Why AI Power Users Will Outpace Everyone Else

David Rice: Most employees are doing about 1% of what's possible with AI. That's not because they're lazy. It's not because they don't have access. It's because nobody's showing them what they could do and that they should be using it 5 to 10 times an hour to analyze everything that they do, summarize their day, and poke holes in all their logic.

Today's guest is Kevin Surace, he's the CEO of Appvance.ai, and he speaks at about 40 to 50 company events a year around the world. And when he asks deep questions about how people are actually using AI, one or 2% of hands go up. Everybody else is using a little Copilot to fix their sentences and thinking, whoa, look at AI.

But meanwhile, there's a 23-year-old at a Silicon Valley startup using it like they have a PhD with 28 doctorates sitting next to them for a penny a minute. And those 10 people, they equal about a hundred of the rest of us. So here's what Kevin wants you to understand. This is not a capability gap. And every year you don't learn this technology, you start to fall a little bit further behind until one day you're no longer relevant.

It happened with a desktop computer in the 1980s. It happened with the internet. It happened with Excel. People either got on the bandwagon or they were phased out. This wave is moving much faster than all of them. So today we're covering why your prompts should be three paragraphs, not five words, how to think strategically about what you want to say instead of typing word by word, why today is the worst it will ever be when it comes to this technology and what that means for your profession, the identity level shift happening across entire industries, and why resistance, well, to quote Star Trek, it's futile.

I'm David Rice. This is People Managing People. And if you used AI once last week while someone next to you is using it 10 times an hour, this conversation is gonna be your last warning. Let's go.

Kevin, welcome. How we doing today?

Kevin Surace: Super happy to be here.

David Rice: Where I wanted to start with this is, you know, when we were talking before this, you had mentioned the gap that a lot of people are experiencing, most employees are using maybe 1% of what's possible with AI. I'm curious, why do you think that's happening and what's sort of the cost of continuing at that pace?

Kevin Surace: It is a very interesting phenomenon. I get to speak at large company events and industry events 40, 50 times a year, right around the world. What you see is very few people raise their hands when you start asking deep questions about how are you using gen AI? It's like one or 2% raise their, oh yeah, I'm doing this.

I'm writing agents, I'm doing, but most people are not doing anything. They're using a little bit of Copilot. It helps 'em correct a few sentences and they go, woo, AI, and they have no idea that you should be using it five or 10 times an hour. To analyze everything that you do to summarize everything that you do to poke holes at everything that you do.

If I'm writing a presentation, I say poke holes at this. Then when you do poke calls and how can I make it better? And then can you help me make it better? And then can you draw this drawing? And then can you analyze this spreadsheet and what does this spreadsheet mean and what does tab three mean?

Because you know, with all of this, you could do every minute of every day. And it's like having a PhD next to you that has 28 doctorates, right? And there are a penny a minute or a penny an hour. Why wouldn't I want to use that? Every hour, multiple times an hour. So I think people don't know, I think a lot of people are only on Copilot, which is, you know, just this very simple little thing to help you along and they're not really leveraging all of the models in the way that they can be leveraged.

And I use all the different models, right? Depending on what I wanna do. Gemini is better certain things than GPD 5.2 and that's better than maybe anthropic at certain things claw, blah, blah, blah, right? So you use them all.

David Rice: I think it's a bit like you put me in a fully loaded cockpit. I don't have any flight training. You know what I mean? So it's like.

Kevin Surace: That is true. Partly what's interesting is I'll give demonstrations on stage. Oh, some things that I've done literally for their company. I said, I don't even have any proprietary information in your company, but lemme show you what I was able to do. Well, here's a brand new product that I think would do well and here's the size of the market and here's the tam and here's how I know.

And they go, what? We would've spent $5 billion figuring that out. I spent five minutes. Or maybe 10. There is no clue that they can do that with AI because nobody's shown them that before. Nobody's trained them how to do it. Most people, they're typing in a, you know, a prompt. Five words or something. You know, my prompts are three paragraphs and very thoughtfully architected for who's my audience, what am I trying to do?

Who am I targeting? What am I trying to say? Why am I trying to say it? What are my opinions on that? Right? There's a lot for me to put in there to get something valuable out. Not write a paragraph on X and that you don't know what you're doing, right? I mean. You misusing the tool. So I think people do need training.

