The AI Impact on the Developer Future with Joseph Morais | Ep. 27

Tim Berglund talks to Joseph Marais (Confluent) about his career in data streaming. Joseph’s first job: SAN administrator. His challenge: AI radically changing how developers build software.

The Pragmatic Engineer episode ft. Grady Booch: https://newsletter.pragmaticengineer.com/p/software-architecture-with-grady-booch

SEASON 2
Hosted by Tim Berglund, Adi Polak and Viktor Gamov
Produced and Edited by Noelle Gallagher, Peter Furia and Nurie Mohamed
Music by Coastal Kites 
Artwork by Phil Vo 

•  🎧 Subscribe to Confluent Developer wherever you listen to podcasts. 
• ▶️ Subscribe on YouTube, and hit the 🔔 to catch new episodes.
• 👍 If you enjoyed this, please leave us a rating. 
• 🎧 Confluent also has a podcast for tech leaders: "Life Is But A Stream" hosted by our friend, Joseph Morais.

Transcript

SPEAKER_00 0:00

I just got access to Claude Code, and uh it's brought like some real joy.

SPEAKER_02 0:04

Does the profession of software developer collapse? Does it shrink? Does it grow? I think it changes. If I can get the computer to do what I want it to do, why do I need to speak its language?

SPEAKER_00 0:13

Why can't it speak mine? Welcome to the Confluent Developer Podcast. I'm your host, Tim Berglund. In today's episode, I got to talk to my friend and colleague, Joseph Morais, and we wanted to spend some time processing what's going on with AI, honestly, tell the stories of our own experiences with uh agentic coding tools and speculate a little bit about where we think things are going. And I know that's speculation a lot of people are doing these days, but Joseph is a pretty interesting guy. He's worth talking to about this kind of thing. So we cover a lot of ground, uh, talk about a lot of other resources that are linked in the show notes. So take a look at that as you listen in. Hey Joseph, welcome to the show. Hi, Tim. How are you doing today? Not doing well. I'm doing all right. You are uh, as we know, the technical champion at Confluent. Yes. And I you know, I would describe your you as uh being possessed of champion status, but what is a technical champion? What do you mean?

SPEAKER_02 1:19

I I I learned later that they gave me the role just because I'm one of the largest people here physically. And so that is actually true.

SPEAKER_00 1:25

Give him like an appropriate role in that joke. If you're if you're listening to the audio version, you can't tell. Even video, you can't.

SPEAKER_02 1:32

You can't tell.

SPEAKER_00 1:32

You can't tell, but I just I want you to know. I want you to know.

SPEAKER_02 1:35

Yeah, I'm I'm I'm I'm larger than average. It's I'm a big guy. But uh, that's not the the only reason. Uh it's a good reason they gave me the title. But uh, what I do here is kind of similar to what you and your team do in DevRel, where we I interface with the community through social media, through a show that I host called Life is But a Stream. Um, except it's you got to squint and you you can kind of see what you guys do, but it's a little different. Whereas a lot of your efforts are rooted in the community around open source uh projects, Apache software projects like Kafka, Flink, and Iceberg. Uh, I'm unabashedly pro-Confluent, right? I uh have a very strong operators background. And you know, the the fully managed and easy-to-deployness of Confluent is something that is really meaningful to me. And uh I tell everyone that if you know you you love this technology, but you're having challenges running it at scale, or you know, maybe there's some deficiencies in some of the features that come with open source, I am all about evangelizing the benefits of Confluent to Enterprise.

SPEAKER_00 2:37

I love it. Given that, what's on your mind? I don't know. What are you what are you thinking about today? I mean, I I know we're all thinking about these days, but but lay it on me.

