The following is a rough transcript which has not been revised by Vanishing Gradients or the guest. Please check with us before using any quotations from this transcript. Thank you.
_____
Hugo Bowne-Anderson (00:00)
Welcome everybody to a special edition, live streamed edition of Vanishing Gradients. I'm Hugo Bowne Anderson. This is Vanishing Gradients, the data science ML AI podcast. Super excited to be here today with Joe Reese. What is up, Joe? I'm well, yourself?

Joe Reis (00:13)
How you doing?

Doing good. good. second podcast of the day. Turned right in between. That's about it. Normal day. Watching a sick kid right now. So it's pretty... It happens, man.

Hugo Bowne-Anderson (00:21)
Very cool, very cool. Sorry to hear that man. There's something going around

here as well. think it's actually at Summit. So I'm in Sydney, Australia everyone at 7am Friday and I can tell you everyone who's on Thursday time, Friday is a beautiful day. It's coming for you. Joe, you're in Utah, SLC or very cool.

Joe Reis (00:38)
Utah, yeah. Yeah, we'll see.

Hugo Bowne-Anderson (00:41)
anyone

joining, please let us know where you're joining from and what your interest in such things are. Do you work in data science or AI? You work in software? And if you enjoy these types of things, please be obligatory, like and subscribe, share with friends. If you don't like it, no need to tell anyone that just don't engage, right? It's the internet. I've just brought on the trolls exactly.

Joe Reis (01:00)
The internet means that they're gonna criticize you and yeah,

that's what it is.

Hugo Bowne-Anderson (01:06)
Exactly.

The host talks too much. Deal with it. I talk a lot. Joe, thanks so much for joining me. We're living in an interesting time of, you know, polarization along some axes and mediocrity among many others. We're here to talk about vibe coding and the Great Pacific garbage patch of code slop. You've called vibe coding an anti-pattern, a joke, a part of the Great Pacific garbage patch of code slop.

Joe Reis (01:30)
Among other things,

yeah.

Hugo Bowne-Anderson (01:31)
Yeah,

yeah, what do you mean by that? What's the vibe?

Joe Reis (01:34)
I mean, it's just vibes, right? So, I mean, let's go through this, though. I think let's define for people what vibe coding is first. And I think that that's kind of paraphrasing from Andre's tweet here. Yeah, let me just read it. I mean, it's a tweet, right? an X. I'm not even on X. I dropped that shit. So I can custom your show, right? No? OK, I'll try not to. OK, so Andre writes, there's a new kind of coding called vibe coding where you give

Hugo Bowne-Anderson (01:46)
It might be worthwhile reading out his original tweet.

Yeah, yeah, yeah, absolutely. Read it out.

Joe Reis (02:01)
where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs, EG cursor, Composer with Sonnet are getting too good. also, I just talked to Composer with Super Whisperer, so I barely even touched the keyboard. I asked for the dumbest things like decrease the padding on the sidebar by half because I'm too lazy to find it. I accept all anyways. I don't read the diffs anymore.

When I get error messages, just copy and paste them in with no comment. Usually that fixes it. The code grows beyond my usual comprehension. I have to usually really read through it for a while. And sometimes the LLMs can't fix a bug, so I just work around it or ask for some random changes until it goes away. It's not too bad for throwaway weekend projects, but still quite amusing. I'm building a project or a web app, but it's not really coding. I just see stuff.

say stuff, run around and copy paste stuff and it mostly works. So I think that that encapsulates maybe what vibe coding is at least according to Audrey and how people are using it. So I want to distinguish that because you brought up a good post by an article by Simon Willis and we sort of dive into the nuance of vibe coding. So I think what this really means is sort of the I think overly promiscuous use of code gen tools to generate code, quote build stuff without putting much thought into it, just going off of vibes. Is that how you read this?

Hugo Bowne-Anderson (03:19)
Yeah.

