Making tech literacy irrelevant | Infactory’s Ken Kocienda

25:35That's the opportunity for AI as well. That's the moment we're in. Yeah, no, I completely agree. I think that right now there's a revolution happening within engineering and a lot of folks are rediscovering basics and bridging into technology things that never touched it before. And it's a really exciting time. And it, you know, talking about your own experiences being back in college and the Macintosh dropping. I wonder about like in those hype cycles that you've lived through, you know, the Macintosh comes out, traditional graphic artists revolt against the idea of creating computer technology.

26:09Even before that, when you had like computer assisted design, like CAD technology, and no longer you needed a room of architects constantly drafting. And these evolutions and how we solve problems, you know, they altered the work that those workers did, but it unlocked for them the ability to do higher level work. And, you know, like you said, we are going to be talking about AI some in this conversation. And so I'm curious, like, what are some of the opportunities you see right now in tech that because we can bridge so much liberal arts into it that there's just so much waiting? So here's how I think about this is that for a very, very long time, we've had algorithms and heuristics, right?

26:51That this is what technologists use to create software. So you write a little bit of math and then you have a graphical user interface like maybe an iPhone with an animation. Well, how does that animation slide to a stop? You need to decide. It's kind of a heuristic. Something, my background, autocorrect. When do you autocorrect a word and when do you not? Right? The algorithm will give you a number, but then it's up to a human to decide where are the cutoffs. Right. And so what AI presents to us now is a couple of new kinds of tools, generative tools, extractive tools, right? So then how do you take those tools and put together, build a new kind of machine?

27:36And that's the opportunity now, is that we have new and different kinds of tools that we're discovering how to use. And the generative aspect of it is the most obvious. the extractive one is slightly less which maybe we can talk about in a little bit but it's like now you have this ability to say to really to talk to the computer whether you type at the computer maybe it's not really actually talking but you're you're able to chat with the computer to maybe have the computer understand what you mean and then go invoke an algorithm yeah and that's the kind of path that then the the chat the chat interaction invokes an algorithm which outputs to a heuristic, to another algorithm, which then draws you a graphic and gives you a visualization of a result that would not have been possible before because of that essential gap that the AI can help to bridge in understanding what you mean.

28:31And that's the kind of opportunity that we're in now. That's the kind of opportunity that we're pursuing at Infactory. Yeah. And that's kind of the holy grail I see is that engineers have always been building, you know, they want to build with intent. They want to have impact on the users, but there's nothing more intentional than understanding very clearly what you want and then asking the computer for it and the computer being equipped with those algorithms, those tools to then solve that problem for you. And you remove so much of this middle step and it allows people to do really transformative work.

29:03And it's opening doors for folks that never before could do that work can now even lead the charge on it. I think we're seeing that a lot at organizations within them when folks are spinning up AI experimentation groups or who's doing all with AI. You get these surprising anecdotes all across the org. Susie in accounting, who's never used a tool before, has built this amazing workflow. And it's because she knows really clearly what she wants and she was able to convey it. Maybe she already had the email draft, right? And that's the power, I think, too, of AI for engineers is it is an extractive tool, but then it's also like a refractive tool.

29:41You can take that intent and then you can shift it into all sorts of other intent using just your own natural language. And that's really, it's unlocking a lot of potential for builders. And you touched on it just a moment ago. I'm curious to know, because we've talked a bit about with Brooke, about in factory and the problem y 'all are solving and about putting power back in the hands of publishers for the content they have. You know, what inspired you Ken to like rise to that mission? What is your goal in factory and what excites you when you wake up every day? Yeah, it gets back to this point that we've been talking about is I am excited on enabling human results, results for humans that they find useful and meaningful.

30:22That's what I'm interested in. And so we go to content owners. We're working with some that have long, long history, large, large catalogs of information. there's a tremendous amount of latent value in all of that work how do we get to it and you know and in some ways i you know in in the conversations that you had with brooke previously with you know talked a lot about data trying to find the value in the data uh and uh both through the generative uh path of ai and the extractive uh capabilities that ai offers we can then point the ai at at at these content archives, at these asset catalogs, and levitate above it.

