Episode 45: Balancing Risk and Reward: The Realities of AI in Business – Ken Pickering

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 let Ken introduce himself, because no one can tell his story better than he can, right? So, this is SphereCast. Let's dive in.

Can you please introduce yourself to our audience in a short few sentences?

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

All right. Thanks for the quick intro, Ken. So, Ken, before we dive into the side, the tech side, you've led engineering teams at places like Hopper, Starburst, I see, and now you're the CTO at Scripta Insights, as you said. So, how would you describe your leadership style and how has it evolved across these different roles that you've been part of?

Yeah. Well, so I think it's, you know, 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 technology org when it's, you know, you and 20 engineers versus you and, you know, 270 engineers. And I think the day-to-day and realities of that shift over time, so expectations modify and everything else.

You know, at 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 individually contribute sometimes. It's, you know, there's ways to still be hands-on. You know, but when you're running a team or an organization of of, you know, hundreds of people, it's just it's just different. You're much more involved in, you know, executive collaboration and process and 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-focused 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, the way I do that 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, 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, 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 am I say as a human happier probably in the earlier phase of companies where things are still a bit chaotic and you're still just really focused on pushing product.

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

Yeah. You know, it's new. Medical's new for me. Medical space, 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 as a contributor. Like what are you doing? Like how are you making a difference? Like what, 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 I think I decided that I really wanted to work on things that 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, the drug prices go up consistently year on year. There's not much consumers can 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, 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? You know, if you're traveling with a family of four and you want to go to Europe, the difference between $1,000 a ticket and $500 a ticket is 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'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 starter with their doctor so they're aware of their options, they're aware of the medical ecosystem and how it operates, and how it functions, and provide some amount of cost transparency in an industry that is purposefully opaque. Like it's purposefully it purposefully keeps people in the dark.

And so 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 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 multi-billion-dollar companies that make money on the fact that like drug pricing is not easy to understand for common folks. And so that's really what I set out to do, and that's what I'm working on.

That's a great mission, I would say. Yeah, thanks thank you for sharing that. Ken, looking back at your career, and you said of course a lot of startups and and everything in between that that you like to do and move things forward and, you know, get into 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 SVP of Engineering that, you know, you can share with our audience?

Sure. You know, I think when you engineer engineering is a lot of times about perfection perfectionism, right? Like building great, brilliant tech, bulletproof technology that's tested and, you know, stands the test of time. 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 you 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 or or a consumer or product learning. And so I would say for me it's like, and 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 it's, you know, push yourself further, take more risks, don't be afraid. Like analytically look at risk management and how you how you do services and try to figure out what you can do in a framework that that you're 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 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 better 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 allow 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 something really goes off the rails you can terminate it, 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.

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

Yeah. I'd say like, you know, if you're not if you're not using, you know, like feature switches and A/B testing frameworks and methodologies, or more advanced testing frameworks and 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 a new algorithm, like how do you enable data scientists to test algorithm in production easily, right? And now now there's plenty of tooling for that that you can use, like SageMaker and Databricks. We didn't have that. We were running on-premise, so we built that. We built basically an internal way to switch algorithm around.

Because it's like how do you actually focus, you know, your most expensive individual contributors on the kernel or or algorithm that makes them the most effective, right? Like 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 our company better, and less and less on like things that don't actually matter, like things that are things that things that are solved problems in the space, or things that, you know, you can bring in a vendor to handle, or you can commoditize with open source. You know, there's all these, you know, I find that especially when you come into startups when when people kind of do things without a really kind of mature framework to solve those things, you're doing everything yourself. 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 abstracting out like what really matters to your business, and and implementing those with with third-party solutions, and then really actually focusing on on your core contributors on like delivering exactly what your company needs them to deliver.

Great. Great. Thanks thanks a lot for sharing those those insights, Ken. All right.

So, Ken, looking at at your, you know, the new technologies and shifting to AI. How are you currently using AI? And when I say AI, you know, all phases of AI, you know, predictive analytics, prescriptive, and, of course, the hot one, you know, GenAI. So, what role does that play in delivering value for your customers?

Yeah. You know, I'd say like we I mean, you know, we we use mostly traditionally 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, straightforward straightforward sort of predictive analytics there. But, you know, I'd say I'm leaning a lot more into generative AI and agentic workflow for a lot of the operations of our business, right? 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 a lot of the systems that adjudicate these things are 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, to 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, 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 agentic workflow is like something that we're we're interested in, like how do you actually Because 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 to be correct. Like 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, rather than actually having someone actually review all of your output? How do you actually have them review just the output that matters? But how do you define what matters is kind of another challenge.

