Balancing Risk and Reward: The Realities of AI in Business - Ken Pickering

Speaker A: Welcome to spherecast. This is Sphere's bi weekly podcast where we sit down with entrepreneurs, business leaders and technology experts to explore what it really takes to build, grow and scale in today's fast moving world. Each episode brings you practical advice on technology, innovation, funding, and the business essentials that matter most. All drawn from years of experience helping companies succeed. If you got a great idea, but you're running into tech challenges, if you're looking for innovations to stay ahead of the competition, or if you just want to keep up with emerging technology, then SphereCast is definitely something for you. Our mission is simple, to educate, inspire and guide you on how to leverage innovation and best practices to turn ideas into impact and challenges into opportunities. I'm Adin Heric, head of marketing at Sphere, and I'm joined by Mario Schwartz, our director of data and AI, who brings over 30 years of experience in the tech industry. And today, we're excited to welcome Ken Pickering, a seasoned technology leader. And instead of me giving you his full background, I'll, uh, let Ken introduce himself because no one can tell his story better than he can. Right, so this is Fearcast. Let's dive in. Can you please introduce yourself to our audience in a short few sentences?

Speaker B: Sure, yeah. Hi, I'm Ken. I am the CTO of a company called Scripta Insights, where we, we try to find people affordable prescription drugs for themselves. Before that, I was SVP of engineering at Starburst Data and CTO at Hopper. Been in engineering leadership now for a little more than a decade and, yeah, happy to be here.

Speaker A: All right, thanks for the quick intro, Ken. So, Ken, before we dive into the side, the text side, you've left engineering teams at, uh, places like Hopper start Bust ic, and now you're the CTO at this crypto Insights, as you said. So how would you describe your leadership style and how has it evolved across these different roles that you be part of?

Speaker B: Yeah, well, so I think it's, you know, so I work at growth phase businesses a lot of times, and I think the leadership role actually changes as you do it at a lot of these companies, you know, just because it's, it's different to lead a, uh, technology org when it's, you know, you and 20 engineers versus you and, you know, 270 engineers. And I think the dayto day and realities of that shift over time. So expectations modify and everything else. You know, with a smaller company, you're much more in the weeds, you're much more technical, you're really kind of helping drive architecture. You're, I I individually contribute sometimes. It's, you know, there's, there's ways to still be hands on, you know, but when you're running a team or an organization of, you know, hundreds of people, it's, it's just different. You're much more involved in, you know, executive, uh, collaboration and process and sort of those sorts of things. So I would say actually, you know, my career is, I consistently reboot and go back to startups because I think for me I am like an objective focus leader. I love accomplishing things. I think I got into engineering in the first place because I like to build or solve problems and that really hasn't, I mean, uh, the way I do that is, is a bit different now that I'm in engineering leadership. But I'm still fundamentally someone who likes to solve a challenge and solve a problem with a team. And so I'd say like, you know, fundamentally I'd say that shapes a lot of my philosophy and that's why I actually like startups a lot, is because startups can be very objective, focused and very focused on outcomes and, and deliverables and you know, you know, you're, in some cases you're fighting for survival and you know, I think that's a very exciting time to be at a company as things get more established and you know, it's much more about the system and process and what processes you're building and you know, uniformity across processes and those sorts of things. And so for me that's, that's, I'd say I am much, I am, I, I say as a human, happier probably in the earlier phase of companies when things are still a bit chaotic and you're still, you're still just really focused on, on pushing product.

Speaker A: All right, yeah, thanks for that. I see some similarities with, with myself as well in that sense. I like, I like the dynamic, sometimes chaotic, you know, and hands on doing, doing it yourself a lot. Yeah. So you recently joined Tripta Insights. Ken, I would love to know what, what excites you the most about the work you're doing there and um, you, you plan to do.

