Your Team's Already Afraid of AI. Now What? with Paco Farias

0:00 Your team's already afraid of AI. Now what? With today's amazing special guest, Paco Farias. Now, Paco, I'm super excited to have you here. You're just such an expert in the space. And before we dive deep into our conversation, I always want to give listeners like a clear roadmap of the incredible value about to receive. For those tuning in right now, what are they going to learn over the next 20 minutes or so? And why do they absolutely need to listen to this episode right now?

1:21 Well, thank you for having here ~ excited about the conversation. I would say a couple of quick things is number one, there's a lot of hype around AI and all of the human impact it's it's having and it's gonna have. So I believe that we're gonna be able to have a a lot of very insightful conversations and aha moments for everyone to get ready so that they don't get caught in the A lot of people who are into AI like me

1:44 just assume everyone else is using it, that every company's fully integrated, but that's not really true. There are a lot of companies out there that know AI is coming, but they haven't really started. And for those companies that haven't jumped in the water yet, where does that culture problem actually live? I would say in the founders, I would say that's where it starts and where it ends, to be honest, or the CEOs or the leadership team, let's just say. because in most companies,

2:10 if they don't take action, even though they may be expressing, doing marketing, doing, you know, all of these cultural pamphlets and whatnot, if they don't really understand, don't really practice, don't really wanna get into it, it's not gonna happen. What's the earliest sign that a team is quietly resisting AI rather than adopting it? I would say a couple of quick quick quick ones. The first one is Believing that AI is just Chat GPT, or it's just copilot, or it's just prompting, asking questions and getting answers.

2:45 So I would say most people that I've known so far on the corporate and non-corporate ~ sphere that you talk to, they're like, yeah, I know AI, I'm using it daily. Okay, good, good. What is it? What are you doing? yeah, I'm asking questions, I'm getting responses. So I would say, yes, that's you know. Step one, or step zero probably of trying to to test it out. But whenever you know you really talk about adoption and and making meaningful changes,

3:13 real changes, a lot of companies just stay there. and a lot of that has to do with people being afraid. I would say that's one of the biggest components. ~ it could be either the conscious or the subconscious. being afraid where they might be even hijacking a little bit because they wanna either again consciously or subconsciously maybe stop a little bit of the change or try to avoid that progression instead of really getting out of their comfort zone and getting you know jumping right in.

3:43 I d my favorite analogy there is in the 2000s. You know, you and I were there and and several of our audience members as well. When we had to do a massive change, rapid change into Wordpress processors or you know ~ spreadsheets and things like that and email. but now it's exponential. So a lot of people are not wanting to do that that ~ that move and they are masking it by saying, yeah, I'm using AI. No, they're not.

4:09 I mean they are, but they are just you know, they're not learning how to prompt, they're not learning how to do, you know, all of the more advanced features. I'm glad you brought up the issue of fear because I think that's a really big part of it. Because a lot of the marketing language from AI is replace, replace, replace. And when I hear replace as an employee, I hear fired. And when you're kind of looking at a rollout or bringing

4:29 AI on board with one of the larger companies you work with, how do you calibrate or tell the difference between healthy skepticism and fear that will actually kill the rollout or stop the project from The complexity here is that everyone listening to us has to make a conscious Their conscious choice is suspending their beliefs for a little bit at least to be more open to understanding and expanding their awareness. Why am I saying that? It's because whenever you're either on the leading

5:02 of a massive company change or you're on the receiving end. Both sides have to have a much broader understanding and openness on how to adapt themselves to the technology to embrace it rather than getting into a defeatist attitude. So that's the biggest, in my opinion, that's the most important thing is changing and opening your A little bit like like, you know, when you when you have the horses that are like this, right? it's trying to either remove the patches or open them up.