SPEAKER_02 2:49

Yeah, so um I just got access to Claude Code, which almost like in like this is 2026, so I feel like I'm a Luddite, but I just got access to it, and uh it's brought like some real joy back. Not that I don't love working with data streaming, I love working with data streaming. Um, but there's been some inflection points in my career that have just been like brought the kid out of me. Like the first time I deployed a container, the first time I got into like Cisco ILS, iOS CLI, and I started, you know, deploying configurations, just kind of different checkpoints in my career and these different kinds of turns of technology. And they've always made me giddy, and I can see this again happening with Claude Code. Um, but what I the reason I experimented with it is I kind of wanted to understand how are developers and engineers and data scientists now and into the future going to be working with SAS and CLI tools. And I really think there's gonna be this intermediary, right? Whether it's Claude or Cursor or something equivalent, where you're not gonna have to learn the CLI. You're gonna have your AI friend learn the CLI. And you're it's gonna learn the CLI faster and more in depth than you ever would. And suddenly you have this tool that's unlike a hammer, right? Hammer can't give you feedback until you're hitting too hard or you're using the wrong nail. It just knows how to hammer. I have this tool now that I can interact with and I can have deliberation with and will catch things that I didn't even think of. So, Tim, what's really exciting to me, and and with the ultimate goal of kind of building and demonstration using Claude and Confluent uh cloud. But what I've noticed is it's kind of changed, it shifted my thinking, and not just professionally, but even in my own personal life, and that I've I'm shifting from thinking objective-based to idea-based, right? Or, you know, we always have ideas, right? But for me, when I would have an idea, it would quickly shift into a set of objectives. And under those objectives, they're tasks. But now, because I know that I have these AI friends, whether they're Gemini or Claude or whichever that comes out of the box that day, I can focus on the bigger idea and let the agent or the coding assistant work on the objectives. And it's a really interesting thing, right? Because as it's starting to go through some of the objectives, I'm sitting there thinking, how can I make this idea bigger? Whereas I never really had the bandwidth to do that, because it goes right from idea to objective to execution. But I can kind of work higher up. And it's exciting to think about that. And again, it's not just work, it's even in my personal life, like car shopping, is now a completely different scenario because I'm having these long chats with Gemini about what I care about, what other features. And and for me to have replicated that level of effort to get the knowledge I needed would have just burnt hours and hours. So this, I don't know, it's it's we're heading into exciting times and uh I'm ready for it.

SPEAKER_00 5:42

Well, I just want to unpack this and talk about it. This is this episode is about Tim and Joseph just thinking about AI.

SPEAKER_01 5:51

Yeah.

SPEAKER_00 5:51

Can I tell you my my story? My I would love it. It's a it's a Claude Code moment, and I, you know, uh uh Anthropic is not a sponsor of this episode. We are we are two guys talking about tools that we've used. There's no commercial relationship here or anything like that. Um about a year ago, uh on my team, Victor Gamov on my team, who is I think competes with you for the title of tallest freestanding sort of technology twin towers in the world. It's just it's unreal. Yeah. Um he was uh kind of an early advocate of AI assisted engineering. Um and it was around a year ago. So it's uh February 27th, 2026, as we record this. Um I was working on a personal project. Um, it was this LED uh uh version of a Nixie clock, and uh maybe I'll I'll put some pictures online, put them in the show notes or something. It's a cool project.

unknown 6:48

Right.

SPEAKER_00 6:49

Um and I wanted it to uh to figure out um whether it whether it was in daylight savings time mode or not, like switch automatically. And it had the microcontroller, had some Wi-Fi, it could connect, and blah, blah, blah. I gotta fight through all that stuff. Um, I was doing all the firmware by hand, which to me is like people who do woodworking. You know, it's it's not because you want the chair, it's because you want to do the thing.

SPEAKER_02 7:12

That's right.

