The Role of AI in Revolutionising Learning And Development: RK’s Interview with Josh Cavalier

CommLab Podcast with Josh Cavalier

Sherna Varayath 2:05
Hey, Learning Champions. Welcome back to the eLearning Champion pod, where we explore the strategies and trends shaping the world of digital learning. Before we dive in, make sure you're a true eLearning champion by hitting that follow button wherever you're listening from. In today's episode, we're going to talk about the role of artificial intelligence in revolutionising learning and development. So buckle up, this is going to be an amazing ride. I'm super thrilled to introduce our speaker for this episode, Josh. Hey, Josh.

Josh Cavalier 2:47
Hey, how you doing?

Sherna Varayath 2:49
Good. Thank you. Josh Cavalier is a pioneering leader and trusted voice in learning and development with over 30 years of experience. As the founder of Josh Cavalier Dot AI, he helps organisations leverage generative AI to transform learning experiences through consulting workshops and courses. Through his popular YouTube channel and weekly live show Brain Power, Josh simplifies complex AI concepts, providing actionable insights for L&D professionals. As a speaker at top conferences, yes, that's him on all the posters like Dev learn and ATD, he inspires audiences with his vision for the future of AI in education. Later this year in November 2025, Josh will release his highly anticipated book Applying AI in Learning and Development, from Platforms to Performance, offering L&D professionals practical strategies to integrate AI and drive exceptional learning outcomes. Welcome again, Josh.

Josh Cavalier 3:56
Great to be here.

Sherna Varayath 3:59
And in conversation with Josh, we have our other speaker doctor, Dr. RK Prasad. He’s the CEO and Co-founder of CommLab India. He nurtured CommLab India from concept to the commercial success that it is today.

RK Prasad 4:07
Hi.

Sherna Varayath 4:21
Additionally, he's responsible for formulating our business strategy and nurturing client relationships. RK is an entrepreneur at heart. He has over 35 years of experience in sales, corporate training, university teaching, and eLearning. He's passionate about technology enabled learning and much more. Welcome, RK. With that, I'd like to let take RK take the lead. RK, the stage is all yours.

RK Prasad 4:49
Thank you. Sherna, thank you very much for that spirited introduction, felt a little energised listening to you. We just ended our day, Josh. Here, it is about 6:00 in the evening. But I would once again like to thank you for so promptly and cheerfully agreeing to speak with us, spend an hour of your time. So to start with, I would like to know a little bit about yourself because I have seen your profile and it's very interesting that you started your career in fine arts. You're a BFA from Fine Arts, then you moved into arts and images and likewise. For the last five years, you have been spending your time in AI. So it's very interesting for us to know how this transition took place for the last 30 years.

Josh Cavalier 5:34
Yeah. It's almost like we need to go out and get an adult beverage at this point to have this conversation, but just to go ahead and summarise, I believe at my core, there's really two main drivers for me. Well, there's three. The 1st is my love of art and science and the balance between those two. The 2nd is helping people, just inherently raising people up, and three, just like yourself, RK, I'm an entrepreneur. And so, I believe those three combined is the essence of my 30 years. To add more clarity to that, I started out as a medical illustrator, I was actually pre-med in art, but in the next two weeks, I'm actually speaking back in my alma mater from the Rochester Institute of Technology for their Imagine Festival. I'm not going to give it away, but one of the situations or one of the things that I experienced way back in the late 80s was, I got my hands on HyperCard. If you remember back then, Macintosh Hypercard was one of the first multimedia applications, black and white, no colour. And it was always fascinating to me the relationship between a human interacting with a computer that always enabled me, and I translated that into working at Handshaw Associates in Charlotte, NC. Dick Handshaw, one of the premier thought leaders around performance consulting and instructional design, thank goodness Dick forced me to learn instructional design performance consulting as an art director, honestly, I was not into it. I was like, why are we doing these surveys? And I just wanted to go ahead and make the thing, and later on it paid huge dividends because it allowed me to get a full picture of what's required. Fast forward, here we are with AI, you need clean data. I'll come back to that. And so, my time with Dick ended, I formed Lodestone, which became one of the top Adobe authorised training providers in the United States. In that capacity as CEO, I would go into learning functions and spin them up on education technology. So if I had to put myself in a box, I'm an education technologist. I've always been working at the intersection of how to use the systems that we have, to increase and improve performance. And so that experience allowed me to get into Fortune 500 companies, military, higher education institutions, and really see how individuals work with the technology. I did a lot of training with Captivate and Storyline and implementing LXP and LMSs. And then, it was time to take a shift. I jumped back into corporate for three years. I work for a $5 billion supply chain company as an individual contributor, which is really odd. But having boots on the ground allowed me to work on high profile projects, one of which was a huge analytics tool that did price point analysis and product placement for products. I worked with frontline sales on the implementation of that, also implementing Teams during COVID. So that was very fascinating. And then in the first week in December of 2022, I got my hands on ChatGPT. And RK, I'm sure you experienced the same thing, I did something novel which was fun.