Speaker B: Of course, yeah, you know, it's, it's new. Medical, Medical is new for me. Medical space, uh, understanding the medical space is new for me. But you know, as I was leaving my last opportunities, one of the things I really thought about is, you know, I'm in my mid-40s as a technology leader, I have two children and I think one of the things you start thinking about is what kind of world are you leaving behind? You know, as you solve problems and you take on challenges as a. As a contributor, like, what are you doing? Like, how are you making a difference? Like, what, you know, you know, engineers and smart engineers and technology leaders can. You can work at a lot of different companies. The skill sets are applicable at a lot of different businesses. And so for me, it was like, well, what am I doing? And, you know, I think. And I decided that I really wanted to work on things that meant something to me. And I think. I think everyone can agree in the United States that our pharmacy and medical systems are broken. You know, we could. We. The drug prices go up consistently year on year. There's not much consumers can do or change about that. Right. And so, you know what? Sometimes you don't fix with policy, you can fix with education and technology. And so for me, it was really about, you know, when I look at what I really want to do at Scripta, uh, I think about what I did at Hopper, which is, how do you demystify flight prices? I love to travel, right? Like, how do you actually know what a good price for a flight is when you should buy a ticket? If, you know you're traveling with a family of four, you want to go to Europe, the difference between a thousand dollars a ticket and 500 a ticket is. Is pretty. It's pretty. It's pretty extensive for a lot of families. And so how do you enable traveling? And so it's the same thing. And I think it's prescriptions. It's a bit higher stakes because, like, you know, there, There's a real problem in this country with people abandoning prescriptions at the pharmacy. They go to their doctors, they seek treatment, they get a prescription, they go to the pharmacy, they can't afford it. Like, and, and. And they don't know what to do beyond that. And so my hope is, like, maybe I can recommend alternative drugs. Give them a conversation. Start with their doctors. They're aware of their options and aware of the medical ecosystem and how it operates and how it functions and verified some amount of cost transparency in an industry that is purposefully opaque. Like, it's purposefully. It purposely keeps people in the dark. And so that's what, that's what, that's what drew me in. And it's hard, I'd say, like, working in an ecosystem where data is by its very nature, like, abstracted from everybody, and everyone hides their. I mean, trying to crack into an ecosystem like that, like medical from a technology perspective, is really hard. You know, there's a lot of, like, Multibillion dollar companies that make money on the fact that like, drug pricing is not easy to understand for common folks. And so, so that's really what I set out to do and that's what I'm working on.

Speaker A: That's a great mission, I would say. Yeah, thank you, thank you for sharing that, Ken. Looking back at your career and you said, of course, a lot of startups and everything in between that you like to do and move things forward and you know, get in the dirt, I would say, what's one, what is the one leadership lesson you wish you had learned maybe earlier as a CTO or SAP of engineering that, you know, you can share with our audience?

Speaker B: Sure. You know, I think when you engineer, uh, engineering is a lot of times about perfection. Perfectionism.

Speaker C: Right.

Speaker B: Like building, building great, grand, great technology. Bulletproof technology that's tested and, you know, stands test now I started working for the military, you know, and I had ISO 9000 quality checks and all that sort of stuff. Right. And so it's kind of bred into early that like, you know, you have to design great quality software. I think it's what I found, especially in consumer businesses is actually you're more incentivized to learn and experiment and sometimes push out more janky software to take risks on like a business learning or consumer or product learning. And so I would say for me it's like, you know, and that kind of flies in the face of what engineering, you know, what you're kind of taught engineering is or should be. And so I'd say for me it's, you know, push yourself further, take more risk, don't be afraid, like, like, you know, analytically look at like risk management and how you, how you do services and try to figure out what you can do in a framework that, that you really are incentivized to test, to test multiple ideas simultaneously. Because especially with consumer, right. You don't have very many. You're just working with a direct market. Like nobody has to download your app, nobody has to use your app. And what makes your app successful can be arbitrary to you. Right. You could have, you think, you could think you have a great idea, you push it out, nobody uses it. Right. Some idea you think is stupid, you push it out and everybody adopts it and loves it. And so what I found is the throughput, like the more things you push out, the more things you test, the more hypotheses you focus on, like the, the bed, the better your product is because you're learning directly from the people around you. So like, you know, I think it took me a while to realize that and start focusing on building systems that allowed that to happen and then, you know, let you, let you push more risky code to production and put rails and kill switches in place. So if it's um, something really goes off the rails, you can termina, you know, and build it. But that, that was like a learning and effort that took years to kind of for me to realize and arrive upon, I would say.

Speaker A: Interesting. Could you share maybe just a little bit deeper into that, Ken? Um, yeah. Can you just name an example of the systems that you put in place that, that, that people can implement themselves, our audience?

Speaker B: Yeah, I'd say like, you know, if you're not, if you're not using, you know, like, like, like, like feature switches and A B testing frameworks and methodologies or more advanced testing frameworks or methodologies, you know, I'd say that's something that you should definitely look into doing. Like, one of the things that we did at Hopper especially was like, you know, when we're testing new algorithm, like, how do you enable data scientists to test algorithm in production? Easily, right? And we, you know, now there, now there's plenty of tooling for that that you can use like, like SageMaker and Databricks. And we, we didn't have that, you know, we're running on premise. So we built that, we built basically an internal way to switch algorithm around. But it was like, because it's like, it's like, how do you actually focus, you know, your most expensive individual contributors on the kernel or algorithm that makes them the most effective, right? Like, like, like how do you, how do you abstract platform, how do you abstract serving, how do you extract analytics, how do you abstract observability? And so that your engineers can actually focus on like the one thing that matters, and that's one thing that I do as an engineering leader, I think all the time is like, how do I actually focus my team working more and more on the core IP that makes our company better and less and less on like, things that don't actually matter. Like things that are things that are things that are solved problems in the space or things that, you know, you can, you can, you can bring in a vendor to handle or you can commoditize with open source. You know, there's like, there's all these, you know, I find that like, you know, especially when you come into startups, when people kind of do things with, you know, without, without a really kind of mature framework to solve those things, you're doing everything yourself. You're, you're writing your own SSO framework, you know, you're hand rolling observability, you know, you're hand rolling your test framework. And so I think it's really kind of extracting out like what really matters to your business and implementing those with third party solutions and really actually focusing on your core contributors on like delivering exactly what your company needs them to deliver.