SPEAKER_00 7:12

Um and I thought, you know, Victor's always on about this AI assisted engineering. Let me get one of the popular tools and say, okay, I need this thing. I happen to have a Netlify account, so Netlify functions, make a Netlify function that will reverse geocode the IP address, figure out if you're in daily savings time, return a little message that has the offset from DST in a super easy to parse way because it's going to be resource constrained C code parsing it. Um no, no, it didn't work. It like it wrote the code correctly, uh, but it couldn't do a Next.js build and it couldn't deploy, once I fixed that manually, couldn't deploy to Netlify. Like, if there's one thing a model has lots of training for, so I'm like, okay, not ready for prime time. And in that same project, I needed some help with HTTPS support because once I got that to work, I could only serve an HTTPS. And again, in this microcontroller, it was it was just HTTP at that point. Ah, I gotta fight through this stack, you know. Uh Claude, help me, right? Give me an example, the Raspberry Pi Pico SDK. It was garbage. The code was just no good at all. I had to fight through it, do it the old-fashioned way. Got it to work, and I thought, okay, you know, it's I'm hardly a skeptic. LLMs are great, but uh, you know, and then uh Christmas, just two months ago. Oh uh I sat back down, people are talking about clawed code. Yeah, I'm like, let me try. And I just prompted out an application that worked, right? Uh related to a similar project. It was a bunch of back end Kotlin code that was doing stuff, talking over Bluetooth to a device, blah blah blah. The details don't matter, right? But it it just it it unfolded, it just worked. And then, you know, my wife and I are building a house, and I thought, okay, I want to be able to take a drone picture and overlay the property line from the plat file, the plat drawing, and all this stuff, you know, and prompt it out. Those actually didn't work quite so well. There's some GPS things that I'm doing wrong there. But um I'm just kind of crapping out the software. Yeah, and if you're watching the video version of this, right, um I'll I'll point this little squiggly green line back here. Yeah, it's right there. Um, is a little thing running on a one gigabyte Raspberry Pi 5 that I prompted out in an afternoon, Sunday afternoon, uh because it was fun and uh based on you know kind of a physical thing that electrical engineering undergrads do with an oscilloscope. All of this is very real. It's it's like the game has definitely changed. That's my own story. Went from a year ago, then it's not ready for prime time. And I'm like, oh, it's prime time now. Oh, it has.

SPEAKER_02 9:58

And and that's December of 25 is when everyone kind of, I don't know if anyone has pinpointed the day, but everyone's kind of pinpointed that a shift happened in the quality of the models. That is really just it's it's it's a it's uh a cognitive evolution, is kind of the terminology I've used to it, because now I get to start thinking in ideas and had the confidence in the coding agent to kind of fill in some of the ambiguity. So when I was building out this demo and I haven't completed it, I've been doing a phase by phase and I'm actually really enjoying it. Um the I wanted to do a uh a narrative around hospitality. So, you know, as I'm plugging away in the prompt, I'm like, hey, I want to build a demo. I need two producers, I need a single consumer, we're gonna give some flink, we're gonna tie things together, but I want to do a hospitality um theme. That's as much as I gave it. And it came out with a really intelligent set of inputs, like check-in times versus like um orders to the room and merging those together to get the 360 to then send an SMS alert to let someone know, hey, your room's ready to check in, or hey, your room is uh or your your your food is on the way. And I didn't give it any of that, but it made a very sensical narrative to go along with this idea I had. And I was kind of left here just feeling, I mean, both shocked and empowered. It's a very weird feeling. And you know, we talk here at Confluent a lot about you know providing a data substrate or um a data plane, right? I love I love the word substrate. It's it doesn't get used to it. Yes, I use that term. I heard you say it first and I reuse it. I think it's very appropriate. Yes, I'm flattered. I did not know that. No, it's good. Substrate's excellent. But you know what's what's fascinating to me is you know, we're evaluating a bunch of tools for marketing that are AI powered, and I realized that these tools like Chat GPT and Claude, they're an intelligence substrate, right? Like you have an idea for something, and now you can plug this brain into it, and now you just gotta tweak many things, often in natural language, and you can build a real product. It's um it's an amazing time.