5:34 I know it's difficult and I know it creates a lot of issues because the media, and we know this the media loves to portray a very, very specific ~ set of information and the algorithms just exploded. So, yes, whenever someone hears AI adoption, new changes, cultural transformation, any of those things, automatically everybody falls into the out direct trap of layoffs. Of course. and why not, right? Because everybody likes to see the big layoffs from the tech companies. Well, in reality,

6:04 I urge everybody to do a deep dive and understand. And in reality, as of today, it's if not the opposite, at least it stays like this, you know, in terms of headcount. So no, it's not really going that way in all companies. because a lot of companies are starting to understand. That they swung to the pendulum too fast, too hard. And they are trying to, you know, there's quite a few companies that already fired a bunch of people and they're trying to rehire

6:30 them because they moved too fast, right? They they didn't understand what was happening. so yes, it there I know it there's a lot going on, and I know there's an old noise. I know that a lot of people are trying to figure things out. I would say do your own deep dive, utilize AI to your advantage. You can have conversations with AI to Help you further your critical thinking about this. You can ask any AI almost now to do deep research for you and understand what

6:58 are the implications of any of these changes and how to tackle them properly. Right. So instead of being absorbed by the AI and just letting the doom come at you, you know, utilize it to your advantage. That would be my my biggest takeaway for anybody, right? utilize the AI to your advantage. It can be your greatest friend, I can tell you. For sure You brought up something really interesting happening, which is that we're seeing rehires when it turns out AI can't

7:23 do what companies thought exactly it could do. Why do so many AI rollets get a huge announcement and a big launch and then six months later they quietly die? I would say ~ my my strongest assumption as of now is that boards ~ board of directors and leadership teams really wanna move as fast as possible on this hype to show early adoption, I would say, or early understanding ~ that they're moving very, very fast to express quote unquote possible signals to the market.

7:55 or to their investors or whomever, right? The stakeholders. And because you know, there's many articles and you you can see it everywhere that board of directors, one of the key continuous conversations they're having is AI. Now here's the fun part and that's what we have been talking about so far. If you really double click on the awareness and understanding of those in the board of directors seats and the decision making seats, If they actually understand what AI is, if they understand how that change

8:23 adoption needs to happen, how those change management processes need to happen, most of them have no clue because they're thinking about it as, you know, it's a new new technology, new wave, let's just adapt. People will know how to do it. it's a resource allocation issue, let's just get it done. But what people don't understand, it's that we in my we we really believe that AI has gotten to a hyper smart space across all domains. And that would be really a AGI,

8:53 right? And it's not. It's really not. It's it cannot hold that intelligence for a prolonged period of time. it continues to hallucinate unless you have it on a very, very well trained, you know, very delimited you know sandbox. So bottom line is there's a lot of the assumptions that Borders of Directors C suite teams And other stakeholders are believing that the AI frontier models or the labs are already at the cutting edge of everything. And they're really understanding that they're not.

9:24 They are at certain parts of it. you know, can they beat one of the top chess players in certain parts of their game if you train that specific AI for a while? Yes. But only that. It doesn't mean that it can do protein decoupling, or it doesn't mean that it can do advanced risk assessments or whatever you want, right? In in any domain. So that I I believe that's one of the biggest issues they believe that just by deleting people out of payroll,

9:52 the systems magically are gonna be overtaken by AI with no hallucination and no risk. And they're very quickly and rapidly understanding that it's going almost the opposite route. It's creating even more problems because they might not have been feeding the models with enough time, enough data, so that it could make the best possible decisions because L LMs, we know this well. They're just trying to predict stuff. And in order to predict, you know, to the best possible way, you need data.

10:21 You need to train them. And they're not training them enough. Yeah, you bring up so something really interesting to me, which is that we keep thinking this is the model that's going to change everything. And three days later, a new model gets announced. There's always a new point one, point two, different name, whether it's named after like a musical note or planet, they're always changing the name and conventions. And to me, that's how we know that it's the opposite.

