the context in which we work or doesn't have that institutional knowledge. And the other thing is we're still wrestling with the talent gap. You know, there's a lot of folks that are experimenting with AI. But on a daily basis and I've been entrenched in AI now for two years where this is all I do day in and day out.

I have days where I feel like I'm an expert in some things, and then I have days when, I feel like a beginner. And it depends on the use case, ultimately, and it depends on the workflow that I'm trying to, automate and use. So we're gonna talk a little bit more about these two areas specifically on what it takes to actually get the talent. And by the way, the talent now is not just the humans. It's how do we get the agents acting in a way that we want them to and get their skill levels up. So it's just a very interesting time to be having this conversation. So let's talk a little bit about the AI evolution.

We've come from AI being content generators to now being told that they are our work companions. And now we are looking at AgenTic AI that's going to be actually autonomous and operating within these workflows, with us overseeing it. And so the future a of AI is really about action and action in the moment as we're doing the work. And I'm gonna use Anthropic as as the example here because I can't speak to all the models. By the way, disclaimer, you know, we we build on top of Anthropic and and use Claude. But if you notice, there was an announcement about a week ago about, this new feature that Claude is has got in its tool set called dreaming.

And it's very interesting because it's a bit of a warning shot across the bow for, you know, traditional learning and development, groups. Agents are beginning to be able to review prior work that they've done, extract patterns, update their memory, and self improve between sessions. This is really quite extraordinary, honestly. And it means that traditional learning and development teams have got to stop acting like content departments and really start acting like capability infrastructure teams. And, you know, many of them will not be able to make that leap. This is almost a complete reimagining of what it means to learn in a corporate environment. And so, you know, what courses, should employees take can't be the question anymore. It's really how do we make sure that people and agents are learning from the right work and in the right context with the right guardrails?

And so we're really seeing a transition from agents being agents that complete tasks to agents that are going to be quality controlling the work. But where does that quality standard come from? And that's kind of the interesting question now. And so as we think about that, we're really moving from this old paradigm of content is being hoarded and stuck in a library that we can go sort through and find later, And it's really all about just in time learning, and that's really about learning in the middle of the task, knowledge being delivered when it's needed.

So the new model, living learning, what does that actually mean? It means that living learning in our definition is continuously updated. It is shaped by the practitioners now, not the instructional designers, not executives, but actually the people that do the work on the front lines. And it's about knowledge being embedded directly into the real work that we do. The disconnect, though, still a little bit. There's an interesting statistic that over 65% of executives believe that AI can complete complex workflows. Less than 10% of the workers that actually do the work believe AI can complete complex workflows. And it's not really surprising. Right? I mean, all the executives I talked to have huge pressure on them to go land AI. They don't have budgets that are increasing.

They don't have headcount that's increasing, and they're being told to figure out how to make AI work. It's the frontline workers that actually have to use these tools, have to figure this out. Another disconnect is if you asked executives in a large company how many applications they have, they would tell you on average 35. If you if you take a look at the number of apps that are actually within these organizations, it's over 600. So it's just not realistic to expect we can just land these tools down on people, and not assist them. But how do you do that now? You know, the future is gonna be about building capability maps, not course catalogs. And, you know, the old way to think about this is here are 40 AI courses.

The new way to think about this is here are the actual AI enabled skills that each role needs to perform, and here's the key, safely and effectively. So the question we're really asking is what must this person do in the workflow? And those are the capabilities that have to get mapped out, not what training module should we assign. So the first thing we have to think about is what is defining what good looks like. And so this is another interesting feature that Anthropic has come out with, and they call it outcomes. And outcomes actually let teams define what they call a rubric so an agent can evaluate its work against a quality bar. And why is that important?

Because that quality bar determines if the agent now needs to go and revise when it misses the mark. It's really, you know, quite unbelievable. And so the question I have is, aren't aren't we as those that are trying to help others learn about AI, isn't our job now then to define what good work looks like, what bad work looks like, what evidence is required, what exactly should be checked, what needs to be escalated, and then what human judgment can't be skipped.