SPEAKER_00 12:03

Now a quick word from our sponsor. Confluent Developer the Podcast is brought to you by Confluent Developer the website, which has everything you need as a developer of data streaming systems. And it's completely free. We've got curriculum, hands-on exercises, executable tutorials, the online data streaming engineer certification, also free, a way to find a meetup near you, those are free. Everything is there. I really want you to be successful in your journey as a data streaming engineer, and this is the site that has what you need. Check it out at developer.confluent.io. That's developer.confluent.io. Now back to the show. Um, let me ask you about the anxiety that I think a lot of people are feeling. Because you and I talking right now are talking as people who earn their living as, you know, communicators, explainers, teachers, however you want to put it. I mean, I frankly, most of my time is as a manager, right? Sure. I haven't I spent the first 20 years of my career writing code, various platforms, various languages. Uh, but it's been a little bit since that's been the way that I earn a living. The way I I obtain food and shelter uh is not code. Likewise, you. Both of us have to know how to do it. That's right. We wouldn't be able to do what we do without that that background, wouldn't be credible at it. But um if if somebody comes in and takes the job of coding, you're fine. I'm fine. We are talking about this, like, oh, I have these cool toys to play with, and I made a better demo faster. That's happening in developer advocacy. Like all of a sudden, kind of last summer, even before you know the the inflection point and coding agents, our demos were way cooler. You know, where did this web front end come from? Um people worry. So what do you think? Um, does the profession of software developer collapse? Does it shrink? Does it grow? I think it changes. So it always it's always been changing. So it's always been changing. It's a big change, though.

SPEAKER_02 14:06

But it right, and I think it's gonna be a scary change. So I'll address this in a couple of ways. So my the bulk of my experience has been operator engineer. I've been a SAN administrator, I've ratten stacked servers, I built data centers. The only thing worse than building a data center is moving a data center, but that's we can spend a whole episode on that. Um, so when cloud became prominent, I kind of had similar thoughts. Well, who's gonna need network engineers? Who's gonna need SAN administrators? Who's gonna need operations people when all this is the other end of an API and AWS or Amazon or Microsoft or Google are gonna manage all that for you? But guess what? Still network engineers, there's still SAN administrators, and you can make an argument that some of those people are doing better now than they were previously, even before the advent of the cloud, because more and more people are focusing towards coding. So those infrastructure people became a little bit more rare. So there's still some, you know, that there's still plenty of skin in the game. I think we're gonna go through something similar with engine software engineering. Um, I think there is gonna be an overreaction, not to say that some of the recent announcements in staff production uh are those overreactions, but I do think there's gonna be some overreactions and over-reliance on the gentic AI, and it's gonna have some some lasting effect. Something's gonna happen. Some business is gonna lose a lot of money. Hopefully, nobody loses their lives in the process. And then I think there'll be a pullback. And then we'll start to see people realize you've had like these are tools. They're the most intelligent tools we've ever developed as humans, but they're still tools. And if you let them go wild, you're gonna have nothing but noise and chaos. I still think you need to have a human in the middle for these things to be truly, truly effective. Now, will there be a reality when that human can be removed from the middle? I hope not. You know, I think these intelligent tools will make us better at what we do. I think, you know, there's this concept of the 10x engineer, right? Like the engineer out there is a like the unicorn, full stack engineer that puts out 10 times as much code as most of their contemporaries. Well, I think what's going to happen is the future is everyone's gonna be a 10x engineer. And maybe there'll be a handful of 100x engineers. But I don't know if that takes away things. Uh maybe initially, but I think eventually we'll hit an equilibrium. And what we'll see is an unprended unprecedented level of productivity across all industries. Um, that could, I don't know, what happens if you reduce release cycles to minutes or you know, put out features in weeks instead of months? I think it could be a really exciting thing for everyone.

