AI Co-Pilots and Autonomous Engineering

Carlos Ponce (00:10):

Good morning everyone. Welcome to another episode of Tech Leaders Unplugged, and I am joined today by my fellow teammate and co-host Wade Erickson. Hi Wade.

Wade Erickson (00:22):

Hello.

Carlos Ponce (00:24):

And of course, we're going to be also joined by our guest Pierce Healy, the c e o and founder of Zeta ai. So welcome both of you to the show. Good to have you here.

Pierce Healy (00:36):

Thanks for having me, Wade Carlos.

Carlos Ponce (00:38):

Absolutely. It's our pleasure, Wade. I'm, I'm sorry, Pierce. So alright, so let's start with you Pierce, let's start with you telling us a little bit about you. Tell about your background, how you got here, and anything you want to say about yourself, now's your time to shine. Welcome to the show once again.

Pierce Healy (00:57):

Yeah. Fantastic. Be here. So I guess quick line on Zelta and I can kind of bridge that into, into peers. So, CEOs, Zelta AI, we help software companies to organize and analyze customer feedback data. So specifically what this means is we're pulling in data from a variety of different customer channels. So, sales and success calls, support tickets, social media, and then basically automatically pulling insights out of that data for product sales and customer success. And really, where this comes from, but before starting XTO as a customer strategy consultant with McKinsey and Deloitte helping companies for years to try and organize customer feedback through the different channels. And really for the entire time in that role, saw how incredibly manual and siloed that data collection exercise was. And a lot of companies really, it's almost word of mouth is trusted to understand kind of what the top customer priorities were. And really it's only kind of been a solution that this problem has only really been possible in the last year. I would say with the advances in AI the data that you're working with in this realm is extremely unstructured. So, call transcripts, and things like that are very difficult to deal with. And it only kind of since the launch, I guess, of, the LLMs that we've had a possible solution. So that was, the starting point for CELTA about 14 months ago. And we've had some really great excitement from customers about the product we're building.

Carlos Ponce (02:31):

Awesome. That's, thank you so much again Pierce. So now let's move on to the topic as chosen. We're going to be talking about AI co-pilots, I believe autonomous engineering. So, we're going to be discussing accelerating software development and enhancing customer focus with AI. This is talk about an exciting and timely topic for today's day and age. So let's start, let's continue with that. Tell us why you chose this particular topic. Why did you feel this is relevant for today's day and age? Tell us from your perspective, why did you choose it?

Pierce Healy (03:14):

Yeah absolutely, and Carlos, why keep me honest here from going into too much detail, but as a company building a product on top of large language models like OpenAI we've also embraced them internally from the perspective of code generation. So a lot of people know about the use cases of OpenAI when it comes to, you know, write me a better email or, you know, draft me this marketing content. Maybe less appreciation for how good it actually is at writing code. It's definitely started to become something that a lot of engineers rely on. And I think as a company we have really just completely embraced this to its full fullest extent. I think anybody who's, who is on tech Twitter knows that you know last week we saw these videos of people drawing a website, uploading at the ChatGPT, and getting working code as an output. And like that's, that's here today. And I think we're only at kind of level one of what that's going to unlock. But there are probably kind of multiple kinds of subsets of use cases within that of how you can use this technology. But within Zelta we've kind of really made an effort to try and use it to its fullest extent and the productivity benefits are, benefits are just absolutely enormous. I would say we've increased our, speed of software development by 10 x easily in the last six months from embracing these.

Carlos Ponce (04:36):

Great, thank you so much again Pierce. Alright, so before we continue Wade, I believe you had some questions for our guests, for Pierce?

Wade Erickson (04:48):

So yeah, so very intriguing. You're one of the earliest companies that I've come across that really has been able to leverage this part and put it into a commercial product. So tell me a little bit about, you know, we have a lot of folks that are in that state, they're programmers doing things, old school, grab a use case from Jira, I mean a user story from Jira, start coding away, put it in the repository, collaborate between folks. Tell me how you use this kind of code generation capability and how did fits into the traditional software development process. You know, did you basically have a bunch of full-stack engineers that just started knocking it out and seeing what they could do and experimenting? Or did you kind of build a processor around that? Because I think that's somewhat missing in the marketplace on how to leverage this kind of technology and very disruptively into a scrum team or something like that.