They need to get comfortable. They need to have access to more models and more tools than just Copilot, right? And then it's game changer and lastly, look, if you're at a Silicon Valley startup, you're doing all these things times 10, this is what you do, right? So they're listening to this going, yeah, of course we do this, Kevin, if you're at a large company in the Midwest.

You go, well, I just got Copilot here and I asked something in Word and it fixed my sentence. And they go, whoa. It's like it's taking the place of Grammarly or something. So. So there is this bifurcation. That's a big problem for a large company because you got startups now that are running at 10 times the speed of the number of people that they have, right?

They're 10 people who are equal a hundred because they're using AI, variety of Gen AI models and gen AI tools for marketing, for sales, for analysis, for poking holes, et cetera, for sparring with. They're using that stuff. 10 times an hour, every person, and you used it once last week, you're done. You know, you're just gonna be wiped out.

David Rice: I think we, you know, there's a lot of talk about capability gaps, but it's, it feels like a bit of a confidence gap here. Like I don't think people are avoiding AI because they don't see the potential. I just think a lot of people don't know where to start sometimes, and. That slow start, do you think that it compounds?

Because if you don't start now, then 12 months from now, you're just.

Kevin Surace: I say this, yeah. You keep getting further and further behind and then it starts to get embarrassing, and then you don't know what to do and at some point you're no longer relevant. When is that? Right. So the desktop computer landed at a point in some people's careers in the mid eighties, the late eighties.

If you were one of those people that said, I don't want one of those things on my desktop every year that went by, you fell further and further behind, and finally you could no longer participate in work you needed to retire. Literally, it was hopeless, right? I mean, how could you live without that? And then the internet came.

There were a lot of people that, I don't need that at work. I don't need it at all. I don't need to do anything with it. I'm not gonna alert it. I'm not gonna, you know, you either got on that bandwagon, or I guess you retired also. And the same thing happened with Excel. Excel shows up in the late eighties and early nineties in every finance department. Right?

And there were lots of people who hopped on and said, this changes the game. I'm going to use Excel in everything I do. And there was others who said, I don't wanna learn this thing. I've got ledger books here and I've got pencils and it all works fine. They retired. There was no other off ramp for, I think the key takeaway here is when you don't learn the latest technologies that are pretty widespread, you know, you head to retirement and that's it.

That's fine if that's what you wanna do, but just know that's where you're going, right? If you're retiring in a year or two, who cares? It's like the doctor that doesn't wanna use the latest medical device. He says, I'm retiring in two years. I don't need to learn this. You're right that you will surely retire in two years.

'cause no one will wants you wrong. And so anyone listening to this that says, I don't really need to learn this. I'm 45. I got another 20 years in industry. No you don't. You have another three. No one will want you working for them. You know, if you're not a master of these tools.

David Rice: Another term we hear a lot now is is shadow AI, right?

And it's this sort of use is everywhere. And for good reason, people are frustrated with a lot of the enterprise tools that they often feel are watered down or slow to deploy. And so I'm curious from your perspective, how do we close that gap between what people need and what they're actually given?

Kevin Surace: Well, what I've said to enterprises is. You need to have all the major models available, not available to everyone in the corporation, but you have your privatized versions, right? You have your corporate accounts and you give access to people who need them in certain departments, right? So drummer comes to you and says, look, I gotta analyze videos and here's the reasons why we're analyzing this line.

We're analyzing. Great. That's Gemini. You now have access to that. We have a corporate account. Right? And so you need to have all the tools available. What happens in a lot of companies today is you go and say, I want to, I go to it. I want to use more AI. Well, we gave you Copilot. Yep. It corrected this sentence.

I wanna do more than that. I'm being a little facetious. Copilot will do more, but you know what I'm saying? And they go, well, we no, and we don't allow you to use anything else. So they go home and do it, or they do it surreptitiously and then they get blocked at work, you know, and all this stuff, right.

It's like if you want to be successful, you're going to have to really empower all of your employees. Right? You have no choice.

David Rice: Yeah. I mean, I think I hear this a lot like where, you know, it's like sort of the approved tools feel like they're built to slow you down, not speed you up. And so people go rogue, right?

Not out of rebellion, but more so out of necessity. 'cause that's the expectation now is you're gonna do more, you're gonna do it faster. And the tools being deployed don't really match the real needs of a lot of people's work. So I think. To me, I don't know. I don't know if it's a compliance, I think it's relevance, right?

Kevin Surace: Look it's easy for it today to set up corporate accounts with 10 different tools, whatever. It's some for sales, some for marketing, some generic, five generic models, et cetera. And then only add people as needed in different departments that need it, right? But let their company know that got lots of tools available you need to ask, right?