Speaker A: Great, great, thanks. Thanks a lot for sharing those, those insights, Ken.

Speaker C: All right, so Ken, looking at uh, your you know, the new technologies and shifting to AI.

Speaker B: Yeah.

Speaker C: How are you currently using AI and when I say uh, you know, all phases of AI, you know, predictive analytics, prescriptive and of course the hot one, you know, gen AI. So what role does that play in delivering value for your customers?

Speaker B: Yeah, you know, I'd say like we're, I mean, you know, we use uh, mostly traditional AI today, right? Recommendation algorithm, processing, you know, prescriptions and deciding what prescriptions to recommend to people based on pricing and those sorts of things. So, so straightforward, straightforward sort of predictive analytic there. But you know, I'd say I'm leaning a lot more into generative AI and a genic workflow for a lot of the operations of our business. Right. Like because, because one of the things we have to do, right, is interpret plan documents and formulary descriptions and drug exclusion lists and those sorts of things and actually the main repository of those because you can't allow the systems that like, like that adjudicate, these things are black boxes that they don't let you in on. So you have to actually process those documents and build, build your own sort of mock adjudication system internally. And so like we have tons of PDFs and document tens and tens or hundreds of pages from clients and we've started using, you know, we've started using LLMs to process those documents and actually extract information from them and normalize them for data inputs. You know, just massive amounts of documentation workflow. There's also stuff like, you know, like, like when new drugs come out or things are pulled off the list, like how are you processing your clinical workflow is also something like FDA comes out with a new recommendation, like how do you process that? How do you actually put that into your system? But also like how do you actually streamline the time that you have on pharmacists checking this data? And so I'd say like looking at like a lot of human in the loop hygienic workflow is like something that we're we're interested in like how do you actually. Because we have, we have a lot of quality checks too. I'd say one of the things that, you know, when you are a medical company, it's higher stakes when you recommend a prescription to somebody than like, you know, like a flight price or a pair of shoes or something like that. You have, you have to be correct. Like you're, you're morally obligated to be correct. And so like, you know, how do we, how do we, how do we detect anomalies? How do we push stuff as things go through a workflow to a human in the loop type situation so they can approve it? And what does that look like? You know, rather than actually having someone actually review all of your output, how you actually have them to review just the output that matters. But how do you define what matters is kind of uh, another challenge. So it's like I'd say by using a, you know, I, I am a fan of not using broad, broad LLMs for these problems, but kind of using like, like, like special purposes. Like if I want, if I want to allow them to operate only on FDA information. Sometimes you actually struggle with like a chat GPT or something or a GPT with like. Because it has so many, has so much data that's built on top of like no, no, no, I just want you to use this FDA data. I need you to pro. It's. It's unstructured data that I need you to process and extract information from. But also like don't augment, like don't augment the source. Don't add additional sources of information. I'm giving you the, you know, it's. And so like trying to figure out how to do workflows with, you know, that actually you can control the inputs and outputs. Is, is something I think is important meta perspective.

Speaker C: So I would go and make a reference. Uh, I worked many years. One of the things that I worked for them was the, you know,

Speaker A: Things. So.

Speaker C: So are you also considering tapping into that and bringing that information to your customers? If something that you have recommended, you know, it gets recall. Is that a service that you can provide? Just giving you some ideas here.