31:06What's in it? What's useful? What do I have? Create tools that can understand what's been extracted, and then make them available to the person in accounting. Make them available to the person who has a question, that has some workflows that they need to accomplish, and yet have those workflows enriched by this greater substrate of data, this more value that is already in the organization and raising it up, using AI to explain it to software so that tools can come along and make new applications and experiences for people. For people. That are useful. Yeah. I'm curious too, Are you vibe coding these days?

31:55I use AI to code all day, every day. Yeah. Okay, but I don't vibe code. No. I don't. What I do is I write very, very detailed specifications. I write paragraphs of paragraphs to the AI, and I monitor everything that it does because it's amazing. It's very, very powerful technology, but it doesn't know what my intent is. I try to communicate my intent, but I have to keep it on the narrow path for what I want. And so it is an extraordinarily powerful tool and allows me to write thousands of lines of code instead of hundreds. But at the end of the day, I'm tired because it takes a tremendous amount of concentration to keep these conversations that I'm having with the computer on track.

32:47Yeah. You know, we've talked a bit about like the vibe coding experience on DevInterrupt. And I myself, I'm a practitioner. I write code every day at my job as DevRel. And I'm using AI every day in a collaborative way. And I'm very much like you. I'm very spec driven. I always start with that really clear markdown document. Hey, you and me and the AI, let's brainstorm. Let's make sure you understand really clearly what I want here. Because that's the power is that it can do so much. And it's ultimately going to be kind of working its way back to like a baseline. a medium, right? If you want to create excellence, then you have to take that excellence in your mind and you have to get the AI to understand what it is you want.

33:24If you just say the most plainly of ways, it's going to give you the most plainly of results. And the cool thing about like spec-based vibe coding or spec-based coding with AI really is that you spend so much time gathering that information into one shared place. You really kind of like nail it down into almost like a world of determinism in a very probabilistic system, right? And by doing that, you create these artifacts of the creation process that I never even had before that are really exciting when I build things now. And I'll turn around a project in a day or I'll create something and throw it up on a website in an afternoon.

33:57And then it's like, sure, great. I have that result at the end. But what I always am most proud of is that spec document that I worked on at the very beginning. It's interesting because I was just having a conversation with a coworker the other day about, well, we need to save these documents. These documents that we make and produce and collaboration with the AR, part of the development process. And so we have a new kind of resource that we need to track and get. Because it's just like, well, why did it turn out this way? Well, we have this, it's not really a specification document. It's an exploratory document that explains part of the why for how things turned out.

34:38Yeah. And it's like a catalyst too, really. And it's like, that's why it's important to include it with source code because it becomes There's a catalyst for other people to work with your same code and get those same results. Yes. Okay, great. Well, you know, we talked a little bit about AI and using it and coding world, but I want to actually just like rewind a little bit. Going back to your time at Apple, I'm kind of curious to know from your own perspective, you know, Apple has had an incredible impact on the world and all the technology we use. But what do you think is maybe the most misunderstood part of Apple's design?

35:09I think people focus a lot on the hardware. That's the exciting thing. When Apple releases a new phone or a new watch or a new tablet or a new Mac, but it's the software that is the real key to the whole experience. It is how the software interacts with the hardware and interacts with what people want to do with it. And again, I go back to this idea of making things that are useful and meaningful. and and you know and i think that people have this this real focus on the gadget but that's not really where the magic is the magic is in the software yeah and so if you think that the most misunderstood part of apple's design from outside looking in is how much of the how much work goes into the software versus the hardware um but i'm curious too like in that apple world that you were in, what's like the most stressful demo that you ever had to give?