So it's like, I'd say by using, you know, I am a fan of not using broad broad LLMs for these problems, but kind of using special purposes. Like if I want if I want an LLM 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 it's built on top of. I'm like, "No, no, no, no, no. I just want you to use this FDA data. I need you to It's unstructured data that I need you to process and extract information from, but also like don't augment. 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 that I think is important from that perspective, so.

I would go and make a reference. I worked many years with that. One of the things that I worked for them was there, you know, in the recall things. So, are you also considering tapping into that and bringing that information to your customers if something that you have recommended, you know, gets recalled? Is that a service that you can provide? Just giving you some ideas here.

Yeah. I mean, you know, it's and so we will we stop recommending stuff that the that that gets recalled or is, you know, flagged for something, and we update our datasets to do that. I'd say part of the benefit of us, though, is that we always encourage somebody to talk to their doctor. Because at the end of the day, we actually can't change people's prescriptions. And so like we do they need to actually seek clinical advice. And working on ways to streamline that and make it easier for doctors to prescribe to patients or change prescriptions for patients, but fundamentally, like 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 targeting recalls or targeting things that are not a being really careful about how we curate what we recommend, because you lose credibility with people the moment you give them a like you've lost a customer the moment you give them a wrong prescription, and they go to their doctor and they say, "Hey, I want to try maybe taking this drug. It's cheaper," and the doctor's like, "That is the craziest thing I've ever heard." Like, you know, like you have lost 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 clinicians to do so many quality checks and those sorts of things. So, 100%.

Yeah. 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, you know, get that reliability and give the information, the right information, to the client. So, if you think about that, from your perspective, what are the, I would say, looking at how you've been proceeding and what you're learning, what should be the first practical steps a company 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?

Yeah. I mean, and so 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 predictive analytics, there's there's machine learning and the traditional ML neural networks and those sorts of things, and now there's like LLMs and GenAI. 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's 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, like everybody, or they're, you know, like a, 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, a natural language dash. Everyone's putting natural like putting some amount of, you know, both traditional and and LLM AI into their products.

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 ChatGPT, right? Like put private medical records into 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 running that spectrum of like, "Well, one, like do people know how to detect hallucinations?" right? Like there's been a number of like kind of embarrassing news articles about like, "Oh, it just hallucinates court cases," right? Like there's there's like 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. All 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? You can't just say, "ChatGPT, recommend me drug alternatives and dosing strategies for this," right? Because because the systems fundamentally are nondeterministic, 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 datasets that it's 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 then there's the the harder ones of like, "Well, no. But 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, where 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 all of that kind of comes into the mix when you're really thinking about like the maturity and how and and sort of the the the matrix of like, you know, is it bad if it, you know, puts a grammatical error in an email? Not as much as if it, you know, wrongly prescribes a prescription drug to somebody, right? And so really kind of evaluating the whole spectrum of what you're using it for is really how I look at it.

Yeah. So, you you mentioned what I call the the twin, right? Which is compliance, and the twin for compliance is governance, you know? It's always We cannot have one without the other, you know? They they're coming hand in hand. So, governance is is really if coming up a lot when people are trying to use GenAI GenAI. 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 talked about compliance, but I would say, you know, let's focus on on governance. How are you trying to put that governance? And also, I commend you that you touched 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 on the governance, because, you know, you give something to the people, you want them to use it the right way, right?

Yeah. Yeah, 100%.

I'd say it's really And 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 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 datasets feed into an LLM, the LLM has 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 in claims files, right? And so automated workflow that specifically is compliant in those cases, and managing the tools themselves. So like you ChatGPT, you have to use our deployment of it where, organizationally, we've disabled training on our data, right? Like you can't you can't bring in your own account. We have to be able to regulate what goes into these systems and regulate what can be used for training.

And so having a system in place where, one, if they pull a report, they have to pull a report from something that has that is scrubbed, right? Making that kind of a thing where if you pull this information, it has to go through these filters, right? And then, two, making sure that, you know, organizationally, your 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? And then you know, then you're really then you're really in trouble because you now have to, you know, you have to tell somebody.

So 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.

So there's, you know, there's a big challenge that that you just mentioned and, you know, the the whole thing what comes with AI, I think that you mentioned to Adin, 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 and how do you think that companies should, you know, embrace this reality instead of resisting it, 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. But 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 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 5 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.

And I'd say also like, you know, but every CEO in the world is also asking every CTO in the world, what our AI 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, you know, one of the things I do think, though, is 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, 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 6 months ago or much better than the models that were used 6 months before that, right? 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 you know, the the days of artisan coding and not driving AI assistance in some capacity, it's just it's done. Like those days are over. And 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 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 the 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 and I think that's really just the that that's sort of how I see it, you know?