Speaker B: Yeah, I mean, you know, it's uh. And so we will we stop recommending stuff that the, that gets recalled or is, you know, flagged for something. We update our data sets to do that. I'd say part of the benefit of us though is that we always encourage somebody talk to their doctor because like at the end of the day we actually can't change people's prescriptions. And so, like, we do, they do need to actually seek clinical advice and work on ways to streamline that and make it easier for doctors to prescribe the patients or change prescriptions for patients. But fundamentally, like, you know, we're not a licensed physician, so we can't actually change people's prescriptions. You know, we employ clinical staff, but we are not necessarily like a doctor or a prescriber ourselves. So it gives us a bit of freedom and flexibility. But I will say, like, you know, people lose trust in your business, though. Like, and that, that's why I would say, like, like, like targeting recalls or targeting things that are not, I think, planner being really careful about how we curate what we recommend, because you lose credibility with people the moment you give them, or, like, you've lost the customer the moment you give them a wrong prescription. They go to their doctor and they say, hey, I want to try, maybe taking this drug. It's cheaper. The doctor's like, that is the craziest thing I've ever heard. Like, you know, like, you have lost time, you've wasted their time, you've lost credibility as a business. And so, like, the precision and accuracy is probably our most important thing that we target at our company, which is why we employ so many, like, clinicians who do so many quality checks and those sorts of things. So. Yeah, yeah.

Speaker C: So I think part of what you're talking about is one of the things that we discussed earlier, which is balancing the risk and the stability. And, you know, how do you use AI to help you get that reliability and give the information, the right information to the client? So if you think about that from your perspective, what are, uh, I would say looking at how you've been proceeding and what you're learning, what should be the first practical steps that companies should be thinking, you know, about when they're trying to adopt AI? What have you learned that you can tell our audience? Hey, you know, I learned that if we do this first, it's more effective than doing this second. Do you have an example like that?

Speaker B: Yeah, I mean, and so, and as you sort of highlighted earlier, right, AI comes in all spectrums, right? And I think, you know, there's. There's traditional sort of predictable analytics.

Speaker A: There's.

Speaker B: There's machine learning and the traditional ML networks and those sorts of things. And now there's like, LLMs. And so they. There's a broad spectrum. And I'd say, like, I bucket a lot of stuff into two things. One is like, well, what does Your product do and how are you adopting AI into your product to make it a better product? And every, every company is doing that. Whether it's, you know, whether they have a website and they're putting in like a chatbot or they're, you know, they're like everybody, you know, or they're you know, like uh, you know, they're like a SQL ide. It's like a snowflake type thing where they'll generate SQL for you or they're a BI tool that can do natural language and natural language death. Everyone's putting natural, like putting some amount of, you know, both traditional and, and, and LLM AI into their product. But you know, the other part of it is, is the workforce improvement and lift, right? Like how are you educating your staff and personnel on it, but actually how are you controlling and regulating it? Right? Like you know, for instance, like I deal with you know, HIPAA compliant information. I can't have an employee push patient records into Chad GPT, right? Like put people, people's private medical records into these. So it's, so there's a lot of education about what you can and can't use it for. And, and I'd say it's like, you know, kind of riding that spectrum of like, well one like do people know how to detect hallucinations? Right. Like there's been a number of like, you know, kind of embarrassing news articles about like oh, it just hallucinates court cases, right? Like there's, there's, there's like, there's, there's all these things that it's both good and not good at. And I think it's really around education, you know, like I would not use it to blindly make drug recommendations for folks, right? It has like, you know, all the modern models have some amount of medical training and medical information training but like you can't just submit that to customers, right? Uh, you can't just say, you know, Chad, GPT, recommend me drug alternatives and dosing strategies for this, right? Because, because, because the systems fundamentally are non deterministic, right? Like you don't know how it's arriving at it without a bunch of prompting. You're not really getting access to the underlying data sets using without really digging. And so I'd say like, you know, it's great if it's writing an email for me where I like, you know, I copy and paste what I want to say and say, hey, make this sound, you know, a little bit more professional. Like so I sound smarter, right? But I mean like there's those kinds of use cases. But there's the, the harder ones of like, well, no, but it. Is it ready for critical business functionality? You know, and then it's also like, how are your engineers using it? Right? Are you. I mean, we use cursor internally to generate a lot of code. But you know, the efficacy of cursor we go back and forth on every day, you know, what use cases can we apply it to? Or is it more of a pain in the butt than not trying to like navigate like AI code versus just writing it ourselves? Right. And so, so all of that kind of comes into the mix when you're really thinking about like the maturity and how, uh, and, and sort of the, the, the, the matrix of like, you know, it's. Is. Is it bad if it, you know, puts a grammatical error in an email? Not as much as it, you know, wrongly prescribes a prescription drug to somebody. Right. Like, and so really kind of evaluating the whole spectrum of what you're using it for at it.

Speaker C: Yeah. So, so you, you mentioned what I call the twin, right, which is compliance. And the twin compliance is governance. You know, it's always. You cannot have one, uh, without the other. You know, they're coming hand in hand. So governance is, is really if coming up a lot when people are trying to use. I mean, you have mentioned it in some of your examples, but not really talking about governance itself. So how do you enforce and build a good governance framework based on, on you already talk about compliance, but I would say, you know, let's focus on governance. How are you trying to put that governance? And I also recommend you that you touch one big point which is change management training your people. I mean, that's something that people sometimes oversee. You know, they just bring the tool in, say, come on, use it, and they don't pay attention to that. But let's focus and the governance because, you know, you give something to the people, you want them to use it right away, right?