36:08Well, you know, I had the opportunity several times to demo directly to Steve Jobs when he was still on the scene. And that was extraordinarily stressful because he was so direct and so clear in whether when he liked something and when he didn't. And his standards were extraordinarily high. And so you could show him something and it would be, well, in no uncertain terms, he would tell you that he didn't like it. You can imagine that sometimes this got pretty colorful, but it was never personal. I remember one time where, um, we were, um, it was the first, uh, retina iPhones were coming out. So the first time that we had pixels that you couldn't see, yeah the pixels were you know beneath the resolution of your your eyes uh and so we wanted to make a new uh system font for uh for the mac uh excuse me for for the iphone uh based on um uh well now we can draw these characters more clearly and so i brought him some mock-ups and he thought they were all terrible and so he told me it's just like these are all junk what are we going to do?

37:20Right. And so it's like in, in one breath, he said, you know, and that's slightly more colorful terms. This work is junk. Yeah. What are we going to do? And so, you know, I think a lot of people feel that he was very much looking down at the people who brought him work, but in that moment, it was like, what are we going to do? And so it was, it was very, very easy in that moment to take that stress and turn it into a focus for whatever the next step was going to be. And this is part of the magic that Apple had as well. And part of the magic that I've tried to bring along with me in my work and other projects and then try to bring through to the work that we do now at InFactory is that everything is provisional.

38:05Things change very, very quickly. And so we need to plan and account for the kind of change and use the work that we do as the stepping stone for the thing that we do next. Yeah. And so that is the opportunity you see right now as an engineer working, a CTO working in a field that's being radically transformed by AI. Is that everything is provisional. Everything, there's a sea change event happening. Yes. Right? How do you manage change? Yeah. Yes. That's what it comes down to at the end of the day. It's what we talk about every week here on Dev Interrupted. Right now, the extraordinary change that's happening within organizations, it can flip them upside down.

38:44And so really getting to the root of that has been really like a powerful thing to learn more about. I want to also spend a moment just to ask about your book, Creative Selection. It's on my bookshelf. I've read it. It's one of my favorites, actually. And I wanted to ask you about your idea of evolutionary design. You talk about this a lot in the book and as part of like your demo driven process. I'm curious, how did you iterate as somebody outside of tech to then create, how did you iterate from being outside of the world of tech to being inside of the world of tech and using that to inspire that evolutionary design?

39:23Yeah. So it really is all about going one step at a time. It's one generation of work, one generation of an idea that you just build step by step. That's what it is. You get an idea and you try to create some concrete artifact that represents that idea. And then you step back and you look at it. Yeah. Is this good? Is this good? What's good about this? What's bad? Get rid of the bad, keep the good, do the next piece. So that's the evolutionary step right there. You can see it in one step. But none of the projects that I've done, none of the pieces of work that I'm proud of, were able to be accomplished in a single step.

40:11So how do you then maintain some continuity? How do you create an arc over those iterations. And that's what it is to try to see where am I trying to head with this? You know, and, and, you know, I look, I'll say it again. It's the, you know, the, the end result is to create something useful and meaningful for other people, for other people. It's not about the tech. It's a, how do you get the tech to transcend itself to be, to, to get it to the point where other people will say, yeah, I want that thing. That thing can help me. And that's what it is. So it's evolving the tech and this desire to accommodate and fill the needs that other human beings have.

40:55And it's those parallel tracks and getting them to converge on an end result. Yeah. And as you describe that, I immediately go to what is something that you designed that's used by so many people autocorrect. You have to guess the intent of something that people do every single day on their phone with all sorts of varying degree. And you're also fighting all sorts of edge cases, like trying to keep the user from accidentally swearing and or like saying things that they didn't intend. And then Apple has since, like, you know, that technology is embedded into everything that we use in the world today to where the technology is always guessing your next step, guessing your intent.

41:37And then this becomes a bridge almost like into what we're seeing with AI hitting the scene. It's always trying to guess your intent or go one step beyond. What are some mistakes that you learned in autocorrect that you think could apply to people building with AI systems? Well, I think remediation is a big lesson because, as you're saying, and starting all the way back with autocorrect, a much, much simpler system than what we have now with AI is that you are trying to divine. You're trying to understand what people's intent are. Well, what if you get it wrong? How does then the person understand that, well, going from this is not what the system did not produce my desired effect, what do I do now?