But I do think it'll be worth exploring at some point in this conversation the spectrum of using AI assistance from, you know, code completion to embracing exponentials and also the type of software that it allows us to create because I think we're in a space where on one end we have grifters who are like, SaaS is dead and we're gonna fucking kill Salesforce by vibe coding. That's clearly bullshit. On the other side, we have, as you would put it, old men shouting at clouds. So, and I'd love to explore

Joe Reis (03:21)
Ahem.

Absolutely.

Hugo Bowne-Anderson (03:48)
or the fertile ground, perhaps give us your take on kind of what the space looks like at the moment, maybe on LinkedIn as well, where we've had these conversations.

Joe Reis (03:55)
Yeah, yeah, so it was one of things where initially when we were talking about vibe coding, I almost didn't want to do this talk because I'm like, can't believe we're actually going have a serious conversation about it. But then I was like, OK, it's the zeitgeist of at least the week until we find something else to get upset about. But I think it more highlights an issue that I've seen with skills as it pertains to code gen. I think vibe coding is sort of

an extreme of this where you're just sort of abdicating all of your quote skill or your lack of skill to AI. But if we take a step back, I mean, this is nothing new. I think it was Aristotle or somebody who was complaining about writing way back in the day and that that would remove the need for humans to remember things and think. So we've been here before and people criticize the printing press for much the same reason that it would mean that writing is

democratize and I think there were some political reasons for those comments too. So history's not, you know, it's, it's, it can be unkind, I would say to new innovation. So I'll get that out of the way. But why do I say the things I do about vibe coding? What do I say? Well, I think, I think in it to its fullest extent, it's an anti pattern if you're going to take it seriously and use it in everyday use. I have no problem if you're vibe coding to toy around with an app. Like we all do that. We're all

Hugo Bowne-Anderson (04:59)
And what do you say?

Joe Reis (05:13)
dorks. So I think we.

Hugo Bowne-Anderson (05:15)
And we've done that historically, right? We just haven't used LLMs, right? Like I've vibe coded shit by Googling constantly, like to try to figure out how to build a React something where I don't care, like I don't want to learn React. Yeah.

Joe Reis (05:17)
We always do that. No.

All the time.

That's it. I mean, we've all done that,

right? So, but I think to take it to an extreme now, the comments that keep hearing are we're sort of in this new age where anyone can build apps. And so, you know, and people point out examples like, yeah, my hairdresser, she didn't even know how to make an iOS app and she made a great app. And I'm like, that's great. But let's distinguish between like things that are apps and projects versus things that go into production that are machine critical systems. And I think this is where I'm starting to see the conversation sort of get lost is we can just vibe code.

Hugo Bowne-Anderson (05:35)
Hmm.

Joe Reis (05:54)
Seemingly anything, even in production, like I've had serious conversations with people. They're like, well, just vibe code. I'm like, are you, like, you aren't even worthy of calling yourself anything remotely close to an engineer if you're going to do that in production, right? I think that's where I got a bit, you know, peeved the other day about it. but it is, it is definitely a spectrum, but I would say it's got it's time and place. I think we all like to quote vibe code with a.

AI these days. mean, I do my fair share of it on the side, but it's mainly it's always for toys, right? And it's for usually use cases where I maybe I'm not totally familiar with that language or, you know, platform or whatever. And it definitely has an assist. But the thing is, people like you and I know how to write code. So you know sort of what you're looking for. Like and you know what anti patterns are when you when you see them. Right. So there's this old term called code smells. If you're familiar with that, I it came out a long time ago, but that's

Hugo Bowne-Anderson (06:38)
Mm.

Absolutely.

I am Yeah.

Joe Reis (06:47)
That's one of the things where you can kind of smell pretty bad code. So, you know, I think that's the big difference.

Hugo Bowne-Anderson (06:53)
Totally,

and I like the distinction between doing stuff for production use and serving users with security, authentication, all of these things, but I...