We're not gonna roll this out across 40,000 employees when 200 need it. We're not gonna do that, but if 200 need it, we're gonna do it because it's gonna make you more productive. So that's the thing. And a lot of the demos I give, you know the, some of the feedback I get is, that was amazing. Unfortunately, the company doesn't give us access to that.

And so I always preface and say, your company may not give you access to everything I'm showing you. And in fact, on my tools page, there's like 40 different tools I used in the presentation and they go, we don't have access to all of those. I said, well, yeah, but it is in the room is so your CEO, so that's between you guys.

I'm here to show you the power and opportunity of Gen AI today, and that will take access to a lot of different things. And you should be asking your exec team, your exec management, your it, whatever for what you think you need, and then you gotta sort that out. Do not go out and use things Rogue. Can't do that.

Shadow AI, shadow it, it's all bad. You can't track it. People are using things and they don't have an account. Now, if you have an account on most of these, it's private, but a lot of people just don't have a paid account and they go, well, it's probably fine. No, that's all public and they can train on and they can come up again.

That's all bad. You shouldn't be doing that. And that is going on and you know, all the work environment to do is try to block those IP addresses, those URLs, right, keep you from going rogue.

David Rice: You're pretty adamant when it comes to agents that companies aren't gonna really move forward with it until they solve for security and control.

But I'm curious what kind of gets lost if we wait too long. Like are we holding back that next big leap just 'cause it's messy?

Kevin Surace: So companies are doing some work with agents right in inappropriate places. Part of what holds companies back is they spent 10 years rolling out RPA Robotic Process automation and RPA is the kind of agent.

That is very regulated by the set of rules, right? It's rule based. It's not bad. It's not AI based, but it's rule based. And so any decisions that are made, it goes to a decision tree and it executes that. It goes through a set of rules or rule based. Those work very well in the very predictable. You understand the security issue, you understand everything about them, right?

Everything is good with that. So now that those work, and in fact you've automated a lot of your workflows with RPA. Agents aren't this miracle. 'cause what you wanna do, what you might do with an agent is say, oh, I can automate this workflow even a little better with more smarts. Yes. But at what security risks and how long does it take to work out the bugs and end Right.

It took a long time for RPA to come to fruition. I think the fact that people have rolled out RPA already is the bigger block or two, oh, I need an agent to do that. They already have an agent doing it. It's an RPA agent, but it works. So an example is, oh, I need to process something in insurance, and it's gotta go between 11 different systems.

RPA does that today. You kick off the thing, it acts as an agent. It does the whole takes in. The data goes between these systems, it files it over there, checks with this. It goes through the rules. It comes back and says, I'm gonna issue a check. And you go, great, right? It does that already. I can deploy an agent to do that, but I'm not gonna save any money.

I might get better decision making with a little less human intervention, but I could also get a bad decision. Because now I've allowed the decision making to AI rather than a set of rules that I created. So as a corporation, I don't have as many things to automate today that require an agent versus the RPA that I already deployed for the last decade and really worked hard to deploy and spent tens of millions of dollars deployed.

It doesn't say there are places for agents to, you know, to take over. I think. A lot of corporations closed those biggest gaps over the last decade. That's my point. There was the technology before agents that works very much like an agent. Good enough. So they're not gonna see the cost savings now, I think, on people's desktops.

What's interesting is with agents is. You get certain workflows that are very unique to just a few people that the corporation would not have automated. That could be interesting, right? But are you really gonna allow people to write their own agents and what's the security impact there? Because these agents are accessing our most critical systems, and I don't know if they have a right to do that.

And I don't know if it's really the agent that's doing it or some rogue hacker, right? So these are all tricky things.

David Rice: Is it maybe a case where that the use of agent sort of starts at the top and then we sort of see where it goes from there? You know, like maybe your C-suite has an agent, but not really anybody else at first.

Kevin Surace: Maybe, but also that's the one that a hacker would like to get into and take over.

I think the aged question is a real tricky question. Because again, you've got RPA that was deployed against the biggest targets anyway. It's working. It's been saving money and increasing productivity, and now you've got this kind of newer technology can take it over and show me the savings.

I've already automated all the things that could have been automated over the last decade, whereas the additional savings, is it 2%? Am I gonna bother? Right? They save the 80% that could be saved. Right? So I think these are tricky questions. Clearly when we look out 10 years, agents are gonna be deployed all over the place.