10:42 And one of the worries is you can train your entire system on an AI model. By the time you've mastered that version of Chat GPT or Claude or whichever model you're using, it's no longer active. As soon, like as soon as I post a review, I try to post a review same day a model comes out. There's a new model half a day or two days later and it's forgotten. Sometimes people go, using Chat GPT, gross. I stopped using it three days ago.

11:04 And a week later they're like, Yeah. You're not using the new chat GPT. What's wrong with you? It's the new cool one. And it's the it's like being in high school again. Like what's cool changes so fast. ~ would you say that this is the most expensive companies make when they try to bring AI into the workflow, or is there a bigger mistake that they're making? question. I would say probably the answer is both and why? Because well, for since companies have been you know around,

11:29 for most companies the largest expense by far is people. You know, for most companies. Having said that, I'm really thinking now because you know just think about the leak from AWS that they spent five hundred million dollars in tokens in one month. So, you know, when you start thinking about it, then tokens are gonna become the next big issue. and that's the other double click that we need to be very aware of. Because when you are you know, one of the other issues that CTOs are

11:56 starting to really understand is they have no idea of the token utilization. They just don't, because a lot of people believed that when you, you know, when you got People on an enterprise plan, yeah, yeah, you know, they're gonna get X amount of tokens who are good. No, no, no. You know, you get more and more and more your token utilization. And the funny part of it is if you ask people, I would you know what? I'm gonna do a survey on LinkedIn after this,

12:26 asking people to be honest on you know, on what they are actually using their company tokens for. at least verbally, a lot of people have told me they're asking all sorts of ridiculous questions. Because they believe that it's unlimited, you know, they already paid for an unlimited amount of tokens. They ask about, you know, predict who the next World Cup you know leader is going to be, rather than doing work related things. So, yes, token utilization is going to be a massive,

12:54 massive expenditure. It's already there. ~ and I believe it's gonna even surpass human cost pretty quickly if it's not really contained. You brought up something that I think is really brilliant. When you have an employee, it's a fixed cost. They ask for a raise once a year and they're not gonna ask for double. Very few employees have the confidence at the end of the year to say, What kind of raise do you want? Double me. Whereas tokens

13:18 You do not know if you're on any type of subscription plan, they never tell you how many tokens they get. So you it's kind of this black box of like, I get a certain amount. It seems like I'm getting less, but they don't tell you because they don't want you to be able prove that they're giving you less. And the token costs are going up. And I've even heard companies using words like token maxing, which whenever I hear the word maxing,

13:36 I know something bad is about to happen. Anything you max is bad. I've never heard something maxing, it turns out to be good. And there's these companies that say you have to spend a certain number of tokens or you're in trouble. So now I think they're incentivizing inefficiency. It's the same problem I see where you pay a person to work 40 hours rather than by the result quality of the results. If you pay me to for to work a certain hour,

13:58 then that's what I'm going to max. If you pay me to spend number tokens, I'm going to spend that many tokens. If you pay me to get as many things done and that's all that matters is project quality, then I'm going to do that as fast as efficient as I can because I want to be maximize my time efficiency. So I think that we're seeing this as the these sudden bills come out, these companies realize, whoa, these tokens are really expensive and it's a non-fixed cost.

14:19 You can have a surprise where they just go, hey, the prices of our tokens are double today, and you're beholden because you're in the middle of a project, you have to pay the new price. I think companies are going start to realize that all of this talk about how human capital is so expensive, at least it's at least you know what you're paying. And I think that's really important. ~ how much do you think? token efficiency is going to become kind of the frontier market.