I probably would push back slightly on not having it for everyday use. what I'm thinking of an example, and this is one I did a live stream yesterday where we chatted about examples like this, but several people recently have told me they'll like spin up bespoke JSON viewers because there isn't one that's really built that is good for their use case. Now, when they do that, they'll need to, they want to do some, some queries on them, that type of stuff. They'll then write a few tests to make sure that they're getting what they expect. So they're not necessarily exploring all the code, but they're writing a few tests

around it. Why that's interesting to me is because I don't come from the software side, right? I come from the data and machine learning and science side. My background's in biology and physics, right? So I do this type of exploration and experimentation all the time. So having tools that I can spin up and spin down in order to serve my ephemeral purposes I think is incredibly useful. I'd love your take just on just how, once again, not to productionize. Of course, if it's something cool that then we want

productionized because it may help external people then I start to explore that but I suppose I want your take on just the increased surface area of what type of software is possible now and I want to frame it a bit historically that one of the reasons we require like such

It's software as we know is incredibly complex to engineer, to build, to make scalable and robust. And it has to handle a lot of edge cases due to the economics of it. Because you have so many software engineers and so much investment in it, you need to serve a lot of people and get significant investment. If I can spin something up myself for my own use, I don't have to do that investment. I don't have to worry about edge cases. So I suppose it's empowering to consider the surface area increase in software development and the surface area of who can do it as well, right?

Joe Reis (08:42)
It is, I would say, the caveat that you know what you're doing. This is a central question, right? And this is a central topic where, as Travis Smith points out in the comments, you have to have the foundational knowledge first, whether it's software or data programming. That's absolutely true. Without that, you can go ahead and spin up the biggest, most complicated system that covers all the edge cases possible. But the question is, how would you know that, one, you've built something that serves the needs of hopefully your users or the systems that you're

Hugo Bowne-Anderson (08:45)
Mmm.

Joe Reis (09:09)
building for it. And second, how would you even know what an edge case looks like? Right. mean, it's, it's, yeah.

Hugo Bowne-Anderson (09:14)
Yeah, so exactly.

And to my point as a data like science nerd, I know what the edge cases will look like for the data stuff, but maybe not all the ins and outs of the software I'm building. So, and we do that in data science and machine learning all the time. I mean, we deploy things, roll them out slowly, deploy them internally. Then we see how they're operating.

in production, look at monitor drift, these types of things, and then recalibrate then. So I suppose when building AI powered apps, there's a kind of, it just feels a bit similar to me is what I've been doing for a long time. Whereas I think you're, presume you're a far better software engineer than I am because I've learned the tools in order to, yeah.

Joe Reis (09:38)
yeah.

I've never seen you self-warned. But the thing is,

it's like, I mean, there's more to it than just writing code, though, right? It's how you think about system design and architecture and all that. And that just requires, I think, cycles of just knowing what works and what doesn't. What I'm saying, though, what you mentioned with data scientists, that's reason why ML engineering and MLOps got into the picture is because most data scientists, I mean, you're good at, I think, using code to

Hugo Bowne-Anderson (10:02)
I love this so much. I, cause, sorry please.

Joe Reis (10:15)
describe data problems and solve those. But the production aspect that you described, that's the entire reason there's a whole ecosystem of tooling and support out there. I to be fair, lot of data scientists, know they can write code, but that doesn't necessarily mean that you know how to write code, if you know what I mean. So that's not a knock against using code to leverage your skills in certain areas. GlassNet has had a meetup

in Salt Lake City and the presenter was talking about was it PGML, so Postgres ML extension and it's pretty dope. I don't really have a use for it, but you you're able to do XGBoost and SKLearnal and Postgres, right? But to use, I mean, it makes it simpler to use, but you still gotta understand what you're doing, I suppose. So it's, yeah. So I guess that's the whole point with vibe coding, right? I think it's, because one of the discussions I'm hearing more broadly is it's just like,

Hugo Bowne-Anderson (10:58)
without a doubt.