They're gonna be highly secure. We're gonna work through all those issues. They'll be replacing RPA or the RPA. Companies themselves are going to be redeploying, RPA as smart agents, all that will happen. But that's, you know, and again, we're talking about large enterprises. A startup is gonna deploy agents doing everything.

Everything they can, right? Because they're cheaper than people and they don't have RPA, they don't have a legacy technology that already automated things. So they're starting from ground zero. They can use agents.

David Rice: Yeah, that's what I was just about to ask was like, you know, the startup space feels like it's just ripe for the experimentation. That's where we learn, that's where we build that muscle.

Kevin Surace: That's right.

David Rice: You said the real superpower isn't the model so much. It's the person who knows how to use the right model for the right job. You mentioned earlier, you know, you, you use this for this and you know, each one has its purpose, but. I see this all the time on like X, right?

Like there's these people with like these huge infographics, you know, and they've got it all mapped out. But I don't think most people, I think most people don't know what they don't know, right? Like what's the skillset?

Kevin Surace: How you learn is due, right? I mean, how you learn any new tool is to go use the tool and learn it, and you spend an hour or two and learn what works there better than work somewhere else, right?

So if it's just a transformer model, LLMA, multimodal, you can, you could do that and start to learn, oh, I. If I'm doing research of this type, I'm gonna go to perplexity. And here's the reasons why, and here's how it gives me the references, right? If I'm gonna do video analysis, I'm gonna go to Gemini, right?

If I'm generating music or trying to work in the music field, I might do here. If I'm trying to work in the video field, I might do this, but it depends on what I'm trying to do. Like D script is better certain things, then the P is another thing that so is another thing. And we could just go down those, right?

But if you haven't taken the time to get those subscriptions and learn all these different tools, then you got a problem. There is no, I see the infographics also. Right. And that's their view. But until you go use it, you don't know what to do. I mean, you literally don't know. And it takes, so these technologies now take a lot of time to understand, right?

Right. If you're a musician and you've been used to a daw, a digital audio workstation, then now you want, you know, have a sense of STEM separation and MIDI export. iNOS Studio you go. It's kinda like a dos online I get. Okay, got it. Right. I kind of get that. If you've never used any of those tools, you can't walk in and say, I'm gonna use pseudo studio, where you don't know what STEM separation is.

You don't know what those individual instruments or stems mean. You don't know how to manipulate them, you don't forget it. I mean, you just kind of host. Right? So you need some background in some of these things too. If you've never edited a video before in your life, you're probably not gonna show up and create a movie with any tool.

You just don't have the skills. But if you go and start to use those tools and work with them for days or weeks, even if you never had any prior experience, you're likely to go, oh, I see how to get it to do this versus that, and then I can see how I can string these together. Then I can see how I can extend a scene.

Then I can see how I can put close captions on it, whatever it is, right? And you have to learn all those pieces, right? You don't walk into Hagen and go, I'm ready to go. But you can. Use Hagen in the four or five different kind of models that they have and within, you know, days many hours, right?

You could start to get really good at a few things there, and you go, okay, I'll give you a power user. And I think that's where people need to get, depending on their career trajectory, what they're trying to do, whether it's at work or at home. What am I trying to accomplish? What are the tools available?

How do I learn those, right? If I'm in marketing. I'm using all kinds of AI tools here to do marketing that are targeted at just that. I don't mean blog posting, I mean email targets, LinkedIn targets, customized emails, all that stuff. There's a bunch of stuff now that is just really, it's powered by AI.

It's very powerful. It's amazing. Right? I was watching a webinar this morning with Zapier and DS script and using the Zapier essentially agentic flows that you can write and create, and it's pretty darn cool. And I'm looking, I haven't used it, right? So I'm looking at it going, oh, I have a lot to learn there.

Right. You know, I haven't done that. And so yet, another thing, I mean, there's a million things I haven't done right. And I try to do this every day. So there's always more to learn, stay curious.

David Rice: Growing a global team shouldn't mean juggling five different systems for HR, payroll, and IT. Deel brings it all together so you can hire, onboard, pay, and equip people anywhere without the usual chaos. Whether you're bringing on employees in 10 countries or managing contractors across time zones, Deel handles compliance, benefits, and payroll in one place, which means less tool sprawl, fewer headaches, and more time focusing on your people.