42:24How do I get from where I am now to this undesirable outcome very, very quickly to what I want, to the desired outcome without frustration, without where the person goes. If you flub this enough, people will just go away. They'll take your gadget and throw it out the window. Exactly. And so it's this kind, it's not about technology. It's very much more about psychology. So giving people the comfort to understand the system that they're working in and understand how to navigate through it to get what they want. even when they don't take a step back and I can go this way. I can pivot really easily.

43:12And that, you know, having things be understandable and explainable and coming up to the level of humanity rather than just leaving it off at the level of the machine is how that's done. Yeah. And you're talking about a technology that's used by people from all walks of life and all levels of technological literacy, right? And so being able to even not frustrate them and offer opportunities for them to convey what they really mean, you know, this is something that looks simple on its surface, but if you get it wrong, like you said, people are going to turn it off. People are going to throw it away.

43:48It's not going to catch on in the way. So, you know, you say this, you know, technology literacy. I feel that it's my job to make that not matter. That's exactly right. It has to not matter. There are people out in the world who are experts in what they do. We work now with companies in the enterprise that they're experts at what they do and have been doing it in some cases for many, many decades. And so how do we make AI now available so they can continue what they're doing only better, right? That's the challenge. That's the opportunity. That's what those companies want. Yeah. Right. And we're still at the beginning of figuring out how that can work, you know, most efficiently work best.

44:39You know, I still feel that we're still, we're very, very young in, you know, in the history of just computers, nevermind just AI. And so we're still figuring it out. But I think that there are some very, very good ideas using some of these ideas that we've already talked about that show the way for how AI will become integrated into people's everyday interaction with the tech that they use. Yeah. And talking about that beginning, I want to know, curiously, having now spoken with you and Brooke, where did InFactory start from? What kind of conversation then led to the creation of this company?

45:14So Brooke and I met at Humane. And so the challenge there was to make a new style hardware device that would open up the potential of AI available to people, to consumers. and what we found, you know, the part that I got into was trying to do that AI portion of it. And we wound up talking to some outside companies, very, very large, you know, data providers. And there was this gap that existed in this sort of the humane stack that was company-sized. this moment where you sort of left the humane cloud and went off into somebody else's cloud to fetch a piece of information and then bring it back.

46:07And that was a part that we didn't maybe understand as well as we might have. And that when Brooke and I started talking about what would come next for us after humane, we looked at that point to say there are people that are producing intents in the enterprise, Not with now a consumer piece of hardware, but now just these intents exist in the enterprise. There are jobs that need to be done and a relatively small number of very, very common, well-understood jobs that people in the enterprise want to do. And then there's the data. And so what can we do to close that gap? How can we make a company that fit into that spot that picks up from the intent and somehow gets through to these vast data resources, these enterprise data lake houses that just have everything that the company ever did?

47:03But what are the nuggets of gold? How can we get people through to the things that they wish they had and they wish they could bring to bear on the problem that they're working on right now? And how do we build a suite of tools that breaks that problem down so that we can apply engineering rigor to it? That's exactly right. You're building a primitive that we're all going to need in order to build and use these systems. And it's kind of like you're crossing an abyss. That wasn't abyss until you started working on the in-factory problem. So, you know, you say an abyss, but what we have, I mean, there's no way that we can get into these in detail during this conversation, but we have prepare, connect, build, deploy, and explore as different modules in our software.

47:47And every one of those enables an engineering touchpoint. And then it's where those different modules interact with each other is where the programming comes in, where you might have some AI assistants, where we can orchestrate, where we can have an agent participate. But it's those solid, those solid foundations that enable us to build applications, to make experiences possible, to get through to that human result that we're trying to deliver. Sure. So fascinating to hear you talk about the problem because I think at the end of the day, data is our biggest problem with AI. You know, we're ultimately having access to those nuggets of gold.

