Artificial Intelligence Podcast: ChatGPT, Claude, Midjourney and all other AI Tools

Is Artificial Intelligence Changing the Role of the Modern CTO with Sergio Pereria

Jonathan Green : Artificial Intelligence Expert and Author of ChatGPT Profits Episode 347

Welcome to the Artificial Intelligence Podcast with Jonathan Green! In this episode, we delve into the transformative role of AI in the tech industry, specifically its impact on modern CTOs, with our insightful guest, Sergio Pereria. Sergio, with a wealth of experience as a fractional CTO, offers a unique lens on how AI is reshaping the landscape for software developers and technology leaders alike.

Sergio dispels the myth that AI is a threat to coders, highlighting instead how tools like GitHub Copilot and cursor enhance productivity and efficiency. He explains the shift in the development process, emphasizing the importance of purposeful instructions and the new opportunities for non-technical stakeholders to produce software.

Notable Quotes:

  • "I think AI is changing everything. It's not necessarily for better or for worse; it's just changing everything." - [Sergio Pereria] 
  • "In a startup, you just know when you start that you will need to change things as you go." - [Sergio Pereria] 
  • "There are people stuck in doing things a certain way, and the danger is thinking, 'AI's gonna replace me, so I'm gonna try and avoid it for as long as possible.'" - [Jonathan Green]

Connect with Sergio Pereria:

Twitter: https://x.com/SergioRocks

LinkedIn: https://www.linkedin.com/in/sergiomcpereira/

Connect with Jonathan Green

Is artificial intelligence changing the role of the modern CTO? Let's find out with today's special guest, Sergio Pereria. Welcome to the Artificial Intelligence Podcast, where we make AI simple, practical, and accessible for small business owners and leaders. Forget the complicated Tech talk or expensive consultants. This is where you'll learn how to implement AI strategies that are easy to understand and can make a big impact for your business. The Artificial Intelligence Podcast is brought to you by fraction, a IO, the trusted partner for AI Digital transformation. At fraction a IO, we help small and medium sized businesses boost revenue by eliminating time wasting non-revenue generating tasks that frustrate your team. With our custom AI bots, tools and automations, we make it easy to shift your team's focus to the task. That matter most. Driving growth and results, we guide you through a smooth, seamless transition to ai, ensuring you avoid policy mistakes and invest in the tools that truly deliver value. Don't get left behind. Let fraction aio help you. Stay ahead in today's AI driven world. Learn more. Get started. Fraction aio.com. Sergio, I'm so glad to have you here today because for a while there was this kind of idea that AI and TTOs were dividing that. AI users were artists or creatives, and that CTOs were technical and lost in the past. But now we have all of these tools. Every day there's a new AI coder, and every day there's this new revolutionary AI language, and we keep hearing the same phrasing. It used to be that AI was gonna replace your workers. Now it's that AI's gonna replace your coders. And that's where I wanna start from. Like, how worried should coders be about their careers? Are they about to get replaced? Hey Jonathan. Thanks for having me and thanks for leading with such a great question. I think AI is changing everything. It's not necessarily for better. or for worse? I think it is just changing everything for coders now, we have cursor, we have GitHub, copilot we have all these amazing tools that allow us to ultimately transform completely to way we produce software code we use to write all the Eves and and the for loops and all this stuff. Now we, the tab key. We need to be more, much more purposeful about the instructions. We give much more purposeful about reviewing the AI's code suggestions. But it ultimately makes software developers much more productive, much more efficient. So I think for anyone listening to these is a software developer, I think it's an exciting time. You can do more in, in less . Time, you can be much more efficient and much more productive now from a business perspective. Yeah, in a sense, these tools also empower non-technical stakeholders to produce software codes without the coders or with fewer coders. So there is an angle of. Let's say threats to, to the software engineering role. So I think with all big change that comes from technology innovations, there are threats to how we used to do work and opportunities to do the work in more efficient ways. So I think for any software engineer looking into this, and for myself as well, I'm much more looking forward to how software is gonna be written in the years to come. And I'm adapting my work and my teams to that rather than just, grieving about what, what changed.'cause I think it's just exciting opportunity all along. I think that's the critical element. It's not like replacing coders, but there are a lot of people who get stuck in doing things a certain way and it's that difficulty of learning the new way. I've always done this way, and I think that's the danger advanced coders have, is that these new tools change their workflow and it's with anyone, once you've been doing something for five or 10 years, completely changing your workflow can be really difficult. We saw the same thing with the advent of the typewriter instead of writing with a quill and with the advent of using a computer that could delete easily rather than a typewriter. So as we've changed technology, each time we have these transitions, there are people that get left behind. Not so much because the technology replaces them, but because they struggle to learn this new way of doing things. My kids can't believe there's such thing as a non-touch screen. So they're already in a world where everything is commanded by touch. Why would you use a keyboard? So we always hit these technological hurdles and I think that the danger for coders and for anyone is to think AI's gonna replace me. So I'm gonna try and avoid it for as long as possible. And that's where you end up at a disadvantage and for a lot of non-technical founders. And it's the chance to, better explain what you want. When you hire an outsource coding team, or you bring on that first CTO or you bring on the developer team, the better you can explain what you want and the more you test it, the faster and the less money you'll waste. Because I can tell you so many projects I've worked on, once you build it, they look at it and go, now that I've seen what you built, I realize I want something completely different. And that ends up exactly massively spending the time. And have you had experiences like that? Absolutely. That's pretty much the core of my work as a fractional CTO. So I usually I'm hired by very early stage startups. Usually at the stage there's just the founder or the founding team, and very few people besides that. So I'm very often the first technical person in, or, one of the first. And I'm really the right hand man of these startup founders to build their product and build the foundations of their engineering teams. And you're right on. I see a lot of companies, especially big ones, like enterprise companies, very concerned about ai. It's not just the individuals who work there. Many of those individuals would actually begin on trying these tools. It's actually the companies, many companies prohibiting these tools like banning Chat, GPT and GitHub copilot and stuff like that from their VPNs, like you cannot, that is grounds for termination. It goes without saying that these companies will be left behind in the next decade or whatever. A number of years ahead. They will be all these new business coming along. Composed by startups, usually very small teams, startups that just chip way faster. They build product much faster. They iterate much faster. They serve their clients better. And the way they do it is that they use these the best tools available. They empower the people who work there to, to use the tools extensively to the best of their abilities and just ship code faster. Produce, the outcomes faster and just iterate faster because in a startup you just know when you start that you will need to change things as you go. You will very likely need to adapt your product, pivot your proposition change the client segment that you are targeting, and just evolve the product as you go with each of these changes. And that means you need