SPEAKER_00 16:38

I tend to agree. Let me can I um I mean you're the guest here, but is I I'm just gonna. No, please. No, I love to hear you go. Um so uh a couple things. Uh uh I mentioned Victor. Um the other day I uh uh had the opportunity to be actually in the same place with him. Oh, that's great. And I was watching him do he was kind of showing me the the thing he was building because he had an idea that morning for a thing that he wanted to make. Um and he's got this you know fairly disciplined process with OpenSpec and uh uh it it you know it's a tool chain, and I'm I'm watching him do AI assisted engineering, and I kind of caught myself sort of stepping back and like it's it's Visual Studio Code, it's markdown, there are folders, they have markdown files in them. This is not code. Okay, I it's not I know what code looks like, but just kind of like blurring your eyes a little bit and and watching it, it looked like engineering was happening. This and I think a non-technical user, um I was I was kind of bummed. My my wife was out of town, uh, and she's tech savvy and very literate. She's not a developer, right? So she could she probably like could eyeball the difference between Kotlin and Markdown and know one is a programming language and one is is an outline with words. Yeah, uh, I think she would probably just kind of nail that one. Uh but I I think to her, this would just look like somebody was coding. Right, right. It seems like to me too. Yes. So number one, it it even with the most advanced tools right now and the most advanced methods, you're an engineer. And then something goes wrong. You have to know and enough about what's going wrong to to dig in there, yeah. Uh and and and fix the problem. And if you're if you don't have a background as an actual software developer, that's gonna be tough.

SPEAKER_02 18:34

I think so. And I I do think there is some it's about elevating. It's right, it's like almost like moving up the stack, but mentally moving up the stack. Because you know, I think of you know, when you and I were started in the industry, people wrote HTML in CSS. I did. Right? But with the advent of things like Wix and Squarespace, do people even do that anymore? Or or these more advanced frameworks that kind of take out the minutiae, right? So you work a level up. You know, uh, I had a guest on Life is But a Stream, um, Alex Hogan, he works for Cursor, and he said something that really especially resonated. Yeah, that was a good episode.

SPEAKER_00 19:06

That's uh they're like a Warp Stream customer, right?

SPEAKER_02 19:07

Yeah, Warp Stream customer. So if you guys don't, if the audience never heard of Warpstream, you should check out Warpstream, it's amazing. And uh some of the telemetry uh and um accounting mechanisms, if you opt in as a cursor customer, go through WarpStream and they help improve their auto tab, which basically you know figures out what you think you want to say, which is pretty impressive. Um, but he said he he made a very audacious claim, and and now I I feel even stronger that he was correct. He said, code is a bug. Like what we really want are computers to do things, and we have to communicate with them because this is the shared language we have. But what if the shared language is closer to how we actually conduct and think? And it really blew my mind, right? Like in people have their favorite, you know, coding languages and they're very feel very strong about C versus Java versus Python. But I think I tend to agree with him, right? If I can get the computer to do what I want it to do, why do I need to speak its language? Why can't it speak mine?

SPEAKER_00 20:03

Yeah. Why can't I just say computer? What I and then Majel Barrett's voice answers back. That's right. Because that's all we want. We just want the computer from Star Trek. You said moving up the stack. Uh I will a thousand percent link in the show notes to the podcast I am about to discuss right now. Uh it was an interview on uh The Pragmatic Engineer with Grady Booch. Okay. So Grady Booch is like one of the godfathers of software engineering, software architecture. Um and I could I could just say I find him to be a deeply enjoyable thinker. He is uh a guy, I I guess it's just this is what a mature intellect is like. Um is kind of how I I think that in the engine. I always said I saw him give a talk in person in the Denver area 25 years ago, and it's the first time I heard him say software, the history of software engineering is one of increasing levels of abstraction. Right. That's what we do, right? Because we're limited in, I'm gonna use a philosophical term, our own noetic equipment that is our minds. Yes. We're not getting any smarter. That's right. Um, this is we got the minds that we have that can tackle problems of a certain complexity. We want to do more and build more, so we need chunkier, bigger, more abstract tools to build more stuff. I like that. Um we don't write everything in assembly language. Now, sometimes we do. Uh I mean you used to write assembly language, right? I mean, you used to write an assembly.