48:25We talked about that a lot with Brooke about fight, like in your huge data lake house or whatever repository. How do you find in there what's actually really useful for your workflows? Because if you treat it like, oh, we're just going to dump all this into whatever, never works. And so I'm curious, too, like when you are when in factory is talking with a large enterprise, someone like sitting someone sitting on a lot of like data and publication history. How do you start to navigate that conversation with them to figure out what is important? So so one of the chief means that we use is enterprise.

49:02And one of the main ways that we do it is not with generative AI, but with extractive AI. is that what we try to do a lot is we have custom models and custom techniques where we can levitate this layer of new data above the primary source data. And we can show, here is the value in some of the technology. You work with a publisher where they publish news stories, and so they have multiple revisions of the news stories. Here is the through line across these stories. And what we do is we cite the stories themselves. We're not producing anything that isn't in the stories, but we say, well, this revision had this, which changed to this, which changed to this.

49:46Did you know that? And it's just like some eyes open. Yeah. Right. And so then we might have a generative process, which then summarizes that. But it's like going and extracting chapter and verse. Right. This document, this line, this is the piece that is actually in common through this revision of different stories or is maybe changed. Yeah. pivoted at a certain moment. And if you can do that in the aggregate over maybe a topic area, maybe a big world event where there has been maybe news reporting that's been going on over some period of time, we can. Again, I like to think about it as just levitating this new data layer above the source data that provides additional insight and understanding, but always maintains the connection to the source data.

50:39And that is eye-opening for a lot of enterprises because, again, we're not telling them something that isn't already there. We're just making it visible. We're surfacing content that is something that they know that they themselves produced and that they can trust. This is amazing. I love the idea of how you've, And this even goes back to when we talked earlier about coding with AI and those artifacts that you get along in the process and why they're important, why they need to be included with things. And this layer that you're describing is very similar to that. You're taking information that's already there, that's already firm and written, and then you're basically turning it into this middle layer that extracts really unique insights and elevates the value of all of that corpus underneath it without making things up.

51:31It dials up the determinism to give you that really critical building platform that you need. So, you know, a big part of what we do is to find structure. We try to find and understand structure in either unstructured data or structured data, but to identify new kinds of structure, new kinds of metadata that then we can write programs against. Exactly. So one of the chief means that we have for this is a technology we call query programs. They're not queries and they're not programs. They're queries and programs together. So we query a particular piece of data and then we run algorithms over it.

52:10It's getting back to what I'm saying before. We have algorithms. We have heuristics. It's a primitive. We have generative. We have extractive. Yeah. And orchestrating all of those together to produce, again, the new kinds of effects, finding new insights, identifying new pieces of metadata, and then searching over it, doing computation with it and producing results. How many of this did we have during this time period? And it's like that information is there to be found, but is very, very difficult to extract when you have a data lake house where things are just as they are, or you have your CMS and no data scientist came over and wrote a specific application for that.

52:53What we're offering is the opportunity to make this application development a lot simpler, a lot more direct, and putting it in the hands of people who maybe are relying on some fundamental work from data scientists in the organization that use our software to make some tools, but make it available to people who are not data scientists themselves. Yeah. Wow. Ken, it's been amazing to sit down with you and learn about how you're creating the building blocks of AI that are enabling organizations to actually to get value out of it. We've been following the story for a lot on Dev Interrupted, and I'm a personal huge fan of you and your background and the impact that you've had on the world.

53:32So I can't thank you enough for sitting down with us in our very iconic Dev Interrupted dome to do this interview on site here at Engineering Leadership Conference. And we're going to include lots of information in our show notes and our newsletter so folks can go learn more about this conversation and continue to follow the transformative work that you're doing on the world. Is there any other final notes you want to leave our listeners with? I want to thank you for the opportunity. Look, I'm an optimist about the future, so I think the future is bright. Yes. Brooke as well, we talked about that on her podcast.

54:01I share the same. So we'll all be looking forward together. And thank you so much for joining us. And to our listeners, that's it for Dev Interrupted.