Ready to see what hiring without borders actually looks like? Visit deel.com/pmp. That's deel.com/pmp to book a demo. That's deel.com/pmp. Deel, hire, manage, and pay anyone anywhere.

And it doesn't really feel like, to me, I guess I've experimented with quite a few tools now and I feel like I, it's not about technical wizardry.

It's if I'm gonna build sort of an ecosystem of tools, it's really just like systems thinking and sort of judgment around what the best thing is. But I think traditionally we've kind of trained people to solve well-defined problems, and I think the value is shifting to like what kind of questions are you asking?

Kevin Surace: Yes.

David Rice: Because you gotta be able to orchestrate the system of tools.

Kevin Surace: No question. And you can certainly write agents that help you orchestrate that set of tools, which is really fascinating. So you go, well, I defined my workflow, let me create an agent to do that workflow for me, because I go from this tool, then I all put that, then I take the file and I put it over here, and then I hager, and then I add, I can create a workflow to do that, right?

Automatically push a button the way. We've kind of had ways to do that before in scripting, both in Windows and Mac, right? There were tools to do this, but now it's smart, much smarter. It can analyze. What was said in a video, for instance, you can analyze an hour and a half video and automatically break that into 20 different social media posts that you don't do any editing on.

It did it for you based on what was said, including in this particular podcast, right?

David Rice: Yeah. I do it all the time. Well, we were talking before, you know, you gave this great example, like what used to take McKinsey six months and millions of dollars you can now do in an hour with AI. Right? But it's not just the tools.

The key is in how you feed it. I'm curious, like what's your process for getting AI to think with you, not just for you?

Kevin Surace: So look, you know, job one is always large and long and thoughtful, opinionated prompts, right? Who's my audience? What do I think about this topic? What I mean, I spend, I could spend 10 minutes, you know, writing several paragraphs.

That's my prompt. And people go, you did what? Well, yes, but now the value that I'm getting out is a mess. And so, again, think about who your audience is. Think about what inputs you want to use. Think about the sources of material. Think about your opinion on the topic. Think about your voice on this topic.

All the above. I do find it interesting. Look, I can do a really complex market study of let's say, vehicles in South America, let's just say, and if I were to hire McKinsey to go do this, to just pick on one, you know, I'm in a company, I'm an Uber or something, right? And I wanna understand the chance of Uber's Uber being successful in South America.

I'm making this up. Well, I will certainly spend $5 million on McKinsey and they will do all this research. They'll put people on the ground, they'll look at all the papers, they'll look at government reports, and they'll put all that together and come back and go, boom, three months later. Right? And for my businesses, for spending the 5 million, except I can assure you I can do that right now and it might take an hour.

'cause you're gonna go back and forth and you're gonna use the thinking model. So it might even take it five to 10 minutes to stew on it. It's gonna go do that research for you. And what it comes back with is amazing. And then you gotta work with it to generate crafts and charts with kind of the reports that you want and put a into a PowerPoint.

But in an hour, I have what used to cost me $5 million and I have it for a dollar in my time. This is true. This is a factual thing. I've done this, I did this for a company recently that said, you know, we had 82,000 warranty claims, you know, over the last few years. What can we do to reduce them? Well, since they were all typed in English, you couldn't really segregate them.

You couldn't easily do traditional analytics in a database. They're just all in Excel and there they are listed and this what wrong and this what wrong and this what wrong, right? But Gen AI can take that in, literally analyze that entire spreadsheet, read all of the English that's there, and come back with five recommendations of changes we can make in production starting tomorrow.

And I did that for a company live for them, and they looked at me and go. You've gotta be kidding. Like, we've been sitting on this data for a couple years. We have no idea what to do 'cause it's just overwhelming. We're gonna hire some agents, you know, some group, some consulting firm to go through and triage it and, you know, analyze and you did this in five minutes.

So this is what people aren't thinking about. I can take in huge amounts of data as long as I haven't exceeded. Its, you know, its aperture, its wind up. Analyze a immense amount of data that I may not have to clean up. Now, in traditional AI, I really have to clean the data, but in general, AI, as long as I, you know, it knows it's not clean and it's just this mishmash and it's all over the place.

It'll just read it all. It'll come to a, you know, a conclusion based on that reading. And the co conclusions were amazing on what the recommendations were, and they went, as far as I know, I think they went and implemented them. Why wouldn't you? You know, we're like run, right. So there's lots of things like that.