14:42 I think that the secondary market is really going to become any software or tool that just says, hey, we can cut your token usage and have. And that becomes like this new, I really see that and cybersecurity are as the two frontiers over the next one to two years. I I think that's a very smart prediction. ~ another one I would like to venture is that LLMs are becoming so commoditized. you know, they're all advancing so fast and and trying to par the edge cases

15:10 of utilization of the LLMs are just for the upper five percent of users, right? Or whatever. It's a very small amount. Everyone else is just using it for the same stuff. So what I'm trying to get into is I believe or one of my my predictions is going to be that it could happen like what happened with Netflix and Hulu and all of those or any of the social media platforms where instead of charging you for the utilization of the model for the average

15:38 they're just gonna go the advertising route. And you know, ChatGPT is already doing it, right? And think about it. If they go the advertising route and they do it quote quote for free, what's gonna happen is yes, you you may still have a token route on on certain parts, right? but it's gonna become so commoditized that A, yes, there's gonna be specialized suppliers that are gonna help you with the most important risk assessments, cyber, of course, being probably the number the most important one.

16:05 but it's just gonna be fascinating for me to see how do they manage this because with all of the open sourced models that are becoming so good so fast, why? Do you want to pay for stuff that people are just spending money? you know, so yeah, it's gonna be very interesting the the balance of when do you really wanna spend the big dollars on something versus just have the average for everyone else? And and that's where I believe the the

16:31 edge cases will come and these companies really have to start getting smart about it. Yeah, I really find that some of the most to me obvious decisions to make, like I've always said there should be automated routing. So when you ask a question of Chat GPT, it routes it to the model that can answer the question best. But they make Yes. you choose. And they I've been saying this for two years. And that's insane. That's I don't know

16:54 which AI knows the answer until I ask it. So we have a lot of people asking the most expensive model. Questions are the cheapest. For certain tasks that I have, the smaller models are better because they don't get creative. They just do the task specifically because they don't have enough intelligence to drift. And I'm always looking for models that are like small, medium, and large for different types of tasks. And there's we've started to see some of the in-between brands.

17:19 Like Open Router has created their own automatic router thing. And I think it's very smart, but these companies aren't doing it. And I think they're not doing it because they want to make you spend as many tokens as possible, which is very silly because the Chinese models are starting to really catch up. Like the it used to be that it was like the American models were maybe six months ahead of the Chinese models. Now it's three to four weeks. And we've even seen a model models

17:39 released now where it's three to four They're catching up within a couple of days. So you can get something that's 99% as good for one hundred like one hundredth of a cost. Of course, you're going to switch. As companies get smarter, they're going to start looking at these things. And I think that's really where we're going to start to see the AI race go. And so a lot of companies are chasing the like they're trying to chase. It's like we're fighting the last war.

18:03 We're so busy thinking about. Token maxing when it's like, no, token efficiency in routing is the future because the company that can get the same task done for less money can win in any market. Right. If I can do the same task as you but for half the price, then I can charge two thirds the price and still be profitable and I can destroy my market. That's what Blockbuster did, and then that's what Netflix did to Blockbuster.

18:22 Yeah, I think the ~ the on top of what you're saying, you know, the the next layer, the intelligence layer, and the specialist layer is probably where a lot of these companies are gonna sit. because again, it might change everything's changing, but the basic layer seems to me that it's gonna get commoditized very quickly, all the way to free. and So it's gonna be fascinating to see where do they wan specialize. it seems to me, you know, for example,

18:51 becoming like the ERPs or the CRMs and all all of those complex, enterprise wide, cyber security heavy applications. So the application layer, that's where a lot of these companies are gonna go, or becoming that. Because I agree with you. It's and a lot of companies also, you know, I'm I'm working with some friends on a company like that right now, where our intention is to go the ~ the non the non-cloud route, right? So you wanna go local for many,

19:22 many, many different reasons. so it's it's gonna be so interesting to see that and and and see where it goes. But yeah, I agree with you. It's lots of noise, lots of movement, but I believe people just need to be breathed. continuously you know, read and and and and find ways to you know scheme through the noise, there's an AI can actually help them do that. You know? I think you're so right about the shift to local. Like we're seeing all of this push to

19:45 build these massive data centers. And within three years, we're gonna see you all videos all the world of like how to turn a data center into a hotel, how to turn a data center into apartments, because they're gonna be useless. Ha. We seem to have forgotten. That computers were huge fifty years ago. They were the size of a room. Now they're the size of a phone. We've forgotten that our greatest expertise as humans is miniaturization. And if you watch any movie about the future,

20:11 when someone's talking to an AI, it's in their phone. They're not talking to their phone to an AI that's far away. That's not really useful. So I love that we're talking about the security, and then you can control costs because you only have to pay for the equipment once. And you kind of have the securities built in. And we used to have intranets and have much higher level security. We used to be very strict about what would get in or out.