SPEAKER_02 21:24

Like, who's doing that now?

SPEAKER_00 21:26

Like, yes, it's fun. Um, but but you're not gonna make a living writing business software that way. You're you're can't you're gonna go out of business. So we we have this thing, and and he uh compellingly, beautifully makes the case that that's what this is. This is another layer of abstraction. And he's like, software engineers, don't panic. Right. Um and so I go listen to that when we're done, you're done listening to me and Joseph's great interview, great thinker on this topic.

SPEAKER_02 21:55

Right, because the more you know about what you want to build, the more effective the tools become, right? If I Said if I gave it like a two-sentence prompt, build me end-to-end a data streaming platform. It might try and do that. It's not going to do a good job at it. But if I said, hey, I need a data substrate, I need a processing substrate, oh, here are some tools that you can already use to build this. They're industry standard, they're well known. Here are the docs. That all comes from at least a place uh you have to be grounded in what you want to do and know what you want to do. You know, if you if you try, you could vibe code something simple, that's really simple, but if it has like production stakes, if it has lives, it has dollars and cents attached to it, that's when you really need to know what you're directing your junior coding assistant to do. And and just to kind of uh tag off of what you mentioned with Victor, you know, he had an idea and he immediately jumped in and started doing it. I'm sure all of us have had ideas for something. I'm like, oh man, you know, two, two, three years ago, that would have taken, that'll take me a week or two. I don't know, I may be able to get to that. Hopefully I can find the time. But now I I've compressed time, and that same objective is possible in an afternoon. Now suddenly I have a lot more opportunities to build cool things.

SPEAKER_00 23:01

Because the things that, for example, yeah, the green squiggly line in the background. Green squiggly line. There's like three or four OpenGL shaders in there to make it look like an oscilloscope trace. Right. I've never written an open GL shader in my life. Am I gonna go look up how to do that or find an example to copy and paste and tweak? I'm just not. It's a fun idea. And and this is just a fun little blinky thing in my background because it makes me happy. Uh, a true hobby project. Right. But it's a good example of a thing that I it just wouldn't have contemplated as a project that I would take on. Right. And now the cost of creating that thing is low enough that I'm definitely gonna contemplate that. I'm gonna build that. That's right. Um and uh this is this is key. So there there are there are some unanswered questions, and I want to be sure clear to everybody who's listening. Um I we don't I don't have this figured out. I don't know what's gonna happen. You don't know what's gonna happen. This is this is this is a significant shift in what we do, software engineering, uh a lot of kinds of economic activity. We we don't know how it's gonna work out, but uh my I'm gonna give my optimistic view, then I'll I'll push back on it a little bit. Um just these examples of hobby projects point out to me that there are so many things lying around on the ground that you can't even notice. You you can't even contemplate building that thing because you're too busy doing everything else the hard way. That's right. Twenty or thirty years ago, I read, I remembered it as an essay uh by Theodore Gray, who's a co-founder of Mathematica with um Wolfram, Stephen Wolfram. Um I found it recently on his website, website still, and it was a like a transcript of an interview. Uh whatever, I maybe we'll link to it. But it was Mathematica um was, as far as I know, the first software that did integration symbolically, like the hard way that you learned in calculus. Right. I mean, the first computers that we made were were made to do numeric integration, right? The ENIAC is electrical numerical integrator and calculator. Uh so you know, we've been doing it through numeric methods since there were digital computers, but now there was a program that would do true symbolic antiderivatives, you know, and math educators freaked out. They're like, uh, we're gonna be stupid now. And he kind of just peels that onion, which by the way, go ask a 12th grade AP calculus student in the second semester whether they get to use Mathematica or they're hating life on trig identities. I dear reader, they are hating life on trig identities. It did not be easier. Don't worry. But what Gray said was just peeling the onion, he's like, okay, you're a hunter-gatherer now. Um all day, you're trying to get food, trying to not die, trying to not get eaten, not get injured. And those weren't stupid human beings. That's right. Or the anyone who is a hunter-gatherer, they're not like a different kind of intellectual stock. It's the same, I use the word noetic equipment. And you have to use all of it to survive there because you just have no other tools or abstractions to work with. And you and I teleported into that life. Like I've spent some time outdoors. Uh, you know, I'm all right. I'd die. I would die. Um yeah. I my and my outdoor skills live in Colorado. You know, I'm not bad. I'd probably just die. Sure. Um because it it takes so much specialty, and we're used to all these levels of abstraction and tools and economic development that let us do all these things. So here's a new layer. Exactly. We get to build more. And the question is gonna be how much software do we need? Because if we can actually build 10x more software, there's there could be uh what economists call a demand elasticity curve there where okay, it's this much cheaper. Uh well, we're gonna make some more, but we actually don't need all that, so employment in software engineering will contract, right? That's a possibility. Um and I I and I I just don't know. Um, but it's also possible that we just get to make so much more stuff now, and that that all of those ideas that just like a hunter-gatherer isn't able to think about a data substrate because that's meaningless, there's all this stuff that was meaningless to us six months ago. It's exactly right. All of a sudden it is attainable.