Speaker B: Yeah, yeah, 100% I'd say. It's really. And so, so, you know, I think, I think the axiom that a lot of people miss in the. In when dealing with generative AI is you still need to have, you know, a data strategy and data governance framework in place regardless of what you're doing. Right. Like, you know, whether, whether you're doing traditional SQL, whether you're training, whether you're tuning and training ML models, like it's all data in, data out, and that goes for LLMs as well. And so I'd say Like, like programmatic access to customer data that is, that is role based and defined is really how we've been trying to approach the problem. Like you know, open and federal data sets feed it into an LLM. The LLM's already seen it. Like that stuff is fine. But like how are you actually scrubbing patient information out if you wanted to look at claims for some reason to identify some anomalies and claims files. Right, like, and so automated workflow that specifically is compliant in those cases and managing the tools themselves. So like you chat GPT, you have to use our deployment of it where organizationally we've disabled training on our data. Right? Like you uh, you can't, you can't bring in your own account. You can't. Like we have to be able to regulate what goes into these systems and regulate what can be used for training. And so like, so like having it, having a system in place where one you're, you're. If they pull a report, they have to pull a report from something that is scrubbed, right? Like, and making that kind of a thing where if you pull this information it has to go through these filters, right? And then to making sure that, you know, the organizationally your security team and compliance team is, is checking these tools for governance to our strategy that one, there's an audit on the privacy policies as you adopt the tool and two, that it has the options you need that are necessary to actually block, you know, unforeseen situations. Like God forbid somebody does push patient information into it. The only thing that's worse than doing that is, is actually, you know, then it's, then it's being trained on it, right? Like, and then you, you know, then you're really, then you're really in trouble because you now have to, you know, have to tell somebody. So, so yeah, yeah, I mean that's, that's how I've been approaching it here is, is I'd say like definitely figuring out how you're controlling data access and then making sure that people follow the systematic design for that.

Speaker C: So there's, you know, there's a big challenge that, that you just mentioned and you know, the whole thing comes with AI. I think that you mentioned to Adam, you know, you said the writing is on the wall, you know, when it comes to AI. So what do you mean by that? And how do you think the company should, you know, embrace this reality instead of resisting it?

Speaker B: Right, Yeah, I mean, you know, I'll just say, you know, I think it's terrifying for a lot of technology employees and engineers to be working with AI. Like, let's just, let's just put our cards on the table. I've been, I've been coding for 25 plus years at this point, right? Like, all of a sudden, new technology comes out. I'm, uh, working with new tooling, you know, and we don't know actually how good it's going to be or, or like how, how prevalent it will be or how we'll work with it long term. I tell everybody, like, five years from now, we'll have a lot more answers than we have today. And I'd say that kind of uncertainty across everything is, is challenging. I'd say also, like, you know, but every CEO in the world is also asking every CTO in the world what our AI, uh, strategy is. Like, what are we doing? How are we using more of it? I hear great things about it, right? Like, like, how are we using it as a business? And I, you know, one of the things I do think, though is that it. Using it is an inevitability at this point in our industry. Like, it is here, it is out, it is driving results, it is productive. It is not, I think the, you know, the magic, the magic solution that, you know, some people hype it up to be. I think it's good at some things and not good at others, but I also know that it's improving quite a bit day in, day out. The models that I'm using today are much better than the models I'm using six months ago are much better than the models that were used six months before that. Right? Like, and so I think it really is, you know, and it's what I tell my engineers. I'm like, look, you're going to have to learn this stuff. Like, I, uh, you know, the days of artesian coding and not driving AI assistance in some capacity, it's just, it's done. Like, those days are over and, you know, Pandora's box is opened and we're not going to get it closed again. And so if you want to be current in this industry, if you want to develop yourself in this profession, like, you're actually doing yourself a disservice by not staying up on this stuff. It's, you know, it's, it's like, it's, it's more transformative than the cloud because it's coming on so much more aggressively than platform as a service in the cloud. And it actually interrupts specifically how we work. Like, specifically how we spend our days, what we write versus what we review. And so for me, it's like, uh, because it's an inevitability. It's like, all right, well, we have to ride the fear out. Like, I get it, it's scary for me too. But also like, it's, it's here and it's going to be here and it's not going anywhere. And like, we're going to have to use it. And I think that's really just the, that that's sort of how I see it, you know. And, you know, I think it's, you know, I was saying like five years. It's, you know, it's going to probably, you know, we're doing the lion's share of coding for us. It could be two years now with the time, with the time collapsing, I'm seeing on progressive models coming out and the market competition around, I'm like, it could be, it could be shorter than that. So if it's two years out, we better have a plan. I guess how I kind of structure my team.