Pierce Healy (05:52):

Yeah, I, I think there's, there's, there are three levels to this as we as, as, as far as kind of we've, we've, we've discovered it's, there's level one, which I would say is kind of like auto-complete, where GitHub copilot is kind of the most used product there. And basically, it's you're typing something out and it can kind of finish the sentence or it can add a comment at the end. The next level beyond that is actually kind of describing something that you want to build, and it will then actually write the code for you. That's probably kind of what ChatGPT is used for a lot today. If you look at any, any engineer sitting down who has the author is, is allowed to use it, you'll see them all day long putting things in the ChatGPT and instead of them having to write it out, they might, they might actually know how to do it, but it's just quicker to get ChatGPT to write it. That's probably level two. And then I think level three, which we've started to play around with, which I think I'm very excited about in terms of kind of future capability is actually embedding the into your codebase itself. So rather than it being something that's written and static and put into the code base, it's something that is on the fly writing code based on some input. And really interesting applications there around basically almost a flexible application for the end user where based on some action, the IM is going, okay, I should actually write some code on the fly here for the thing that they're trying to do. And we've done some small use cases around that, and I think over the next few years that's going to become a very big deal for a lot of software products. But internally I would say the predominant thing we're doing is offer engineers are not writing code that they could easily get written by ChatGPT. It's also probably allowed like we're a nine-person company out of that five are engineers. The other four of us non-engineers are actually essentially becoming engineers because we can write things that we want to do in the ChatGPT use that either kind of locally to try and test something out as r and d or even actually put that into the code base. So I'd say it's a combination of basically 10 x-ing the productivity of existing engineers and actually giving everybody the ability to become an engineer.

Wade Erickson (08:00):

That is very intriguing. What are you finding that learning curve for, so I would imagine the engineers, like you said, it's synonymous with grabbing stuff out of a code library, pulling it into yours, tweaking it, and making the changes this is just generating it based on your prompts. Tell me a little bit about how you see this some call it citizen development, low code, no code, but this is really code, but you didn't write it code. What's the learning curve? I mean, what are some of the backgrounds that you think the people need to be able to transition into an AI-supported software engineer versus a traditional full stack that's just using this as a tool, bringing it in, and like you said, tweaking it, inserting it into the code base and it becomes static code at that base, you know, just kind of a, what type of a person do you think would be a good fit for that?

Pierce Healy (08:58):

I think understanding the fundamentals of course makes it substantially easier because you can kind of do a little bit of debugging yourself and kind of understand what you should be asking, ChatGPT in terms of kind of, you know, applicable language. But really I genuinely believe that anyone can do this. If you're willing to just sit down, you have curiosity, you're willing to keep going. I think anybody could sit down and just start doing this. You know, your first question is actually could be like, I want to, I want to build a product, what do we need to download? And it will tell you, to download VS code or download or open up a Google collab notebook, right? Okay, I'm in Google collab, what do I do next? Like, it's, it's really end to end. I think anyone who's curious enough and has the you know, the, the willingness to get stuck in can start using this.

Wade Erickson (09:48):

Where do you see you know, there's a lot of discussion around company privacy data, you know, IP, intellectual property stuff, getting into these public AI engines? Obviously, you're prompting it, and it's spitting code out, so that's kind of the reverse. You're getting information more so than pushing in unless I think there is going to be a time where people drop in chunks of code modules and say, write a better version of this code module, you know, and then have it come out and it'll optimize things. So, when do you kind of see how Zelta fits into this? You're now obviously an aggregator of customer feed, feed information, and then using AI to parse all of that and make some better sense of it all. Where do you see Zelta, do you see yourself championing these kinds of development processes well, alongside your product or service or spinning something out separately? Or what are you thinking?

Pierce Healy (10:53):

Yeah, for, for me, so like the, the impact of this on the software industry is obviously going to be radical. I mean, it already is, but a, a couple of years in the line, who knows how the, how the landscape looks with, with this technology like enabling teams to, you know, 10 x there, their engineering capacity. What it's going to do is, first of all, allow very small teams to release very sophisticated products in a short amount of time, which is going to flood the market with new software products. It's also going to allow bigger companies to operate with significantly fewer engineers. It's going to remove the probably 15-year bottleneck that software companies face around engineering capacity. Today any software company's ability to move faster is really very firmly linked to how much engineering capacity and how much engineering ability they have that's going to go away. And if those two things happen, so if we have a float in the market of new products and new software companies popping up overnight, and we also have these bigger companies who can just move lightning fast on any new thing they want to build, the competitive edge is going to have to move somewhere else, right? If it's not, if it's not engineering capacity, what else is it going to be? And for us, it's going to be basically making good decisions. So GPT can write the code, but it can't tell you that what you're writing is a good idea, you still need to do the thinking around, okay, we're building a good product here for what the end user's use case is, we're responding to their needs very quickly. And for us basically, that's going to create a much higher emphasis on products like Seltzer, where it's creating this live understanding of customer needs and giving you the ability to quickly act on those based on this kind of infinite engineering capacity you might have.