SPEAKER_02 27:42

Yeah, it's it's really incredible. And I think there is some risk. And so this is what's so fascinating about the tool. I mean, because i I I it's to to kind of address what you said. I do think there'll be a glut of software we don't need for sure. Uh, I think there'll be a lot of bespoke software that people will just build themselves, which I think will be interesting, but also very dangerous. Um, but my hope is that ultimately, you know, we there's a once again an equilibrium in how much software we build, and what we instead see the software that is necessary gets higher and higher quality. Right? Like again, just to pick on uh an example that's close to you and I, you know, when I got here almost six years ago, Confluent Cloud was a shell of what it is today in terms of features. We've built insane. So much cool stuff. Now, what if we what if we were able to say, okay, it took six years to get this? What if we were able to do with the same amount of work in two? Right. What uh like what downstream possibilities would that have provided to our customers by having table flow four years earlier or things like that? You know, and that's where I'm that's where I think the optimistic Joseph is really excited because I think we're gonna get levels of quality and features that we that maybe would have taken us 10 years from now, and we'll see them sometime next year or in Q4 or Q3. So that's exciting to me. But then there's also this risk of cognitive dissonance, right? We've we've unloaded so much to the agents that um people don't really start to, but maybe within a generation or two, really don't understand how anything works and you have the wall-y scenario, right? And I worry about that. But here's the fascinating thing about these tools. There is a skill you can install for Claude called learning opportunities. It was put together by a professor of psychology. I can't remember where um he teaches, but it's pretty easy to find the skill. And loading skills in the claude is very easy because it's on a marketplace. And basically at the end of a claude session, you just invoke this skill and it looks at everything you did in Claude and it starts to challenge you and spends a 10, 15 minute session learning about what you just did. Like, hey, why did we choose to remove this variable? Why was it better to time down this? What did why did we put the skills.md in this folder? And not only have I learned from it and it reinforced the things I did, I found flaws in my design just by it asking me questions. I was like, you know what? We really don't need that thing. We should get rid of that. And it's like, you know what? You just identified an enhancement. That is great. And the fact that the tool at the end of the conclusion of building what I ultimately want to build can have a conversation with me about what I just learned. Yeah, I don't even know how to react to that. Like, that's what's to exist today. How many other skills are just gonna either you know prop us up or or spark innovation? It's an exciting time.

SPEAKER_00 30:16

It is a little scary, but sky, but exciting. A little scary. One one I got one more thought. I get it. It's this is uh talking a lot for you being the guest. But thanks for thanks for.

SPEAKER_02 30:26

I do that. See, that's what happens. Two you get two podcast hosts on a talking.