Joe Reis (11:04)
You know, we're going to have this new era of software where anyone can write code. Anyone can write code. That's just it. Or anything can write code these days. That's the whole point. But just because you can, you know, does it mean that, you know, well, I'm not going to tell you what you can and can't do. Go for it, right? But I would say the old expression is, you know, fuck around, find out. And that's kind of what you're going to be dealing with. Like every

Every engineer that I know who's been in the game for a bit is excited about this moment of VypCoding. We're we're so excited because this, the amount of technical debt that's about to be unleashed on the world is, you know, it's like, so I equate it to the Great Pacific Garbage Patch, right? And so for context, reason I think about it in this framework is, you know, if you go back to how things were crafted back in the day, there was a sense of quality, there was a of craftsmanship, there was a sense of,

pride in one's work, right? And I think as we became, you know, much more of a commercial society where things, you know, started making everything out of cheap plastics, and everything was disposable, right? And that, you know, there's no free lunch here. Things go when you throw things away. Sometimes they end up, I don't know, in a landfill. Sometimes they end up in the ocean, right? And so the Pacific Garbage Patch is like, I don't know, 1.6.

million square kilometers, something like that, 1.8 trillion pieces of plastic and counting 99 % of it is plastic. And I think to me that represents just sort of the consequence of this disposable culture that we're in, you know, where we can just throw things outside unseen, you don't see the garbage patch. But it's ironic because as we adopt these practices like Agile and DevOps and XOps, whatever, these are these take their inspiration from Lean.

Hugo Bowne-Anderson (12:22)
Something like that.

Joe Reis (12:48)
And the whole idea of lean is actually reduce errors and defects and produce things of high quality in a repeatable, you know, fast way. And this has been lost, I think, and the industry, like we think that we're being agile because we can pump out more code. Whether or not the code adds any value or not, and whether or not it's actually, you know, going to produce more debt down the road. This is besides the point. Nobody seems to care about this. It's how many lines of code do you ship? And so I think this is the danger we're now where you can, can,

The marginal cost of producing code is essentially zero at this point. Just like when the internet ushered in content, every newspaper in the universe just died, pretty much. Slow death, but it's not relevant because the cost of distributing content is zero now. And the same thing is going to happen with code. so what they obviously, the side effect of this is the great Pacific garbage patch of Slopware, I suppose, Slopware was actually set in an urban dictionary in 2005.

Hugo Bowne-Anderson (13:18)
Absolutely.

Mm.

Joe Reis (13:39)
but it means much the same thing. but yeah. And then somebody's mentioning security here. And this is something I'm talking to security people about too. Sorry to cut you off, but like security is a big, thing where everyone's like, this is like the big, big thing. Like the consequence of getting a code wrong and it blowing up, I mean, that depends on what it's used for. The consequence of security errors, far, far greater, right? Most people don't understand security to begin with. And now you take an untrained person trying to do a...

Hugo Bowne-Anderson (13:40)
Awesome. Awesome. I also.

No please.

Joe Reis (14:02)
Hackey software and put that in quote production. It's going to be a pretty exciting time.

Hugo Bowne-Anderson (14:06)
Definitely, mean, across a lot of dimensions in a lot of directions. do, there are several things in the chat I want to get to a point, Brian, but there is something I want to mention first. do like how you framed it in terms of around programming, building software isn't just about writing code, it's about designing systems. And I do think there's a big opportunity here.

for us to actually, this has revealed kind of what the job really is. And I'm actually, I've posted in the chat a wonderful essay by Tim O'Reilly called The End of Programming as We Know It, which you and I actually corresponded about when it first came out. But he actually mentions an email from our mutual friend Chip Huynh, the author of AI Engineering Among Just Being Legend. And Chip wrote to Tim and said, I don't think AI intro...

uses a new kind of thinking, it reveals what actually requires thinking. No matter how manual, if a task can only be done by a handful of those most educated, that task is considered intellectual. One example is writing, the physical act of copying words onto paper. In the past, when only a small portion of the population was literate, writing was considered intellectual. People even took pride in their calligraphy. Nowadays, the word writing no longer refers to this physical act, but the higher abstraction of arranging ideas into a readable format. Similarly, once the physical act of coding can be

the meaning of programming will change to refer to the act of arranging ideas into executable programs. So I wonder from your perspective how AI assistants can play into this and if at least initially prototyping with vibe coding can actually help us with this new paradigm of how we think about building software.