to be nimble, you need to be fast, and using these tools makes you much, much faster in each of these situations. So I, I think in my perspective. Of course I might be biased because I worked with startups every day and I used these tools to build their products faster every day. But I think these startups that are being founded today this year they will win in the years to come against these existing companies that just fight such major innovations that are up to up for the grabs. Where do you think a lot of companies get stuck? When they're trying to navigate a technologically, we saw this happen 25 years ago when the internet suddenly became worldwide and accessible. Every company said, I need a website. Why? I don't know. I need a MySpace. Why exactly? I don't know. But my nephew said, I need one. And now we're seeing every company says, I need ai. Why?'cause it's on the news all the time. I hear about it all the time. Every government's talking about it. We get so caught up in the hype we don't think about, is it? It's not practical and useful in every single individual use case. And it's about deploying as with any tool efficiently. That's where you get the maximum use case. So when you're building a plan, when you're working on a new project or with a new CEO or a new founder, how do you strategize where to implement these tools and where to use the classic ways or the more manual processes? Ex, exactly. Like one thing I see a lot is companies want to build chatbots for some reason. There's a lot of people who got trapped or stuck in the okay, generative AI is a chatbot. I need a chat bot. Trained on my data and, fine tuned for my specific use case. I mean nothing against it. I've built a bunch of those and for many use cases that makes sense. Like for, customer support, internal tools and whatever. But I feel like there's an over indexation to the chatbot use case. Not everything can be solved with chatbot and no amount of LLMs and training data will be a hundred percent a replacement of. Sometimes a human team that operates on very vague information and not very written down guidelines, I think, as opposed to that one, one. Sort of architectural pattern that has been under explored over at least the past year or two, is fully agentic workflows using LMS for parts of the decision making in a workflow but putting these different parts together in a way that. At the agentic workflow collaborates towards an end. I think now that's become becoming much more trendy, but I don't think people fully understand, or at least many people talking about that fully understand. For example, I can give you some specific examples from my own. Just over the past year, I built a job copilot, which is an agentic job search agent. So you just put your resume or LinkedIn. And it goes out and finds jobs for you and applies on your behalf. A lot of people a lot of customers there are getting interviews getting hired and they simply uploaded their LinkedIn, their cv to this agent, and the agent went out to, to find jobs for them and to apply. So they just get the scheduling invites for the interviews on their email. It's quite amazing. I have also built marketing agency composed of AI agents for a client an e-commerce group out of Austin at Texas. And they basically we built the different agencies with different roles. Like one does the market research, one does the strategization of the. What the ad should look like. What is the channel we using? And then there's a bunch of agents which are copywriters for different channels like email marketing, Facebook, TikTok, and so on and so forth. Which is quite amazing to see like how you can build this with technology. A few years ago this would not be possible. Now it is not only possible, but just a huge, massive opportunity for businesses out there. So I think to the question if, businesses are, leveraging AI enough or too much I think they are over-indexing to some things that are, might not be the perfect fit for them and overlooking other things that could be game changer in their industries. Like I'm seeing every day with the startups I work with as a fractional CTO. I think that when a lot of people hear agent or agentic, they just think smarter chat bott or trained chat bot. And one of the challenges that a lot of the language word an agent is used in so many different ways in AI from a simple chat bot all the way to something fully autonomous. And I found that a lot of people, when they say they want an ai, they really want an automation where an AI is a small component and that most people. The way they're using AI right now, it's a lot of copy and paste. You take the task you copied into the ai, it gives you an answer. You copy into the next thing and the next evolution. I think you're right in agentic and what's happening this year is to merge that with automations into an agent. Where it does the whole task without having to copy and paste. We're seeing this now where first they gave Claude access to your computer and now chat. GPT has the ability to move your keyboard and mouse and do things. And obviously this takes a great deal of trust and one of the challenges. With AI is that it's so exciting. I've seen a lot of founders jump past the compliance or the security phase because they're so excited and they don't think about how using different ais, like a lot of people are really excited by some of Theis coming out of China that are super cheap, and I always say if something's really cheap, there's, you're just not seeing the payment. There's something else going on, right? If it seems free, Facebook seems free until you start seeing all the ads, there's always another element. They have to fund their business some way, so there's something going on, but. One of the challenges I face with some of the projects clients I work on is that it's very hard to explain that you can use AI without breaching your security. Like they either go too far in one direction, like super secure and never use AI or the other direction of let's open the doors, feed all of our customer data and just see what happens. Yeah, that's a big risk. And I think you're right, many founders and even teams, like in general, I think on a micro level, product managers of some initiatives in bigger companies all these stakeholders who are typically non-technical backgrounds. They get very excited. They want to build something really fast. They use all these tools. And I think there's a risk of these initiatives skipping the typical engineering and technology stakeholders. Because indeed, like if you are training an LLM. On all the sensitive data in your business, like the code base, the client data, all this stuff. There's a catch to that. If that data leaks, you are immediately out of business. There will be court cases that you cannot really win and you'll be out of business in no time. I think there's a huge opportunity. There are also significant risks to account for, especially. How to handle the data because if the data is so good to train the LLM and make it great, then it, if it's lost