Speaker C: Yeah, I mean, this coming from me. I'm a little older than you, so, uh, I've been doing this a little longer. So I've seen many other transformations that, uh, you probably have not seen. I'm dating myself that I even, you know, punch cards. So that's. I've been doing this. So I. Trust me, I've seen the evolution of how the technology is bringing, and we still haven't even touched network and uh, how competing with neural networks are going to come and make even a bigger difference on how fast things are. But I think the speed is something that we have to have some control over. I think that, that what's happening is the speed to market and putting things out there that are not, you know, tested the right way and people having the wrong impressions about what you get, it gives you like you see in Forrester and in other publications. You know, why 80% of that, you know, POCs, uh, that they tried to do with AI fail is because they're focusing on the tool you mentioned earlier, your workflows. And, and, and one of the things that we have found out is that if you focus your solution looking at the workflows and understanding how you can really implement them in different functions will bring you better success. So what I would ask you then, uh, understanding this is. I know you are saying it's inevitable, but I think the question is how do we make sure that speed doesn't kill? Because what we have seen right now, speed is killing. And if you think that the spend is close to a trillion dollars on, you know, AI and new technologies. And 80% of, of that is being thrown away. You know, I think a little caution comes into play. But how do you communicate that as a CTO to your CEO? So that would be my last question, just to, you know, bring you into not trying to put you on the spot. But you know, I think that's a challenge that you're, you have right now. And it's, you know, how do you proceed? But how do you proceed with caution?

Speaker B: Well, and you know, I think, and that's uh, that's why you pilot, you know, that's why you test it out, that's why you kick the tires, that's why you work on it. You know, I don't. And for what it's worth, like I am, I am not one of those people that says AI is going to replace engineers. I just don't think that that's going to be the case, at least for a very long amount of time. You know, I think it still requires humans to, which is why I think it's important to generate code because it still requires humans to review. It still requires some sort of output that like we can verify. Because like I sort of said before, we can't really deal with non deterministic systems, writing deterministic systems in some capacity. Right. We have to uh, like we need a way to review the output of those systems so we can verify its functionality. You know, like I was reading a report like, I think Jellyfish published something recently. It said, you know, like, people are coding, you know, 25% faster, but they're doing, it's about 10 more bugs in software, like uh, 10 more quality issues. And you can't, you can't really make that trade off, especially if you're working on anything that's like remotely like critical for people to actually utilize as a service. You know, like, so maybe some apps have that kind of tolerance, but, but for instance, I don't. Right? And so it's, so it's, it's, it's not, it's, it's trying to actually look at it with a critical eye and determine like, yes, it can help me with this. Right? Like, you know, we've spent a lot of time, for instance, working on our cursor rules files and stuff so that Cursor can write APIs. Like, and, and, and at this point, like Cursor can write APIs. But I mean it's got five pages of prompting that took a very long time for us to produce. Right? Like and, and, and systems and frameworks and, and caveats and gotchas that as it. As a chickens, as we Learned in generated APIs, like no, no, no, no, no, no. All right, let's go back to the rules file. You know, and, but, you know, and so it's. If you're not putting that effort in, if you're just vibe coding your way through scenarios like you're not going to be successful because it's great for a very simple, like I want to connect to this API and pull this data and put it into a database. Sure, right. But really complex systems interactions, it loses itself. It's, you know, it's. I actually find that I talk to it like a junior engineer a lot. Like when I'm working with Cursor, I'm like correcting it. Like I actually, it's like, it's funny because I'm talking to it like, hey, you actually forgot a lot of database constraints or hey, I don't think having a new data store for this service is actually appropriate. We should probably just use the one that we already, you know, it's like, things like that. It's like pretty big misses. And so, uh, you know, I, and so I think it's, I think it's, I think it's going into it with eyes wide open about what it can provide efficiency for. And like, you know, I think it's actually like we, we use it to review code a lot to do like surface level reviews on stuff and it's actually been really good at catching pedantic stuff. Like last week it caught someone switching switch to uh, two variables in an ap, in a, in a, in a function, in a function call, which by the way is really tough to detect in a code review because it's just, you know, like, and, and also like loves to pollute data. Like if that went into production, who the, who the heck knows what would have happened to our pipelines. Right? And so it's like, so, you know, but thank you Cursor, for finding that bug. Right. And so like, you know, I think it's, I think it is really approaching it with a like sample and if it's good enough for your use cases, use it. If it's not, come back in six months. Like, it's not, it's not, it's not science fiction. It's not, we don't have, you know, AGI yet. Right. Like, it's, it's, it's a tool that is, you know, trying to parrot what it's learned from other People, but sometimes that's not good enough. And so uh, that, that's, you know, for me that's the use cases. It's like there are certain things I would trust it with and then a bunch of stuff that I wouldn't trust it with, you know, and, and so I really do think it's trying to find the right, right, right model for your business on how to actually, how to, how to approach that problem.