Carlos Ponce (12:44):

I have I got a question for you, Pierce, if I may Wade.

Wade Erickson (12:47):

Sure.

Carlos Ponce (12:48):

So okay. Again, from the user perspective, my question is from the user perspective, and as someone who is also linked, very closely tied to, you know, the sales, the marketing space my question is in terms of adoption, because I see that, I mean, if you look out there, there is a very broad spectrum of possibilities and tools and choices, right? So there's a lot of effervescence about what needs to happen, especially with the advent of AI and all that, right? So, my question is, in terms of adoption, what has been the greatest challenge for you, for Zelta as a young company to overcome and or what need, and what do you need to overcome in terms of standing out against the, the noise or the multitude of options? So what else, what's your greatest challenge in terms of adoption?

Pierce Healy (13:48):

I mean, I think on that question, nothing's really changed. It's just about being extremely focused on what your customer and user needs are, right? Adding more companies to the mix doesn't change that. The companies that win are still going to be the ones that basically build the best product for the problem you're trying to solve. And I think that that's still a hard challenge. And, you know, adding more co-adding, adding, say more competition doesn't necessarily make that anymore, any easier, any more difficult. So, I would say from our perspective, that's still a journey that we're on. But we've definitely made great strides, I think.

Carlos Ponce (14:26):

Yeah. All right. Thank you so much. Back to you, Wade.

Wade Erickson (14:29):

Yeah, I see a question that came in from Nick. I think it's really pertinent talk about, you know, the impact on the designer. Obviously, a lot of you know, to just pose a whole question and expect the whole system to be built is, we're not there yet obviously. But the question was, you know, what's the impact on the designer, especially designers that can code? How have, you know, and we talked about that, that, you know, the not educated and experienced engineer can play a role in getting this code generated. How do you see that transition between the designer who's typically a UX UI guy or girl, and then how that shows up in basically a prototype almost for the full engineering team? Can you talk a little bit about what your thoughts are on that?

Pierce Healy (15:22):

Yeah, so it, 's funny, we actually find that right now the co-generation is better for backend work than the front end. I think the front-end dev is safer for the moment in terms of kind of GBT not as good at understanding styling and designs. I've no doubt in the near future that is you're going to be turning a Figma into working code in one click. I mean, that just seems like an obvious near-term possibility. So designers are going to be massively upleveled by this. They're going to be able to, you know, whether it's in Figma or some new application that has this hooked up in the background, they're going to be able to design out frames and have that turned into working code with a click of a button. And so it's definitely going to massively improve the lives of designers, I guess from respective of kind of that engineering slash design handoff.

Wade Erickson (16:16):

Yeah, I can, I can definitely see where it basically extends their value within the overall team. Because obviously, as you said, Figma, those other tools, you can get Pixel perfect code that comes out of those that drop right in for the UI designers, and then they start to just add the code that attaches to the back end, you know, and the services back there. So, you're finding those, the back end code generation, which is obviously a lot of the heavy lifting is back there. And yeah, what's interesting, you know, Logigear is an offshoring company. All our people are in Asia, and we do that, just like you said, it's because of the challenge of not only finding resources, but the cost. You know, sometimes there's just not enough budget. You have too much work to do and you need to offshore and look at some lower-cost locations to find folks. How do you see this impacting that whole industry, nearshoring, offshoring I mean, obviously they can just grab the tools and be just as fast and still maintain the cost-effectiveness, just be more efficient as well. What are your thoughts on that?