Then I asked it to write a learning objective and everything changed for me. Because then I asked it to convert the learning objective to a video script and then asked to convert the video script to a storyboard. And in that moment, in my heart of hearts, having gone through the personal computer, the mobile phone, the Internet, it was an inflection point. I knew it and so I went down the rabbit hole because as an education technologist, I was asking myself the question: How does this work?

I knew it wasn't deterministic, something else was happening, and so I spent weeks watching YouTube videos, understanding vector databases and calculus, and all of the nuances of the transformer, and how does this thing work. It blew my mind, probably in the same way that these individuals in the labs themselves, it blew their mind when it started acting like a human. And it was so fascinating to me because again, in my mind, I thought, well, how is this going to transform the industry? How is this going to move the conversation around performance, not just of humans, but of also of machines and the relationship between the two? And I latched on to that and, a lot of it when you when you first start out, it shows up as tactics because that's really all you have. You don't have the systems in place to do strategic moves. Now those systems are coming into place. It's getting really exciting but also really scary. I'm pragmatic when it comes to technology. I'm a realist, but I also look at opportunities, at ways that we can drive the industry and make massive transformations. And so hopefully, that gives you a snapshot of the 30 years, I mean in the last two years since I formed Josh Caviller dot AI, I've been doing webinars and workshops and speaking around the world, doing large implementations of AI in L&D functions and understanding exactly the challenges. It's not easy. There's so many different facets to an AI ops implementation, at the organisational level, and how it appears inside of the L&D function. I've been trying to navigate that, but I can honestly tell you that ,as a professional, this is the most engaged that I have ever been. And so I'm having an immense amount of fun, because one of the cores of the way that I operate is to help people out, and when I began to draw a vector on the ability to go ahead and create hyper personalised learning experiences at scale, you begin to meet people where they are at. RK, when I started, you create the CD-ROM, it went to the training computer in the middle, and that was the only experience everybody had. It was horrible. It was probably, yeah, OK, it was cool. You did multimedia and you clicked next to continue. It took a quiz. OK, got it. But now we're talking about meeting individuals that are on the spectrum of ADHD, being able to address accessibility issues, being able to challenge high performers at a level they’ve never been challenged before. This is fascinating. When it comes to human performance, we are on a golden . And the only way that we're able to do this is through AI, various forms of it, especially generative AI. So again, I can't wait to continue the conversation, so that's my background.

RK Prasad 13:15
Yes. Yeah. Your enthusiasm is so contagious. Actually, I share a lot of commonalities. I also stumbled upon ChatGPT at roughly the same time. And I've been in training and development most of my career, which is about more than 30 years. Earlier, I was doing other stuff like marketing, sales, advertising, and stuff like that. But I would say that majority is in corporate training and university teaching. So I think both of us share the privilege of straddling both the eras, before Internet and mobile technologies and after Internet and mobile technologies. That time in India, we never had good Internet. It took a lot of time for us to get mobile technology. So we could see both, as you were saying, in Rochester Institute of Technology where you studied, everything was so primitive, comparatively. And I totally agree with you that this is the age for L&D professionals, for that matter, any professional. But because we are from that fraternity, I keep telling my people that this is going to go beyond your imagination. However much wild imagining you do, it'll still be wilder. That's what I keep telling, and I don't know whether they believe me. OK, coming back to our conversation, you see most of the L&D folks are either subject matter experts who have come in from sales or quality or engineering, or they are from HR and learning background. So, the computer science part of it, they may not be very, very comfortable, or they may be knowing only pieces of it. So I would like to ask you to explain the fundamental elementary things like:

What is AI? What is machine learning? What is deep learning? How does this work? What is generative AI? Is there a difference or is it built on one another? So a little clarity there.