I had another one with injuries in the factory. Here's all the injuries that occurred in the factory. There were thousands and thousands. How do we reduce them? Make recommendations, you know, it was minutes. Boom. Here's the recommendations. What, just do these things. You'll cut your injuries by 80%.

What? And I poked at that a lot and I'd say it was right. I mean, you never know if it's right until you implement it, but it was on the right track and humans could not have come up with that. McKinsey would for $5 million. But it does give you an idea of what's changing these days. Right. And what we can do.

These are very powerful tools you have at your disposal if you know how to use them right.

David Rice: Yeah. I love this example. I think it goes back to what we talked about at the start with people, you know, using for maybe 10% of what's possible. But it also captures the fact that, you know, it's fast, but it's not magic.

You know, the output reflects the quality of the input, so to speak. It's easy to kind of confuse speed with insight, I think. But I like that you're getting at something deeper, which is you're co-creating. You're thinking together. You're not just delegating. That's where you start to get in trouble is when you do that.

Kevin Surace: It's a sparring partner that's smarter than you are, but it's a sparring partner and you wanna poke at it and say, why? You think, by the way, I tell it sometimes it's wrong. I said, that's wrong. You didn't think about this. Right. And the good models, pretty much all of 'em now will come back and say, you're right.

I didn't consider that. Now that I consider that this is the answer, I go, that makes more sense, right? So I spar with it. I tell it's wrong. I ask it about my recommendations, my opinions. Is this opinion valid? Is this opinion, right? I was analyzing something yesterday is the whole article on the fact that AI can't make a moral judgment.

And I go, huh, I don't agree by the way. 'cause I think today we can. So I said, here's the article. Explain to me why this author thinks that AI cannot make a moral judgment and explain why you think you can't because right. He says, but I can, because I've learned everything that humans have done. So I understand what humans would think is moral right, and therefore I can make a moral judgment.

I can't make a judgment on behalf of you. You'll have to live with the consequences. But I can tell you what is likely to be considered moral in these outcomes, right in these outputs. And so I thought that was a great back and forth to say, you know, defend yourself here. Right? Tell me if you can or can't make moral judgment.

Of course it can, based on what it's read, human think is moral. Does that make sense? Like it's clearly immoral to kill people. And then I posed the question to it and said, help me understand the morality of just 'cause it's timely right now. The US invading Iran. Just help me understand the morality of that, not whether we should or shouldn't.

Just what are the moral implications? Boom. Here are the moral implications of doing so. And here's two sides of that equation, and here's how the Iranian people will feel. Here's how the world may feel based on this data. And you go, this is freaking great. Don't tell me you can't suggest how, you know humans are gonna feel from a moral judgment standpoint.

Should we do? I'm not suggesting we invade anything, but should we invade something? What does that look like? Right? And how are we gonna be judged in the world? That's fascinating. If I am in the CIA, if I'm in the government, I actually wanna ask the machine, based on everything you've read, how's this gonna look?

And then poke at it, right? And go back and forth it. So again, I'm saying these to people and people in startups are going, you know, of course you can do that. Most other people listening are going, you've gotta be crazy. You can do that. Of course I can, by the way. I can go further and say. Based on every known war, you know, in the last 30 years or whatever it is.

If the US were to do that, how should we do it? What equipment should we use? How many people, what are the risks? You can ask that. They'll say, well, sure, based on what we know, based on all of this, here's how it should proceed. I have a thousand analysts in the Pentagon working on, I can just ask this model and I'll tell you, it's ridiculously good.

David Rice: When we were chatting before this, you know, we were talking about the fact that we're kind of headed towards this area where it's maybe just a few models, three or four that dominate because once accuracy hits a certain threshold, right? The difference to most of us becomes pretty much invisible, like.

Kevin Surace: It's almost invisible now, right?

Unless like, unless you know that Gemini have to be better at this one thing, then you wouldn't, you know, you'd just use GPT 5.2 or ChatGPT if it's the public one. Yeah, I think we're gonna ask them tote to where all these things have encompassed, you know, a hundred percent of all humanity and there's nothing more to do.

And we're already doing that, right? So if you think three years ago there was kind of one model. It was ChatGPT, it was 3.0 or something, 2.5, 3.0. Somewhere in there where it first was available and it made lots of mistakes and it had lots of problems, but it was pretty impressive. And then there's Claude, and then there's Gemini, you know, and then there's.

X.ai and then there, you know, you just go down the line and now there's, and then there's the ones from China and then there's llama. These are all performing very well. And if you look at the tests on them, you know, they're all now within a few percent of each other. They're all very good in most questions that they're gonna be asked, right?