Speaker A: Yes. Sorry Mario, I just wanted to ask actually on that, uh, could you put that maybe in, in the two buckets?

Speaker C: Right.

Speaker A: So what do you would trust it with and what do you don't trust it with in general? Like of course.

Speaker B: Yeah. So I'd say like. And this is actually, you know, and this is actually where I think like a good architecture can actually help you enable AI, uh, in your environment. Because I find it's good for like simple data in data out scenarios. Like if you have a very complex monolithic service that does a bunch of stuff, I promise you the agents are going to get lost in there. Right? Like, I promise you that. You know, like it's, it's like if there's too much going on, like it's going to start getting lost in that ecosystem. So for like, for simple services where like I have data in this database, I have a JSON format, I want to dump it out to a website. It's great for that, right? It's actually really great at react, react, native programming. You can feed it figma screens and get reasonably okay UI code. Right. But like, so I find like edge level conditions or simple programming exercises where it's really good at like you, you have clearly, clearly defined interfaces. It's a pretty well worn pattern in the industry. It's been pretty great. I'd say like, you know, stuff where it's really complex system interaction or you're trying to do more and more with something, you know, or you're doing like a complex workflow. It's not, it's not, it's definitely not as good at that. You know, it's. But you know, I think, I think it will actually lend itself to people producing architectures that are cleaner, like actually having cleaner system boundaries and systems design and architecture because it's actually going to be easier for AI bots maintaining that kind of ecosystem or adding features to that kind of ecosystem than just kind of working with big monoliths or big complex code bases, you know. And so I think there's ways you can actually like future proof your architecture to make it easier for These sorts of things to the drive efficiency. But that, that's what I've seen today, you know, on the business cases side, like I said, I, I use it for, it's great at like research, like market research and stuff like that. If I want to understand like vendor ecosystems or landscapes for a particular technology. Like in the past I would have had to, you know, Google that and go to a bunch of websites and you know, like find pricing lists or talk to an AE or whatever. Right? Like, if I just want to know like, what is an SSO solution going to cost me? Like, you know, what's an SSO for a million users cost? Like, it's been great at doing like sort of research like that, you know, and so, you know, kind of like assembling price lists and making it easier to do that kind of thing. So I think on the business side it's, it's gotten a lot more, a lot more beneficial.

Speaker C: I think you have answered part of the question I'm going to ask you right now. But I, I think that the question is, you know, looking ahead, how do you see the role of CTO changing over the next few years given that AI, ah, becomes even more central to, you know, operations and product development?

Speaker B: It's one of those things where I think it changes a moderate amount, but it really doesn't change a moderate amount. It just, it changes, it changes it from a, you know, like, you know, I'd say you might be managing less humans or, or different kinds of humans, right? Like you might be managing your senior engineers, might be working with four bots, you know, simultaneously as opposed to, you know, a team of like, you know, two junior engineers and some AI or something. Right. I think it changes team structure and dynamic, but actually like, I don't think it changes the strategic or like it's not determining your architecture for you. It's not that smart yet, right? Like it's not. And I don't think it would be that smart for a while. It's not, it's not really able to turn business requirements into reality in uh, a, in ah, a, in a, in a. In a truly nuanced way of like deeply understanding your business and deeply understanding your tech stack and deeply understanding your customers. I think there's a lot of work to do there. And so I think like a lot of the strategic aspects of the CTO role are actually virtually unchanged. I think when you look at how we produce code and how we work through systems designs and how we work with teams, I think that gets modified To a certain extent. But I don't think it's like the massive disruptive shift for executive leadership that other people kind of are kind of forecasting. I don't think it's going to be me and AI like coding for script in five years. Right. I don't think that's going to be the situation. Right. I think our team is actually going to be maybe more effective, more efficient, probably the same team. We might not like aggressively grow as many people, you know, as, as, as we might have had to when we hyper scaled or hyper growth companies. But, but I don't actually really see my job like wildly, wildly changing when I, when I look at like what my job function is, like what the requirements of my job are and you know, how I accomplish them to a certain extent. So I don't know. What do you guys think? This uh, is a good question.