Josh Cavalier 15:41
Yeah. You know, it's fascinating because we talk about artificial intelligence. It's been around for a long time, the concept of it has since the 1950s and 60s. And I think that the problem was it didn't have all of the elements that it needed to be successful, which is compute, access to clean data, and so on. So the concepts were there, but the technology wasn't. You'd find little glimpses of it, whether it be going in and do medical imaging analysis or on a factory floor with vision technology, kicking things off the line.

But again, all those things are just highly programmed. And so with machine learning, looking at just straight data patterns and writing custom software to identify those data patterns and doing actions against it. But then the concept of the neural network and processing data through a neural network, when we got into deep learning was a radical shift in thinking about processing data through that network and using weights and biases and activation functions and that changed everything. But then, in 2017, when Google DeepMind came out with the transformer and the ability to go ahead and take a corpus of data and send it through a large language model and get a response on the other end, that truly changed everything. And I view it like a tiered structure where you have machine learning as the foundation, deep learning brings the neural network into play, and then generative AI, especially multimodal generative AI is a complete unlock of the whole entire stack of artificial intelligence.

RK Prasad 17:41
Right. I think that's quite enough as far as fundamentals are concerned. So when we look behind, sometime in the 80s and 90s, we had this ERP coming in, enterprise resource planning software, you have all these integrating software which pervaded into the business processes and operations. So do you think that AI will now do the same thing? Will it enter each and every nook and corner of business environment?

Josh Cavalier 18:55
Well, RK, you just teed me up because that's what my book's about. So the subtitle of the book is from Platforms to Performance and I go in to describe how within the whole entire learning ecosystem, the core functionality is going to come from the top down, meaning that your Microsoft Tenant or your Google Workspace essentially is going to go ahead and drive the way that you do work.

Especially when it comes to integrating with other systems like knowledge management, human capital management systems, it's going to be a top down experience with agents talking to each other. So there's going to be various configurations depending upon a business. You may use a domain model to go ahead and do large work, but there's going to be special customised models within your domain if there's an area of knowledge. And then also vendors are actually going to have their own models too depending upon the proficiencies required within their application. So if you're using salesforce.com and using agentforce, there's going to be probably specific models around that, and sales and marketing. If you're using ServiceNow for knowledge management, odds are there's probably going to be a specific agent architecture within ServiceNow. Or if we take a human Capital Management system like Workday, and the specific functions around talent management, there's going to be architectures there. All of these systems will talk to each other. The question is, what type of consolidation is going to happen in the marketplace? That is the unknown. And so what we see, at least from the bottom up in the learning and development industry, is consolidation with LXP LMS vendors. We see consolidation with skill ontology, indexing companies. It's happening. The question is, what is the velocity of the consolidation? How does it show up and will vendors in current state be able to interact with the larger learning ecosystem which at the top is going to be Microsoft and Google?

RK Prasad 21:13
OK, let me take a step back. Can you expand a little about what this agent is, because we keep hearing this word agent, usually in different contexts in different functions. But in this context, what is it?