Or most interaction or most sparring that you're going to do. So there's going to be less and less differentiation. There has to be, 'cause they're getting better. None of them can get any better than the human knowledge that we've given. At least with a transformer model, it simply learns on everything.

It's read right or seen. So I think we do ask them to and maybe it won't matter. And then when you ask them to, of course the cost of these things starts coming down. Everybody starts working on inference costs get less and less because it becomes a highly competitive business and maybe nobody's paying anything for the things or they're paying a dollar a month or something because gets highly competitive.

David Rice: That was my next question is sort of what does that mean for how companies build and buy AI sort of in 2026, but also it's kind of when they think long term.

Kevin Surace: You know, hopefully nobody's building their own gen AI models in the enterprise. You go and leverage those, right? You leverage what's out there.

But I'd say how does it impact what they do? Well, I look costs are going to continue to come down. So negotiate your contracts wisely. You may renew them manually and look at the cost. But there, you know, this is following Moore's Law in general. And so costs 10 years from now had better be, you know, 5% of what they are today.

Costs are gonna fall by 90 to 95%. The compute costs are gonna fall that much. It's just what it is. 'cause in the end, this is running on GPUs instead of CPUs, but it's running on silicon. And silicon costs get cut in half every 12 to 18 months and still do. They get cut in half every 12 to 18 months, and this is gonna happen.

So you do that math and you go, I get that this cost reduction compounds on each other and we're gonna be 90 to 95% cheaper in 10 years. So just, that's what you're gonna follow. And I think that's good. That really makes this, it's already accessible to everyone, almost to the world. And so this just becomes even more and more accessible, even for really complex tests.

David Rice: I'm curious because it feels like this one's different. That's what we keep, we, I feel like we all keep saying that, right.

Kevin Surace: Say then every tax cycle, this one's different. The internet one is different, the smartphone one is different. PC on the desk is different. They were all different.

David Rice: I'm curious though, like is there sort of a signal that you're seeing right now that tells you like this is a moment that leaders just can't afford to sit on the sidelines?

Kevin Surace: So was the PC on the desktop? So was Excel. So was word. So was, I mean, that's the thing. When a word processor showed up, you could not afford to sit on the sideline. I mean, you couldn't or else you have to retire. Right? And this is one of those cycles that if you sit on the sideline, you're retired and that's fine too, but you can't sit on the sideline.

I think people want to keep saying it's bigger than the rest. Well, I don't know the inter, well, the internet was bigger than everything that came before it. Arguably, smartphone was bigger than everything that came before it because it was connected to the internet and built on everything we'd built on.

This one is bigger than everything that came before it because it's built on the internet and smartphones and PCs on the desktop, right? It's built on everything we did before this, so that allowed it to get out to three or 4 billion people in a matter of we, because we had the internet. Can you imagine if we had this technology and there was no internet, five people at a university could use?

You'd have to walk out to the machine, right? But there is the internet and that allows us to talk regardless of where we are in the world at high speed, I have, you know, gigabit symmetrical here and works great, right? So I don't think about that anymore. We don't think about the cost of transmitting data anymore.

I can send terabytes across the world basically for free. And 20 years ago, that might've been $20,000 to set a terabyte or two, and now it's zero, right? So this has happened. Over and over again. We're building on those prior waves and I think that we're building on it quite successfully. And you certainly don't wanna miss this wave or miss out in any way.

The person next to you using this five times an hour is 10 times smarter than you are. 10 times more productive than you are, like it or not. Very big arguments in the music community right now of leveraging AI tools to help generate music, to generate parts, to generate new arrangements, orchestrations, things like that.

Those who want to and those who don't, and even the music community has been through multiple stages of technology. Before it used to be everything was pencil and paper. Then we had notation programs that made you much, much more productive. Then we had something called digital audio workstation. Then we added samples or digital kind of digital recordings, digital samples of instruments, and all of a sudden I didn't need a guitar player.

I could make this thing play guitar digitally. Now we've got AI and I say, just make this guitar part. Make it sound like this and go do it. And boom, there it is for the whole song. And you go, oh wow, I didn't have to write a note. So you've got a bifurcation in the music community. Half the music community says, I will never use that thing.

It is a horrific tool and takes away our creativity. You got another half that says it's making more and more creative because I can put out 10 songs an hour if I want to even, you know, even with my proprietary melodies. This is fascinating and I can tell you guarantee the winners are using the latest technology every time.