And and, you know, I think it's, you know, I I was saying like 5 years, it's, you know, it's going to probably, you know, we're doing doing the lion's share of coding for us. It could be 2 years now with the with the time with with the time collapsing I'm seeing on progressive models coming out, and the market competition around them. Like it could be it could be shorter than that. So if it's 2 years out, we better have a plan. I guess how I how I kind of structure it with my team.

Yeah, I mean this coming from me, I'm a little older than you, so I've been doing this a little longer, so I've seen many other transformations that that you probably have not seen. I'm I'm dating myself that I even seeing, you know, punch cards. So that's how long I've been doing this. So trust me, I've seen the evolution of how the technology is bringing, and we still haven't even touched, you know, the neural network and how computing with neural networks is going to come and make even a bigger difference on how fast things are. But, 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 the, you know, POCs that they try 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 understanding this is, I know you're 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 you have right now, and it's, you know, how do you proceed, but how do you proceed with caution?

Well, and you know, I think and that's 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 to which is why I think it's when it generates 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 like we need a way to review the output of those systems so we can verify its functionality.

You know, like I I was reading a a report like I think Jellyfish published something recently that said, you know, like people are coding, you know, 25% faster, but they're doing it's about 10% more bugs in software. Like there I mean 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 as a service. You know, like some 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 determining like, "Yes, it can help me with this," right?

Like, you know, we've spent a lot of time, for instance, working on our 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 it as we learn it 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 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 I you know, 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 it caught someone switched to two variables in a 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 it 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 6 months. Like it's not it's not it's not science fiction. It's not when 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 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.

Ken, could you, could Yeah, sorry, Mario. I just wanted to ask actually on that, could you put that maybe in in the two buckets, right? So, what do you would trust it with and what do you don't trust it with? In general, like, of course.

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 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, and 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, 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 to have 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 codebases. 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 to drive efficiency, but that that that's what I've seen today.

You know, and on on the business cases side, like I said, I 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 solution for a million users cost? Like it's 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 and making it easier to do that kind of thing. So I think on the business side it's it's gone a lot more a lot more a lot more beneficial.

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 becomes even more central to, you know, operations and product development?

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 will be that smart for a while. It's not it's not really able to turn business requirements into reality in a in 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 and how we, you know, 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 and, you know, AI like coding for Scripta in 5 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 is this is a good question.

Well, 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 don't foresee it in even with the evolution of technology and everything. I don't see it substituting a CTO strategy, you know. Knowledge of how to do things and anything is 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 of how technology has changed, you know. I come from, you know, big rooms that were, you know, huge, you know, with that you can needed to host a mainframe, right? To what what we do right now on the same space, you know, how many computers I can host in there, servers and virtual machines.

So 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 start people using the technology, and bringing the technology to be you know, a day-to-day, you know, like you use a spreadsheet, like you use SQL, like you use Python, you know, 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 and correct directions faster. So, you know, it'll help you to be more proactive than reactive. That, I wish think, is how I see the the role evolving. It will help you be more proactive and bringing your strategy under control faster than currently we can.

Many times, you know, we're we're waiting for things to happen. We can use now the technology to help us, you know, predict if we're going in the right trend or not, you know? So so we're bringing AI to AI, I would say. So that's a that would be my my parting, you know, thoughts.

I agree with Mario, I would say. So, I'm speaking definitely from business and marketing, sales perspective. So so in in this field, I would say marketing and sales is is still mainly going to be done by humans, definitely. So the the the one thing that Ken, you said is, I think across departments, it's really the preparation, the research that you want to have, you know, the analytics, that that's definitely much more faster, efficient. I think each team member in marketing, sales, can definitely be two, three times efficient today already, then in 5 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 it falls so much more than 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, Mario. 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 to our SphereCast, who would that be? Name them now, and we will make sure we have them on the SphereCast.

Awesome. You know what? I will recommend someone I worked with in my Hopper days. His name is Juan Liu. He's the CTO 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, of making Medicare a better experience for a lot of people. But yeah, I I recommend recommend him. He's super smart guy, former Facebook. He's, you know, led engineering at a couple different shops now. I've really enjoyed working with him. I think he'll have a lot of good insights, so.

All right. Well, thanks a lot for that, and we will definitely haunt 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 that which which definitely like made me think as well how I can use it actually in my department as well. So, thank you very much for sharing those insights, Ken. It's it's been really wonderful, and I I know the audience will will definitely enjoyed it listening to it as well.

Awesome. A pleasure pleasure meeting both of you guys. This is this is great.

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

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 kind of actually 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 you know, I think it is 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, so.

We are we are all on that crazy raft, I would say, going down downhill. All right. Well, thanks a lot. I will stop the recording now, Ken.