Josh Cavalier 21:23
Yeah, apologies to all the listeners because I went from just zero to agent. So let's just take a step back. I always like to start with a prompt. Let's go there. Everybody is prompting these days. And a prompt is a way that you are trying to extract out of the model a response that is going to do a high value piece of work. Of course we can use it for entertainment and other purposes. For this conversation, we'll just talk about work or doing a task or a subatomic task. And so there's a maturation level because I can go in and ask the model to go in and write a video. Now, if I just ask it to write a video script, I’m going to get a mediocre response. It's in the nuances of the prompt where I go in and give it additional information, a role, I give it a very specific task, I give it an architecture of that video. Start with a hook, give me initial information that we're going to cover, get into the content, give me a reflection, give me an emotional push off the back end. That's a great educational video architecture. Well, that's going to be in the prompt. And now I have a very cohesive output. But let's take it a step further. Now I want to go ahead. Once I have a great educational video, every single time, I want to put in a workflow. I now want to go ahead and take the output of that model and that script and send it to Synthesia or Colossyan, HeyGen . And now we begin to talk about automations. Now I want to go ahead and vet that whole process out. We could begin with an interview with a subject matter expert. It confirms it and converts it into a video script. That video script then goes to Synthesia or Colossyan or HeyGen, and then as an instructional designer, I review the output. That is an automation.
If all of those things align and everything has vetted, we can begin to talk about an agent, and an agent is 95% software engineering. And so you're going to go in and determine, do I need to go in and hook in with a model? What model is it? Is it Gemini 2.5? Is it GPT, 4 0, 4.5, whatever. That's one. Second is a tool. So the tool would be Synthesia or Colossyan for generating a video. And then, do I need to go ahead and store information in the memory, long term memory, short term memory. But then the agent itself, in this example, is very specific. Agents can do things like orchestration or other higher functions of reasoning. It just so happens that the agent we're talking about here is very task based. And so one of the little nuances of an agent because it can talk to the model, it can be very proactive in completing a task, or can do planning and reasoning and then orchestrate other agents. And as I go in and teach individuals, this whole entire process is that as we move from a prompt to a structured prompt, to a workflow, to an automation, to an agent, the complexity begins to increase. The level of knowledge that you need to monitor and maintain this software begins to increase and also the level of risk begins to increase as we get to an agent. And so the proficiencies that you have to go from an automation to the agent, it's a big jump. But here's a beautiful thing. If you can get it to work, the results are incredible.

RK Prasad 25:25
So let me see if I understood you correctly. An agent is on top of the pyramid.

Josh Cavalier 25:33
No, it's not. It's not yet. We can get into it, but that's just the start.

RK Prasad 25:36
OK. So what I mean to say is that it will coordinate, orchestrate, use, select other AI tools, to give us the output. Is that correct?

Josh Cavalier 25:50
Yep, yeah, it can use other tools. One example I give is Google Calendar, right? Very simple. The request is what events do I have coming up this week? And so the agent's going to ahead and leverage large language model to have that conversation. So if it's going to go ahead and have a tone of voice, it's going to go ahead and respond to me a certain way. The tool is Google Calendar. And so when the API call goes to Google Calendar, it's going to get the event, it's going to bring it back. Now depending upon the transformation of the information, it may put that information in short term memory. Then the information is transformed and then sent back to the chat bot, and I get my response. And so there's more to it than just a prompt and response. The agent itself has many unique facets that make extremely powerful.

RK Prasad 26:48
So do we call ChatGPT an agent?

Josh Cavalier 26:52
I would not. Well, hold on. Was it 2 weeks ago or this week? I don't know. OK. So if we look at a GPT, you can go ahead and do an API call out to an external service
I can go in and use now memory. So it knows who I am and responds according. That's fine, but can I connect multiple GPTs? I cannot. The question is does open AI have this capability? I believe they do. They haven't released it yet, and so it's just a matter of time before open AI releases their agent architecture. And then this conversation we're having would now be available on open AI systems.

RK Prasad 27:41
OK, now let us move to the core of our combined interest. If we take a simple model like the ADDIE model, analysis design, development, implementation, evaluation,

how do you foresee that AI will contribute or control or totally change these various steps in L&D?

Josh Cavalier 28:11
That's huge question. So I always bring it back to the individual. When I got started in the business, I learned ADDIE, I was part of the process, and I knew my role. I knew what I had to do as an art director. I work with the developer, I work with the designer, I work with the project manager. It just so happened that over time, these roles consolidated to some degree, and it got a little messy depending upon the scope of work. Sometimes the roles separate, depending upon what you're doing, but at the end of the day, these are all these tasks that need to occur across ADDIE or whatever workflow you have. So if we look at artificial intelligence, it's really good at doing certain things. It's not so good at the human stuff, right? So empathy and creativity, and just having an intimate conversation as far as like a human to human conversation. And so, if you look at what AI, especially generative AI, is really good at, what I see happening is a complete and utter collapse of creation, meaning that the time from idea to execution will be when you communicate it to the model, that includes learning interactions. Now there is some facets to that. You'll be able to go ahead and communicate to the model and get a very nuanced one off learning interaction. You will have to go ahead and orchestrate with pre-existing applications. Again it's going to transform and change over time, but let's just take Storyline. What will happen there is that you'll be able to go ahead and talk to a model, that model's going to talk to Storyline, and Storyline is going to execute and create the course. It's not there yet, but odds are, it will be. Or you take a version of Captivate, like Captivate for the web, that version of Captivate can become headless, meaning that you don't go into the IDE or the interface of Captivate, and then you'll be able to talk to a model and that model will then talk to Captivate and create the course for you, based upon other bits of information, branding and assets, and what the topic is, but it will pretty much generate the whole course for you. So if we take that in consideration, what does that mean for an instructional designer? How does that show up? And so I have a new role that I created, and I call it the human machine performance analyst.