SPEAKER_00 30:32

So Dario Amadei, um co-founder and CEO of Anthropic. Um love the last name. Uh a lot to love about that guy. I I don't know Dario, but yes, uh, there's there's uh it's there's just a lot that's admirable about the way that company conducts itself, in my personal opinion. Again, nothing to do with confluent here. Um and uh he said recently software engineering will be fully automatable within 12 months. Now I I don't know about that. Yeah, I am I am in fact a skeptic of that kind of claim, which has kind of helped me be late to the party in AI here, because I hear claims like that, like, oh, we're cooked, everything is super intelligence around the corner, the singularity. I'm like, um, which gets us back to I think you used the word intent earlier, or you got close to it. Um large language models and the agents that that wrap them um are capable of some incredible things, like we talked about. Uh certainly very good at writing a lot of kinds of software right now, not not everything and not perfect, and they'll get a little better or a lot better. Uh but one thing they don't do is want.

SPEAKER_02 31:40

That's true. They don't have a desire.

SPEAKER_00 31:43

They do not intend, they do not desire. Uh there is no real goal. I mean, you give them a goal, there's specs, and they go through the steps they love, you know, step step things. I'm just anthropomorphizing the thing they love. Um, but the intent, the desire, the goal, the thing that we want to build, we're the only ones that can want. That's right. You know, it's a fundamental kind of principle of economics that that wants are unlimited and the ability to supply them is is limited. Um this has made it cheaper to supply certain kinds of wants. We still get to intend and desire and plan and goal and build, uh, and we get a cheaper way to do it. And you know, yeah, we do have to be careful and think about disruption and and jobs that are eliminated and people that that that need help as a result, and there's that's beyond our scope here, and it's those are real things. But um this to me just refocuses my mind on on what it means to be a human. Because now you've got this machine that's not human, that's really good at using language a lot like a human, and really good at turning natural language into code, which I've always said is kind of the fundamental alchemy that programmers do. You take the mental states of like what operates a business and turn that into machine.

unknown 33:07

Right.

SPEAKER_00 33:08

And now we've got a machine that's that's that's kind of good at that. That's right. And so, you know, I've got some thinking to do about that, but it it refocuses me on, you know, the humans are the only ones that that that long, that desire, that that have a thing that they're upset doesn't exist and they want to bring it into existence.

SPEAKER_02 33:29

Right. Their only need is to execute whatever we supplied for it, but all of the genesis of creativity is still bound to us humans. And until someone can until someone can show that they've built some type of agentic or AI system that is coming up with unique thought completely unprompted, then we start to get into Skynet. That's that'll be a different technology. That'll be a different technology for podcasts. And you know, there's another aspect I just I don't know why uh Skynet made me think of Alex Hogan again. But he mentioned that they're using cursor to build cursor. And that's kind of fascinating, right? Is the fact that you build a tool, the tool gets better, and the tool makes the tool better, turtles all the way down. You know, I I don't know if there's another time in history we've ever had like a self-fulfilling, self-improving tool like that. I don't know. It's it just seems kind of magical.

SPEAKER_00 34:23

Well, yeah, I mean, in little ways, like compilers compile compile themselves. That's true. I mean, you probably use an ID to improve your ID, but machine tools, you use machine tools to make better machine tools. Yeah. Um, but what's cool, you can be like, hey, cursor, what what what sucks about you?

SPEAKER_02 34:39

I can tell you.

SPEAKER_00 34:41

Right, right. And suddenly you now have things to work on. But but you can't say cursor, what do you want to be? That's true. Yes. That is what a person does, and now we have better tools for turning those wants into real things. And I like it.

SPEAKER_02 34:56

The the optimum, the most optimistic view I can think, you know, kind of piggybacking off of what you just said, is that um we now have these tools that help us fulfill our human desires in a in an easier, faster, better way.

SPEAKER_00 35:10

My guest today has been Joseph Morris. Joseph, thanks for being a part of the Confluent Developer Podcast.

SPEAKER_02 35:15

And thank you for having me, Tim. It's wonderful. And if uh your audience enjoyed our banter, please check out Life's about a stream.