Pierce Healy (17:27):

I mean, near term, what they're probably going to do is offer the same service at the same price, but remove 50% of their headcounts after that, I mean, eventually, I'm sure companies are going to have, you know, less need for dev full stop is kind of what we've, what we've spoken about. But yeah, for service providers that are able to say, you know, this is our price, this is what you get, get at the end, massive opportunity for them to use this kind of automation. What I would say that was the most interesting end state or kind of impact of this technology in this, area is, is actually completely taking the engineer out of the loop of the product. Not in terms of the kind of AI running a product, you know, end to end and itself, but having a more flexible experience for the end user because of the fact we can almost have like an on-the-fly engineer within the application. And I think in the same way that today, YouTube and Instagram are almost personalized to me when I log in, in terms of the content I see, I think that CRMs and ERPs in the future are going to be personalized to the end user. If you only need to see contacts in your CRM, the buttons for everything else will just be removed. They won't, they won't be visible to me. So I won't, I won't log in and see all this clutter that I have today of all these things that I don't necessarily need to do. It'll be responsive enough to remove those things. And then if I have a chat window and I'm like, Hey, I actually want to see X, Y, Z, it'll be able to formulate that on the spot for me. But like, get the data, visualize it in kind of flexible ways that maybe the company itself didn't even think about or carry out any task automatically from there. And that, that for me is the really, really groundbreaking implication of this beyond just, you know, increasing productivity is like actually using this as an embedded dynamic co-generating part of their product.

Wade Erickson (19:28):

Yeah, I would imagine, you know being a testing company that it, as far as our origins, we are, you know, the demands of AI in software testing are, is really huge right now. They want to just, like you said, build the code from that handwritten UI screen. The testing tool companies are looking at generating all the test cases and the testing scripts from looking at that finished screen by the programmers so that they can validate that. I'm just wondering how you know, as these libraries that are generated from AI become refined, it's kind of like a lot of people that use Salesforce, you know, why test Salesforce? They manage all the code. I'm just using configurations here, but in reality, there are full-on companies that have software testing for past tools where that code is really being pulled in from libraries and you're doing almost like a drag-and-drop for functionality. But still in the, there's some testing. Do you see any impacts from what you've generated? Did you still have to do significant testing on that code, to be able to release the product, whether it's manual testing or automation testing?

Pierce Healy (20:46):

Yeah, well, one, 100%, and I mean, it's like GPT will almost debug itself and it'll give you something that doesn't work. And then you, if you, if your prompt it saying that doesn't work, it'll go, oh, I missed X, Y, Z. And so that's, that's still a huge element of working with this, that debugging is essential. There are certain tasks that we've found it's pretty reliable for an interesting one that we've tried out in beta is having more customized email templates for our users. So instead of us saying, okay, when you click these buttons, this data gets pulled into your automatic email that gets sent out with your notifications. Instead of hooking that all up in a kind of a templatized way is to say, okay, here's the data. This person wants an email about some topic they're interested in GPT just writes the HTML for the email. And that's a kind of a simple example. It's pretty reliable at doing it but is showing us this window into a scenario where if you consider email to be an interface as a front end, it's actually being completely written by GBT on the fly for every user instead of some kind of template that everyone has to subscribe to.

Wade Erickson (22:00):

Yeah. I think you're on mute, Carlos.

Carlos Ponce (22:04):

Yes. I'm sorry. It just, there's a, there was some background noise here and I need to shut the windows. That's the end. I'm sorry. I apologize. Oh, well, well, Pierce, unfortunately, we're coming up on time and of course, I wouldn't want to end the conversation without encouraging our viewers to visit Zelta.ai. And so you can see right down below you'll see a ticker with the address. So go there at just excuse me, Zelta.ai, and you'll find all the info that you want about the company and how to reach, of course, our guest Pierce. And last but not least here's Pierce's LinkedIn information. You can also reach out to Pierce right there, write it down, and immediately go ahead and reach out. And I'm sure there's going to be a lot of good communication going forward. So Pierce, the only thing left for me again, is thank you for having been with us on the show, and I really look forward to learning about what Zelta is about and how you know how it works and all that. I, personally, am thinking of just giving it a try and, hopefully providing you with some raving reviews about the tool and the platform, and the solution that you're providing for the space. So thank you for that. Before we go stay with us as we go off the air. I have a quick announcement which is about tomorrow's guest. So tomorrow we're going to be speaking with, excuse me, [INAUDIBLE], the product marketing manager at SetSail. And the topic is going to be sales superpowers with generative AI, the hidden gold mine, and practical generative AI strategies for Sales teams. So there you go, Wade. We talk about an interesting and relevant topic for us, right? So we're looking forward to this conversation. And with that being said, again, thank you so much Pierce for having been with us on the show, and see you next time right here on Tech Leaders Unplugged.