And what this is that for the decades that we've been doing this RK, we've been worried about human performance. Well, now there's a new player in town, and it's the machine. And so, we now need to be able to go ahead and determine what is the performance of the machine and what is the relationship between the human and the machine, which could be a robot, so just not even on a computer, I know that's forward thinking, but could be a robot. So with that, essentially what's going to happen is this. I call it the centre, if you look at the process of ADDIE, if you go before ADDIE, there's all of the data that you have to collect as a performance consultant to understand what is the truth? Is it truly a training situation or is it an environmental issue in the front line of our factory?
Right. It's not a training issue, it's something else. And then when we take somebody through a learning journey, we have data at the end that we have to go ahead and analyse and say are we moving the needle? And that takes all kinds of other forms in regard to measurement. So what I see though is at the centre of where you go in and you architect a learning experience which includes awareness, the marketing of it, the creation of it. You have the performance support off the back end, you have social interaction between peers. So that could be automated. Essentially what's going to happen is all of those elements I just talked about in the middle are going to collapse down, and the role's going to get pushed to the edges. We will be working higher up in the business as the performance analyst working with the business, understanding the data that's coming in from the front lines.
What is the business problem? How does that translate to the learning solution? Then we have to go ahead and look at the numbers off the back end. Are we moving the needle? So, there is this almost a policy like role that is now created when you begin orchestrating agents talking to each other, and I believe that's the end goal for us as professionals. So we're at early days here, and I didn't mention it earlier, because at this point I'm pretty sure we're blowing everyone's mind. We didn't even talk about agents talking to each other, and there's various stages of that. So I don't know if we should leap that far ahead, but that's my lens on it. That's where I think everything is headed. I don't know how fast that happens though. That's the question.

RK Prasad 33:53
OK, it prompted me to ask you two questions. One question is we are talking a lot about the agents talking to each other or talking to different tools, like an agent talks with Synthesia or Articulate Storyline, Vyond, whatever. There are different kinds of architecture, different kinds of tools built by different companies. Like AICC and SCORM, do you have some kind of interoperability protocols?

Josh Cavalier 34:34
This is where XAPI comes in. So XAPI is interesting, and I remember when it first came out and I was like, oh, this is kind of novel, we can now go ahead and track things not just within our system, but externally.

So what I truly believe is that XAPI is going to be the conduit or going to allow us that when we create hyper personalised learning experiences, they can live wherever they need to live. And what I mean by that is that for decades we have been putting things in a box. We've been putting things in an LMS, in an LXP, and they serve purposes, I get it. I totally get it, but if you go back to your advertising and marketing days, when you talk about multi-channel delivery, people want to consume content where they are at, not in an LMS, right? Whether it's in an application on the desktop, on the mobile phone, wherever, or out in the real world, maybe on a screen, somewhere on the real world. This is our opportunity because XAPI is going to allow us to go ahead and track hyper personalised learning experiences, deliver them where they need to be and not even worry about an LMS or LXP. Now again, there are times where you have to have actual data of people taking the course. With compliance, that's going to be the toughest nut to crack out of all of this, maybe the laggard of all this whole entire situation. But imagine for a moment, when I talked about hyper personalization, being able to go back to that video. It's a food safety video on shellfish, very specific like cooking shrimp at a restaurant. And so we realise that, oh, our new hires, they love their mobile phone, we're going to go and push it to a mobile phone. It's going to be very specific. It's going to be a bidirectional chatbot-based video-based interaction. Perfect. All right, well, AI builds that interaction, it pushes it to an architecture, not an LMS or LXP, but independent. It's tracking all the data through XAPI, but then something happens. Immediately, based upon the availability of shrimp, we have to go ahead and push people to a different type of fish on Monday. I don't know, maybe supply chains are disrupted. There could be tariffs or something happens where things are massively disrupted. So we have to move on a Monday. Well, no problem. Because with an agent architecture, it understands the business situation. It goes through the human machine performance analyst. It's vetted, yes, this is the current business situation. This is our source of truth. The agent then updates, almost self-heals the experience, and pushes it back out on the front lines after it's been reviewed, or not. You can have agent self-review, depends upon the level of ability. But then it goes back out. It automatically uploads it, and it's not in an LMS or an LXP. It lives where it needs to live. And so this is where we're headed, where you'll be able to go ahead and turn the dials as far as the learning experience, but not create the learning experience itself.