it's definitely a big deal. The that, there will be liabilities on that. So yeah, I think companies and founders should never skip doing things properly, like securing the data and building the guardrails for. Whatever LLM integration they build either using an API of a provider or using self deploying an LLM on their own infrastructure or using some infrastructure vendor for that. Now we have such a rich market for all of these. We have so many providers competing for each of these different use cases. So I think depending on what the actual requirements are. I can immediately tell what is the right option and what is the cost structure for that. I think though these risks should never undermine actually going for it and developing a roadmap of ai. And especially LLM assisted workflows. You're right on. We said chat bots and agents and people listening to these might assume that any ai tool or LLM empowered. Workflow needs to be client facing and promoted as an AI for that, for this or for that? Not at all. I'm doing I'm having tremendous technical success and taking some really interesting breakthroughs in use cases where LLM is really down, under really under the hood in some clunky part of a use case. And the company, if you look at the website or whatever. You cannot really tell if that company is even using LLMs or AI or whatever, because it's just a part, it's just a cog in the machine. And I'm talking things like for instance data cleaning, like ETL type pipelines where there's a lot of unstructured data coming in and we use just LLM to LLM like a, a self hosted LLM to. Clean the data, find patterns, cluster, the data really shiply really nicely really immediately, like low latency stuff. It's amazing. Like this work would either not get done at all, which would cause poorer output. Or it would be done by human, like it would be done like next day because this team is is sleeping right now or whatever. So now we have this 24 7 really sheep cog in the machine that does this piece of work really well. So not every LLM based product needs to be outspoken as AI for these, ai for that, sometimes these AI tools are just a cog in a machine. Either agent workflows or just a wrapper on top of G-P-T-A-P-I kind of thing. I think you have brought up something really important, which is that the be most of the best use cases for AI are not customer facing. They're interior, so we see a lot of things. A lot of the clients I work with, for example, like one department needs data and their customer support, and then they have to message the tech team to find out how long it'll take to fix the problem. And then the tech team gives them a very complicated answer or a link to their card and their project manager. And then the sports team doesn't understand what it means. So you have this translation problem. So one of the projects we worked on was let's build a bot that can translate all the tech team jargon and tell you based on like how long it'll take to fix and what it means and what you should tell to the customer. Because this little breakdown in communication means that two, two things happen. The customer support team is annoyed 'cause they keep having to ask what do you, what means? What does what, how? I'm sure you're like this. Every coder I've ever worked with hates when you break their workflow. Like when they're in the zone, there's nothing they want less than to get disturbed. So two people are getting frustrated and there's a simple solution, which is to put an AI in the middle just to keep track of the data. And I love what you talk, 'cause I think data organization is so important. Most companies are so obsessed with collecting data, but what do they do with it? You have these giant piles and it's not useful. It's like buying a thousand books, but not reading them. Being able to organize the data is the first step into usefulness, and that often gets missed. Or whether it's filing the data or I see it, there's a lot of problems with record merging. Someone calls in, does a sales call, then they call into customer support. Two separate records are created and it's the beginning of a problem. And a lot of companies actually use two separate systems for tracking the customer. So now when the customer calls in, I've worked in customer support, I've worked in sales, you're calling the other department to try and find the data. And if we could just merge the data right, have a single point of truth, it would make it so much easier. I have this philosophy I say to my clients, and I wanna get to a point where you never have to ask someone else at the company when you're looking for something. If you can get to that point where you don't have to be annoyed, you're looking for it and be annoyed when someone bothers you to ask for something that should be easy to find. That's when you achieve the first level of efficiency. I. Exactly and you are, you're right on. One of the biggest I think not so much talked about use cases for LLM type workflows is actually making data human readable. For instance, if I'm working with either split records across the spreadsheet or Just a huge J file that is, just busy working lot of fields, just making that available for humans to read. Usually you would build a dashboard and map one field to which, filled in dashboard and it's like a list of, 50, a hundred, 200 fields. And even that, even if it's like a pleasant ui, it's just a lot of data or like a spreadsheet with a hundred columns. It's humans have gotten. Used to this kind of stuff, but that's not how it should be. Because what the human will be doing is crawling the eyes through all of that data, finding patterns. You can actually use an an LLM type workflow to find the patterns on those things. To brush these kind of, lots of data clunky outputs, and just give the human what he's already looking for, or even take a next step in the workflow okay, you found the patterns, you found the figures, or whatever. You can actually feed that in the next workflow action. For example. Imagine you have usage from customers. That is just a huge loss. Like it's really hard for human to understand, even for one customer what's going on, but then you have 10,000 customers. You can actually have LM par through all these and find signals to feed to customer support team yeah. I think that it's so critical to. Find the right data.'cause the other direction you can go in is there's too much data. Sometimes you have so many dashboards, it doesn't fit on a single monitor. Now you have two monitors and three monitors, and sometimes it feels like going from a car to an airplane. There's a lot of switches, but there's too many to keep track of unless you go through years and years of training. And this is where. Using the data efficiently goes to that next level, which is now we have all the data, we have it organized. How can we simplify it to what's useful? What does someone need to know? Because when someone calls into customer support, exactly. If you know what, if they've called in this week, what their previous problems were, where they, have they ever logged into the website? Whenever I get a customer support issue, that data is so important. Have they ever actually logged in? Have they gone through the training? When they say the thing doesn't work, did they ever try? Did they ever open the instruction made or did they talk to the chat bot? If, you know they tried eight times, it's a different conversation than if they've never tried and it's a different problem. When I first started business 10 years ago, the most common customer support request I got was I never got my login details and they'd misspelled their email address. 