Speaker C: M going back to, you know, to your answer. I think the, the evolution of the job, like you said, strategy is not going to be substituted by AI. It won't be. At least not I'll foresee it in. Even with the evolution of technology and everything.

Speaker B: Yeah.

Speaker C: I don't see it substituting a CTO strategy. You know, knowledge of how to do things and everything is, is, has an instinct that you will not be able to replicate yet. I'm not saying that they won't be able to do it in 10 years or 15 years. Trust me. I've seen too many changes in my lifetime or how technology has changed. You know, I come from big rooms that were huge that you needed to host a mainframe to what we do right now on the same space, how many computers I can host in there and servers and virtual machines. I do think that it will help be more efficient. One of the things that I tell people when we're talking about the new tool in the box is that you can use a lot of the speed of Genai to make your team more efficient. If you focus on that, that's a big first step on your strategy. How do I make people efficient with this? Because you started people using the technology and bringing the technology to be uh, a day to day. Like you use a spreadsheet, you use SQL, like you use Python, part of your regular toolbox. Right. And it helps you make that. Plus it helps I would say management in your role be able to have a little more insight of what your team is doing and correct directions faster. So it will help you to be more proactive than reactive. That I would think is how I see the Role evolving you will help you be more proactive and bringing your strategy under control faster beyond currently we can, many times we're waiting for things to happen. We can use now the technology to help us predict if we're going on the right trend or not. So we're bringing AI to AI. I would say so. Uh, that would be my parting thoughts. Uh,

Speaker A: I agree with Mario. I would say so. I'm speaking definitely from business and marketing sales perspective. So in this field, I would say marketing and sales is still mainly going to be done by humans, definitely. So the one thing that Ken, you said is, I think across departments, it's really the preparation, the research that you want to have, the analytics that's definitely much more faster efficient. I think each team member in marketing sales and definitely be two, three times efficient today already. Then in five years it's going to be probably, you know, multiply that as well. I think so that's, that's my opinion. And I think still, as you said with your subscriptions, for example, you still need the validation of a human before you launch your marketing campaign because it involves emotions. It falls so much more than just, you know, what they know from, you know, scanning the Internet, basically. Right. The LLMs. So that's, that's on my, on my, I think my input on that one. And I have a last question as well. So my last question is. The last question I ask, actually all my guests is if you think of one other tech leader you would love to tell us to invite to this, uh, to our sphere cast. Who would that be? Name them now and, um, we will make sure we have them on the sphere cast.

Speaker B: Awesome. You know what? I will recommend someone I work with in my Hopper days. His name is Juan Lai. He's of Connie Health. He's someone else who went into medical software to try to make it. He's working on the Medicare, Medicaid, Medicare side of the equation and making Medicare a better experience for a lot of people. But yeah, I recommend, recommend him. He's a super smart guy, you know, former Facebook, he's, you know, sharing at a couple different shops now. I've really enjoyed working with him. I think he'll have a lot of good insights. So.

Speaker A: All right, well, thanks a lot for that and, uh, yeah, we will definitely hunt him. So we will, we will of course do this after the podcast to make the intro. Ken, that would be really appreciated. And I want to say just from my end, and then I leave it also to Mario and you. I really enjoyed this conversation. I Learned many things who were really interesting and who. Which. Which definitely, like, made me think as well, how I can use it. Actually, my department as well. So thank you very much for sharing those insights, Ken. It's been really wonderful, and I know the audience will definitely enjoy listening to it as well.

Speaker B: Awesome. A pleasure. Pleasure meeting both you guys.

Speaker C: This is.

Speaker B: This is great.

Speaker C: No, thank you, Ken, and for your time. I mean, for letting us pick your brain a little bit. This is always the fun part, you know, sharing some war stories and.

Speaker B: Yeah.

Speaker C: M. And. And, yeah. At least, you know, feel like we're not alone trying to control this craziness that AI is bringing to the world, you know?

Speaker B: Yeah. I actually. I went to an engineering leadership conference in San Francisco last week, and I'm not a conference guy. Like, I'm not, like, I love. I work from home, love being here, but I was like, I should go, because I want to. I just want to confirm that everybody else is confused and terrified, and the answer is everybody else is confused and terrified. Like, that kind of entry leaders across the board, you know, it's just like, all of us are just trying to figure it out, which is, you know. But, you know, I think it's good to. Yeah, I think it is a good time for the community to kind of come together, too, because I think there's a lot of stuff we can learn from each other as we're all kind of, like, fumbling with it, with AI Strategy at this point.

Speaker A: We are. We are all on that crazy raft, I will say, going down, down, downhill.

Speaker C: Um.

Speaker A: All right, well, thanks a lot. I will stop the recording now, Ken.