20:33 Now we just try to do blocking. We've reversed our security posture, which is much worse. this is so So useful. The kind of last thing I want to dive into is we have a lot of companies. Everyone wants to say they're AI native. And there's a lot of companies that say we're AI native or we're an AI first company, but they haven't actually changed anything about how work actually happens. They've just kind of made a branding decision. What's broken?

20:54 Well, first of all, I truly believe that a lot of people don't even understand what that means. as easy as that. So let's just start with the the realization that a lot of people just truly do not understand what that really means. because it is i I I'm learning, my you know, I I I have enough awareness to understand that I believe I can grasp what it is, but I am not You know, the the most knowledgeable person in the world or a researcher

21:20 to get all of the details of what that means. So there's a lot of hoveries. There's a lot of people being, you know, their egos and and other things are getting in the way of really going against their curiosity and open-mindedness. where, yes, in the culture side, one of the key things of becoming AI native is that you have to teach people AI is not the enemy. AI is the enabler. AI is your body, is your help. You know, it's gonna make

21:45 you a thousand times more productive if you take the time to understand how to enable it. So it's it's a significant effort going against your ego, going against even your survival instincts as human, because our brains, you know, are wired to survive. So some people are like, no, no, this thing is gonna go against me. I'm not gonna teach it, right? Instead of okay, this thing is gonna become my my neighbor. So yes, it's it's it's a lot of ~ cultural shift that needs to happen.

22:18 And and in my opinion, on top, of course, of the real AI layers that you have to create when you're a native AI native, so the technical components are you know crucial. But on the culture side, I would say are equally important because you could have a genius CTO, a genius co-founder. that just does everything for you. but if the team does not even understand how to leverage that AI nativeness, they're gonna continue reverting to their usual, you know, normal company, you know,

22:49 operation. So it's it's a significant reboot of their human operating systems, right? Yeah, I love that you talk about that and focus on the human part of the AI transition. I think that's so important to pay attention to staff morale and how to get these programs adopted because a lot of these things fail because someone thinks if it succeeds, I'm going to lose my job. So I love the things you're working on. And I really appreciate everything we've talked about today.

23:13 For the listeners who want to follow your work or dig deeper into what we talked about. Where should they go to find you online? Where can they see some of the work you're doing and some of the projects you're working on? And what should they do when they get there? Absolutely. well, the easiest one is my LinkedIn account. I've been doing a lot of work on that. So just look for Paco Farias, or I'm sure they're gonna be able to click on

23:33 a link ~ here in the description. So that's the one I try to be the most active on. I'm starting to ramp up my activity on X as well with Paco Faria C. So any any of the social media that's the easiest and the fastest. ~ you know, I I'm working on a couple of companies. One of them is Glinks, GI D-L-I-N-X.ai, sorry for pronunciation, glinks.ai. So getting there. we're we're definitely building something super interesting on the intelligence layer on the local side.

24:02 So you guys can follow our work there and and are gonna be quite surprised. And yeah, so ~ you know, following through that, ~ we're gonna do lots of announcements. I love interacting with with the team, I love helping ~ other humans as much as possible because I wanna learn and through interaction that's how you learn the most. Amazing. I'll be sure to keep following you myself as well. I love the stuff that you're working on. And thank you so much for being here today

24:23 for another amazing episode of the Artificial Intelligence Podcast. Thanks so much.