90% of my customer support requests were people misspelling their own email address. That's gone down a lot recently, but for a long time that was a big problem and it's really hard to solve it 'cause you're emailing them all the answers. Here's a login, here's your password, but it's the address. So sometimes. A tiny thing can cause a huge problem for the thing I think that's really beautiful right now is that you can see if an idea is a good idea in a much shorter timeframe. One of the things that I've seen destroy a lot of businesses is when the founder gets in love with an idea before they find out if the market wants it. When you see this founder and they spent like dollars on patents. Got a deal from a factory. So they printed a hundred thousand units and then now they have a warehouse they're renting. And how many of you sold none? So you don't know if anyone wants it. And that's the worst situation to be in. The beauty of minimum viable product. Not just that you could do it fast with ai, but that you can do it before you develop that emotional attachment. Which causes poor decision making.'cause that's really the scariest thing when someone's in love with an idea and it's a bad idea, but they're they've spent so much time on it.'cause you don't wanna say I've wasted six months or two years of my life. Exactly. I see that a lot with first time founders, to be honest. And sometimes they fell in love with like B two B use cases. Like they see the end group product they have in mind. They build it, they go to market and they spend a lot of money like. Ads and customer campaigns and influencer marketing and all this stuff, just to find out that no one's really willing to pay for that because and then they pivot to a B2B proposition. Like they work with the tools that people are already using and then they succeed with those. What that means in terms of product is that under the hood it might be the same thing. But like it's really different. Build a client facing app, let's say to a B two, BAPI that you expose to integrate with B2B clients, sometimes enterprise companies. So even the sales to that is fundamentally different. Instead of spending a. On, on ads, you will do B2B, like outreach, you go to business events in the field, and so on and so forth. So I always tell founders to do this exploration as early as they can. And you're right. If you, if we can build an MVP in one or two months rather than six months or a hundred like 50 k, a hundred k deals with an agency, and if we can build it with 10 or 20 K or whatever. We are in a match bear position to find out what works, find out what does not work, and iterate to the next thing. Because in a startup, it's a game of iterations. The more iterations you can have, the more likely you are to find success. If you spend too much money and too much time in iteration number one. You are very likely to depend on the success of that first situation because you'll not afford to take a next one. So this is, I think, a mindset that many times, first time founders are not really equipped with because as you say, they are so in love with their idea that they don't even. They don't ev they don't even account for the possibility of it failing and well, numbers are what they are, 90 plus percent of startups fail in the first year. So I think the biggest challenge is to not be one of them and to be the 10% and then the 1% that actually not only doesn't die, but actually succeeds and thrives. Yeah. I think you brought up something really important. There's this mistake a lot of people make from the outside, which is that I. Everyone else's first idea was their winner. That the first thing they tried, the very first effort. And so they expect their first try to work and it's, I wish that were true, but my best idea was like my 70th idea and my hundred 50th idea and all the ones in between for failures. One of my favorite questions to ask really successful peoples, how many good ideas do you have per year? And they always say one or two. No one has ever said three to me, and that's an important lesson to realize that. The odds of your first idea of being a good one are not very high. Having the ability to iterate, having the ability to pivot both emotionally and financially, so you haven't gone too far down that path, that's what really that agility is what really is the advantage smaller companies have right now. So I think you're dead on. I think this has been really amazing. I know our listeners are gonna love this episode. People are looking for a fractional CTO or kind of interested even in your job seeking app, they can help them to apply for jobs automatically. Where can they find out more about you online, connect with you online and maybe see if they even wanna work with you? Absolutely. I've worked with with dozens of startups, dozens of founders as a startup CTO for the past 10 years. I do work as a fractional CTO, so I help startup founders build their first product hired their first engineers. I'm writing online about all of these stuff every day on Twitter. You can follow me at sergio rocks. And on LinkedIn, you can just search for my name and I will probably come up as that the first result. Connect with me, reach out if there's anything you feel I can help you with in terms of building your company, building your product, hiring engineers and so on, that'll be amazing. If you are searching for a job just find my app. My AI agent jobs copilot.ai. I think you'll, it'll help you a lot. Find your jobs faster. That's amazing. I'll make sure to put all those links below the episode into the show notes. Thank you so much for being here for Sergio for another amazing episode of the Artificial Intelligence Podcast. Thank you so much for having me, Jonathan. This has been fun and have a great day. Thank you for listening to this week's episode of the Artificial Intelligence Podcast. Make sure to subscribe so you never miss another episode. We'll be back next Monday with more tips and strategies on how to leverage AI to grow your business and achieve better results. In the meantime, if you're curious about how AI can boost your business' revenue, head over to artificial intelligence pod.com/calculator. Use our AI revenue calculator to discover the potential impact AI can have on your bottom line. It's quick, easy, and might just change the way. Think about your bid. Business while you're there, catch up on past episodes. Leave a review and check out our socials.