Joe Reis (15:39)
I mean, it's interesting because I'm using, you know, a few elements right now to help with my own thinking about writing my book. Right. And so what I think it's uncovering for me is all the blind spots that I have and what it's uncovering for me is all the blind spots of my friends. So, you know, with my book, I'm in a fortunate position where if I have a question, I can go ask probably the world's top expert and have them review what I wrote. That's great.

Hugo Bowne-Anderson (15:49)
Mmm.

Yeah.

Joe Reis (16:01)
And they have a lot of context. say what the challenge of that is, you know, I'm noticing the LLMs are catching things that even top experts are missing sometimes in conversations. And so that's because it's, you know, it's been trending the entire Internet until it sees maybe some nuance that even experts and obviously experts know a lot of things that LLMs don't. So I think it's definitely complementary. Like I'm using it to push my thinking forward on lot of topics. I was on a plane back from Europe last week and I was chatting with Claude for

Hugo Bowne-Anderson (16:22)
Hmm.

Joe Reis (16:27)
for hours about actually vibe coding and just the issues of it. This is before our conversation actually. I was asking Claude, what's the consequence if everything is basically, if all data in the systems and code or just all generated by AI. So I'm thinking about the logical extension of where companies are saying that 99 % of code is gonna be written by AI and agents and so forth. I'm like, okay, so.

At the biggest extreme here, what does this look like and what are the consequences of this? Is this a good thing? Is it a bad thing? you know, Claude brought up lot of good points either way. It's like it could be good. Here's the downsides, enormous amounts of technical debt. But then the counter argument as well, but wouldn't AI just like catch the technical debt and fix it? So it's revealing a lot of questions, I think, to Chip's point, under the underlying ways we think about things. But also, I think we've lived in a world of

Hugo Bowne-Anderson (17:00)
Mm.

Joe Reis (17:20)
serious constraints, mainly labor constraints in terms of the amount of code that we can generate and things that we can build. I don't think we're anywhere near there yet. And if you talk about enterprises, for example, it's like the sheer amount of complexity in any enterprise code base is insane. It's not fathomable even by probably a single person. So maybe

Maybe AI does it. I don't know. But I think that it's not there.

Hugo Bowne-Anderson (17:45)
meta point you made earlier, currently the way we work and the incentive systems are kind of screwed. I'm going to paraphrase, but you spoke about wanting people to take pride in their work and that I have paraphrased that but we do live in a kind of a skewed system where a lot of us a lot of people are forced to do jobs that they're not able to necessarily take pride in and do tasks that don't serve them as individuals in corporate structures. I don't want to get to like I'm not a punch card, but there is that that that vibe. The other thing is there are incentives.

particularly in software, or least ML and AI, to build and not maintain, right? So that we, you know, the amount of technical debt, spaghetti code out there and...

Joe Reis (18:19)
yeah.

Hugo Bowne-Anderson (18:24)
essentially unmaintainable systems and castles of sand is super concerning. Brian Bishop had a really good point, which is the original tweet from Carpathi's specified side project. So it's not clear to him that this isn't all a straw man. And I suppose I would say I agree with that take from Carpathi's tweet. But I think what's being exposed now is that, of course, like all the people saying we can build SAS by vibe coding, that's nonsense. I do think there is some middle ground. And that's why I brought up

Joe Reis (18:28)
tons.

Hugo Bowne-Anderson (18:51)
example of bespoke JSON viewers and that type of stuff and run some tests yourself. So I do wonder what else it allows us to do. And William Horton in the chat spoke to this. He said he thinks production might be ill-defined. And I actually agree with that. mean, for me, working in ML, productionization is a huge, there's a graded approach to it where maybe you deploy, then scale, then have more reliability, then do online experimentation.

Joe Reis (18:55)
Yeah.

Hugo Bowne-Anderson (19:19)
then horizontal scaling, whatever it is. And so he gave the example that he vibe coded a chess training tool that's deployed on the internet used by a few of his friends. So he doesn't know why that wouldn't be production or partial production. It's a real live app on the web with users. It doesn't necessarily have security authentication or all of these types of things. But I suppose what we're exposing now is just kind of new.

ideas of what productionized software can be and it can have more characteristics, right?