RK Prasad 38:17
OK, so that is left to the agents and the generative AI.

Josh Cavalier 38:22
I think eventually. That could be five or six years out. In the near term it's very much hands on for L&D because you have to be realistic that we're working with systems that have been around for 10/20/30 years. You're not going to go ahead and steer the ship immediately. So there is an opportunity in the marketplace for new systems, new platforms, new tools, that lean into AI, which I think you're going to see very quick here. And I'm fearful that some of these vendors that we've been working with for a long time, and more fearful about the individuals who have a relationship with these tools, people have been working with Storyline for years, or Captivate for years. I tell them, you got to be all right with not working in a timeline, you got to be all right with not creating the thing. Are you OK with that? If you are, then let's continue the conversation. If not, you may want to look for another line of work. That's not an easy conversation. That's a very difficult conversation. I know for myself if somebody just said, everything that you were passionate about and put all your energy in for 20 or 30 years, you're not going to really be doing that anymore.

This is one of the biggest behaviour changes our industry has ever faced and I make fun of it a little bit. When I go and talk, the analogy I give is the mouse, this thing right here that didn't exist before. Do you know how many people had to train to use a mouse, that thought it was a remote control, and they were like using it like this? This is another radical way that we interface with machines, but at a scale we've never had before. Not only that, but the impact on skills and skill degradation, and reskilling and upskilling, massive conversation because we don't know the balance of tasks that are shared between a human and machine.

When machines fail, will humans be there to pick up the pieces? And what does that look like? When you get an AI enabled platform and not even talking L&D, it could be the IT department, it could be sales, it could be marketing, it could be comms, it could be talent management. You get an agent in house that does X amount of tasks and then humans deskill or don't do the thing and then the technology fails. Who is there to pick up the pieces? And so we have a huge discussion to have, in our industry around performance, the relationship between a human and machine, and how do we analyse skills to a point that we have a fall back which I believe is going to be the humanization of work.

And what I mean by that is, we kind of equate. The machine is going to do all of these things, but the machine can also put on your schedule a coaching session, a mentorship, a 360 feedback, the human to human connections that need to be there to maintain tribal knowledge within a business, right? I mean, yes, a machine can do all the work, but you and I have been in business long enough to know there is those inside things that are not talked about to the machine, that the way that business works. So what does that look like? How does that show up? When you have a customer, that's been your customer for 25 years? A machine is not going to understand those little things that you do for that customer, that have kept them on board for 25 years, at least yet.

I won't throw that out just yet, but it's a really fascinating dynamic that we're up against and a challenging one for human resources.

RK Prasad 42:36
That's very interesting as well as a bit scary because maybe the customer also will hand over vendor relationships to another agent and it's only the agents that should be talking.
So I want to spend the last part of our conversation on what you say as hyper personalization of training, I understand that it is highly personalised to an individual.
And it happens not on some platform or a monitor or a device, but it happens where he needs it. That is what I think you mean by hyper personalization, which is possible with AI.
So the question here is, let's say I'm a salesperson or whichever, IT guy, sales guy, is easy for me. So I'm selling stuff. I may be using machines, I may be doing personal selling, I may be interacting with human beings or other machines, mine as well as the customers’. So something like Salesforce will track all my behaviours, all my actions, my tasks. And it will plan for me, when is my next call follow up call with this customer. What happened? What is the history? All that is done, and it also has my targets and how I am performing. So the question comes.
Will AI be able to diagnose my problem? If it is a training problem that I am falling behind my targets or something is going wrong somewhere? Can it identify that it is a training problem and if it is a training problem, can it devise a solution and make it available to me?