This is really the future, in terms of setting standards for how people learn. You have to define what good looks like. So if agents can dream, if they can remember, if they can self improve, then we have to be thinking about how to care deeply about what they're learning from. And so what that really means is if you're not focusing on that, you might have an agent that's learning a shortcut or a policy that's outdated, or they may be learning from a biased decision pattern or a risky approval habit. And so now if we define the standard, we have to continue to monitor that and make sure that what they're learning from continues to be the right thing. And if it's not, we have to be correcting that both for the agent as well as, you know, for humans. And we're gonna get into some practical examples here in a minute so we don't stay, you know, at the level of this being theory.

But it really means that now we have to get into embedding learning into workflows. And you have to ask, well, how do we do that? I mean, we're partnering with the engineering teams now to figure out how to do that. And, you know, it it means a couple of things. To the extent that we can control our copilots, our autopilots that we are licensing from companies like Anthropic or Microsoft, the answer is yes. We have direct control over that. But the question we have when we evaluate vendors is what are you doing to embed learning into the workflow of the AI tool you're trying to sell me? Because in order for my team to adopt it and for me to go to bat for the budget and for me to actually get the ROI, I need to know that you're doing more than offering me consulting services or channel partners to provide consulting services to actually help my team understand.

I need to know that you have the learning built into the workflow just in time to help my users because them moving out of this into some other AI tool or going to these old legacy ways of learning and taking courses is just old thinking. So we really need to ask you to raise the bar and make sure that you're embedding that in the workflows. The same way we're about to do that, you know, with our anthropic agents, we are making sure that those folders exist, that help those agents self improve. So to do all of this, we have to start with looking at a workflow, picking a workflow, and then deciding what is AI led, what is human led, and what is AI augmented. And why is this? Because we now have two roles, in learning. We have to now actually make sure that the agents are learning the right thing, and we have to make sure that the humans are learning the right thing.

So our job just got incredibly more difficult because we have to be doing that with two times the work. And so that really is the the the challenge and the opportunity as we go forward because the classroom is moving into the workflow. So the practical actions that can be taken is, number one, just pick a workflow, and I'm gonna give you an example. Let's take sales proposal drafting. The capability is not use AI. The capability is use AI to draft a customer proposal using approved sources without inventing ROI claims, offering unauthorized discounts, or bypassing legal security review. That is very specific to embed. And then you have to define the AI boundary. So terminating a customer is a human led decision.

Summarize a meeting is an AI led with review boundary. And then you have to create the learning object. So each workflow is going to have one learning object associated with it. You're going to have the purpose, the approved use, the prohibited use, the required sources, the prompt scaffold, the quality rubric, the escalation triggers, the good example, the bad example, the human approval stuff, that becomes the curriculum. I look at that, and I go, gosh. That is a lot of work. But imagine the effort being put in upfront to get that right and and what kind of efficiency, productivity, and adoption are we going to get off the other side. You know, the fifth step is embed it where the work happens.

So in our example, we're using Claude and Cowork, you know, in our company, and so we are putting these learning objects as project instructions. If you have Microsoft Copilot, you can do the same. You can also put these into Slack and other communication channels. If you're using Google, you can put it in a workspace template. So the last step is really about tracking behavior and how do we know we're improving. Everyone that I talk to is really complaining about the lack of dashboards and ways to measure adoption, but measuring adoption is is also sort of an old SaaS way of looking at success. At the end of the day, we can look at the output quality. We can look at the number of policy violations. We can look at time saved. We can look at the number of human overrides. We can look at user confidence. What does the user think about it in the moment of the workflow?

These are the things that actually matter. It's is the outcome being achieved based on the learning packet that we put in front of the agent and in front of our workers. And so the key message is you have to measure the capability, not attendance. Traditional L and D measures course completion or time spent or what's the score on the quiz. Even Microsoft and these companies are looking at, you know, certification earned. So in closing, what I would say is l and d is evolving, and the old l and d role builds courses and manages big clunky LMS, models. The new l and d role is building capability systems and making sure that learning is embedded, you know, where people learn. With that, thank you so much.

It was a pleasure to be here, and I hope you check us out at aimonster.ai. We are all about helping people learn about AI. Thank you so much.