The people, when digital audio workstations came and samples came that said, I'm never gonna use that. I'm gonna play this instrument and I will never use anything digitally, they got left behind the digital music revolution of today. Most film scores and all TV scores are done digitally. I'm sorry.

There's no tropic player in that studio. You heard the tropic, nobody played it. Right. That is the digital representation of a trumpet and it's perfect, near perfect. So that's the direction of these things. And there's gonna be a bunch of people left behind. And I have arguments with them every day. And the, and I honor it, they are very serious about, I will never use that stuff.

And it takes away our livelihood, our creativity, or, okay, but there's a guy next to you using it, or a gal next to you using it, and they're gonna take away your job. I mean, that's the thing. So if you're a blog writer in marketing and you go, this is what I've done. I write blog posts, I'm really good at, you better be using AI and writing 52 blog posts a day now.

'cause if you're still writing one every three days, okay, there's someone next to you writing an entire year's worth at a day, you're dead, you're done. I can't change the fact that you're mad that guy took this wonderful skill that you had to type away from you. That's gone. Nobody's gonna pay you to type word by word anymore.

They're gonna pay you to really think about strategically what you wanna say, get it said, come back and edit it and push that out. And that takes 15 minutes. It doesn't take 15 days or 15 hours. That's the way it is. I'm sorry about that. That just like, Hey, I'm really good with a typewriter. Okay. We don't use typewriters anymore.

Right. We don't need that. I know how to use the Faxa machine. We don't need them. What are you talking about? So that's true with every technology. It's very true today.

David Rice: Yeah. I think that's for me, the, you got at it with the musician thing, right? Like the difference is that it's like sparks an emotional reaction.

It's not a business initiative, it's an identity level shift for people. And even like whole industries like you said there with like marketing, right? So we're seeing this I think it's natural to some extent that there's gonna be sort of that sort of kickback, I think with leaders. We mentioned that compounding effect before, it's the same thing, right?

Like it's this way for everybody. And if you're not keeping up with it right now, the speed at which it's moving, because I think that's another thing a little, it's a little bit different, like, and it's harder to catch up than that it's ever been, I'd say.

Kevin Surace: Yeah, no question. Look, the it's moving at rapid speed because it's building on the prior technologies, and we're now in an internet and wifi phase that everybody's got access to everything immediately.

And so in three years. It has moved incredibly rapidly and the technology's gotten better and better. Like I like to say, I'll, you know, I'll generate something in video and they go, wow, that's, it's kind of indistinguishable. I, as far as I know, you shot that with a camera? I said, yeah, and today is the worst it'll ever be.

That's when it hits them. And same with music. I had seven songs I was working on, right? Some were actual humans in a studio and musicians, some were fully generated by AI. And I sent it to musicians. I said, just tell me which ones were which. They all give up. I said, they may all be AI, none of 'em may be AI.

Some may be just tell me what you wish. Because they insisted that, well, AI doesn't have a soul. And I can tell. Okay, go. Can't tell. I said, okay. This is the worst it will ever get. It doesn't get worse than today. Tomorrow's generation will be slightly better than today's, and then they kind of go, I'm screwed.

No, you're not screwed. Just learn to use the tools, use your experience in music to produce something better than someone who has no experience. Right? Same with Hollywood. Hollywood is, you know, they saw Tilly, the fake actress, and they go, we're screwed. Well. I can't answer for every role in Hollywood, but I can tell you stories are going to be made and they're going to be made with digital people.

They're going to be made with real people. They're gonna be made with cameras. They're gonna be made without cameras, and there'll be a roll for everyone. But there's no question that the million dollar a minute film budget, which is what it is today in Hollywood, about a million a minute, could be a dollar a minute, and still have a hit.

And it'll look Hollywood quality within three years by 2030. It's interesting that Roku, CEO, at CES recently said the same thing, said within three years there will be a hit that costs a hundred bucks to make no question fully on a screen like I have here.

David Rice: Well, Kevin, thanks for coming on today. Appreciate it. Enjoyed the discussion.

Kevin Surace: Yeah, fun stuff. It's a fun topic.

David Rice: All right. Well, listeners if you haven't done so already, head on over to peoplemanagingpeople.com/subscribe. Get subscribed to the newsletter, get updates on all our upcoming events, podcasts, everything that we're doing here, all the latest articles that we're writing.

So get ahold of that and until next time, just learn how to do what you do better.