Josh Cavalier 44:59
I believe so. And so let's talk a hypothetical. Let's say that on a Friday, a new tariff is enabled, it completely jacks up our supply chain. We have to go to another supplier in country. So Monday morning, our frontline sales team needs to go and talk to a customer about this new product, benefits, cost, large purchases, that kind of thing well. That's Monday, and so the AI, in coordination with a human, because we're going to go ahead and validate the new business reality. An agent eventually could do it. I don't know the timing of that, but a human's going to go in and say yes, this is our new reality. OK, feed the new product into the agent, it knows everything about the product. This is definitely a training situation. That salesperson’s driving up to the customer at 9:00 AM. The agent's going to go ahead and create three different scenarios based upon what it knows about the customer and give live training even Socratic conversations, or bidirectional video based conversations about the new product, what's the history of the customer objections if that's in the sales data. Now we have a hyper personalised training experience that the salesperson's going to have in the car before they walk into the customer’s, that's the vision. And RK, we have all the pieces. They're all there today. They're just not connected. And when you have Google putting together model control protocol, being able to go ahead and have AI talk to our existing tools, when you have agent to agent protocol that Google just created, now they can begin to talk to each other in an efficient way. These things are going to happen, and they will happen at scale. The question is who's going to go ahead and be the first out to try it and fail and lean into the failures? And then, how does it scale? And what is the risk involved in the scenario I just described. Because if that frontline salesperson gets the wrong information, integrity begins to break down. And integrity is paramount, you think that humans are going to trust machines, there's no way. And so it's going to take a long time to build that up. And so do your systems have integrity? Because when you have human to human connection and that person is going to make the business decision, not his agent, maybe eventually the agent will say, sure, let's go and purchase the product based upon the numbers we got. That needs to be front and centre. It needs to be part of this conversation.

RK Prasad 47:48
OK so, what I could gather from your conversation is that we need to think about the backup, the latency, we need to build that in case the front end system fails. And that can be human, that can be another machine. So I got this now because you are from the medical background, medical imaging, and all this stuff, these trainings you said will be pushed not on something, but where he needs it. So do you think Internet of Things variables is something which will be the vehicle?

Josh Cavalier 48:38
One of the inputs that I have in my multimodal model is sensors. And so in robotics it's tough for me to have a conversation with people because they're just not there yet. It's something I am thinking about, and so yes, having world view or having world knowledge within a model is essential for the translation of what we do every day. Otherwise we are going to be limited by our phones, our screens, our devices that do sensory based information when you want to actually do the work and have a robot out in the real world. I'll give you another one. You're going to love this one. Recently Llama 4 came out, and Llama 4 has a 10 million, at least the scout the smaller model has a 10 million token window.

What are you going to do with 10 million tokens? I'll tell you what we can do. Let's go ahead and upload 20 hours of video of a human performing a task over and over again, edge cases, repetitive movements, mistakes, all of it. That information then gets analysed by the model to train a robot. So the robot will eventually be able to understand all of those physical actions, and the human will then shift their role to working with the robot, making sure and validating that they are doing the task, again, yes robot performance. And then eventually, maybe they'll be there to assist in fixing the robot or improving the robot or improving the environment. They have a role to play because they're experts in that process. It's funny because years ago I had a wonderful opportunity to walk the floor of a Toyota plant in Kentucky. And even at that time, this is 10, 11 years ago, humans and robots are working together to spot weld parts of the vehicle. The human had to take the piece of metal, put it into the rack. The robot grabbed the piece of metal and dropped it down and spot welded it. So even 10, 11 years ago we had humans and robots working together to become more efficient on that car line. That's where I see robots happening and again sensory based information whether it be through video or through actual physical sensors as sensing temperature or GPS or any other kinds of worldview sensors, is going to play a huge part. But I think I might be getting a slight bit ahead of myself because I don't know how long it takes to translate that into every physical business environment. But I believe the Nvidia's and the Boston, the company that's doing the robots, Boston Consulting Group, maybe, they're doing all the different Boston dynamics. They're doing all of the robot based work and so, where are the price points at? How long does it take for robots to actually? It's funny? Wherever I go and talk, everyone’s like when are the robots going to do my laundry, when are they going to do the dishes? Now we're talking, I'm ready to buy a robot, right? I mean, we're human, that's what we want. So it's fascinating to actually put it out that far and believe those things are going to happen, but we just don't know. We don't have a timeline. We just don't know.

RK Prasad 52:01
Yeah, but it'll it is going definitely going to be sooner than later.

Josh Cavalier 52:16
I think so, yeah. But it's funny, you have a situation like what's happening with the tariffs, it completely impacts CapEx off the top. When you take a lot of money off the table, it slows things down in regard to the velocity of data centre and the build of those data centres. On the flip side, how efficient can they be with their engineering and use the existing data centres that they have, all it takes is 1 unlock. One unlock and we have a whole different conversation because now the GPT 5s, the Gemini 3s, those models can scale up relatively quick and get out into the marketplace. Now we have a different conversation.

RK Prasad 53:01
Right, right. Wonderful. There are so many more questions I'd love to ask you because it has been highly educative to me and opened so many windows of thought, but I think we have just about a couple of minutes. Is there anything else you'd like to share with our listeners?

Josh Cavalier 53:23
A lot of things that we talked about, when people listen to us are going to be scared and I want everyone to know that this is about agency. It's about understanding your relationship with technology and everyone listening to this is in the driver's seat. Do not think that you cannot own your destiny or own your future. When it comes to working with AI or working with machines, you are in control. It may feel like you're not in control, but you are in control. And these are the conversations that we need to have in HR, across function, because at the end of the day we have to make sure that we understand what our relationships are with machines and do it in a positive ethical fashion moving forward.

RK Prasad 54:16
Thank you very much, Josh. It's a privilege meeting and speaking with you, where do you live in the US?

Josh Cavalier54:26
I'm in Charlotte, NC, and if you want to find me online, joshcavaliere.com or josh cavalier.AI.

RK Prasad 54:28
Oh, OK, I have seen your thing. I'll send an invite. Yeah, we do have such some customers there and we keep visiting North Carolina, Charlotte and so maybe we'll meet for over a cup of coffee soon.

Josh Cavalier 54:51
I don't it's a maybe. I think we will.

RK Prasad 54:53
Yeah, it'll be wonderful. And I look forward to reading your book. I'm sure it'll be a big hit because with all your experience and the timing of the book, I think because last ATD conference I've been to New Orleans. So when I went there and I saw the the books on sale, there was not a single book on AI as related to L&D. Michael Allen was there and I went and introduced myself and then.

Funnily, he lives also in Saint Paul. So we caught up after the conference and we went for lunch and all that. And so I asked him, why don't you write? I mean, there is not a single book here. You wrote so many books on instructional design.
He says I won't write a book on AI because I think still the human being is a must for instructional design. So how do you create that learning experience? He was still very focused on the human aspect of it, so this book will beautifully complement what he has said. So I think it will be a big hit, Josh, and I'll be one of the first guys to buy it.

Josh Cavalier 56:35
You'll get a signed copy, my friend. That is the arc of the book. The arc of the whole book, is the transformation of an instructional designer into a human machine, performance analyst. And what does that look like? It is a blueprint for individuals who still want to be in this industry to impact human performance and now machine performance.

RK Prasad 56:59
Right, right. This is a very good title you have given, where the person has to concentrate on the beginning and the end. The middle part is taken care of.

Josh Cavalier 57:13
Well, you can thank ATD Press, my editors for that title, because I had other titles and they reeled me in. This is a perfect example of letting the professionals do their work. Right? So yeah, I can't take credit for that. I got to thank the folks at ATD Press for at least guiding me in the right direction on that title.

RK Prasad 57:13
Good, wonderful. Wonderful. Thank you very much once again, Josh. And you enjoy your week ahead and enjoy your Monday. Thank you very much and I hope we'll be in touch, and we will probably collaborate in this area.

Josh Cavalier 57:39
Sounds good, RK. Take care. Bye.