Joe Reis (19:47)
Yeah, it's an interesting question. suppose, you know, I would find productions anything that's being used by an end user. So congrats on writing a production grade app with FIBE coding. But you also got to assess it on the spectrum of risk. So what's the risk for a chess app? Who cares? They could put it on, you know, probably this Raspberry Pi and hook it up and nobody would really care either way. Right. So I think you got to assess like what's the risk profile.

Hugo Bowne-Anderson (20:06)
So yeah.

Joe Reis (20:09)
But what, and Brian's comment about Andre's, you know, saying this for projects is correct. That's why we read it, the tweet before we started, so we can establish exactly what this is. So, I mean, those are all fair points here. The interesting thing is, again, it's, what usually happens with these kinds of things and what I see happening with vibe coding, you know, because what was it? The tweet was from February 25th, so we're not even a month into it.

Hugo Bowne-Anderson (20:19)
Mm.

Joe Reis (20:35)
But things get distorted, right? Because it's sort of the game of telephone after about the third call, whatever, not that people use calls anymore. you know, every bit of information is handed down and then pretty soon people are saying, yeah, we should totally do this in production for jobs. these are serious conversations that I've had with people who saying, well, why couldn't I just do this for my job, actually? You know, I'm like, OK, I guess you should try it. You should try it. Go for it.

Hugo Bowne-Anderson (20:41)
Mm.

Yeah.

and

see what happens. So I actually, I've pulled up and pasted in the chat a recent blog post from two days ago by Simon Willison. Actually, I just posted on Twitter that we were going live and he was like, hey, what does Joe think of this? And so in this post, Simon Willison, Simon wrote, does Joe consider all forms of AI assisted programming to be vibe coding? Yeah.

Joe Reis (21:06)
Yeah.

Wait, who asked that?

That's what I think about it. Weird.

No, I don't think

so. I think it's how Andre defined it, right? So I don't think that I would, I'm not going to distort what Andre put out there. I mean, he at least coined the current manifestation of this charm. So I'm not going to distort it for my own gain. mean, it's what, no, I don't, I don't think so. I mean, I agree very much with what Simon said, there, Simon says, there's definitely a spectrum of it. I mean, I use AI coding all the time for stuff. I think that it's,

Hugo Bowne-Anderson (21:40)
You

Mmm.

Joe Reis (21:46)
And I, you know, on the weekend, I also used it and I found myself having to correct like 90 % of its output. you know, I was like, and at the end of the day, just did it by hand because I knew what I was going for. You know, but that's, but that's in Python, right? I can write Python, you know, like pretty much like blackout drunk at this point. Like it doesn't matter. I've been doing it for so long. It's like, or SQL, it's the other one, right? Like it would be really hard to throw a SQL problem at me that I probably couldn't figure out. But that's because I've

You know, but I also know like when it was giving me output for Python, was like, that's this isn't correct, nor would it meet, you know, standards, right? Like it's given me like, you know, in this case, I had to spin something up. It's like, here's here's the Postgres connection, connection credentials in like the code and not in like a secret file. And it's like, OK, I'll just do this myself. Thank you very much. So it's you know, but it was interesting. So, you know, I gave this talk in Austria about

Hugo Bowne-Anderson (22:27)
Mm.

Yeah.

Joe Reis (22:36)
almost two weeks ago now. Where it's very much improvised talk and it was where I started to formulate this idea of the Great Pacific Garbage Patch where it's even without the notion of vibe coding that aside, what I see happening right now is, you know, I think there's unencumbered use of AI, co-gen tools and data gen tools and tools where you get, you know, I see some vendors have tools where you can spin up data pipelines and data infrastructure with an LLM now. All great on the surface.

Hugo Bowne-Anderson (23:02)
Hmm.

Joe Reis (23:04)
Why wouldn't we want to do that? It makes life a lot easier. But what I sense is going to happen as these tools proliferate is just the sheer amount of, like you said, just cruft and things out there that are running in, I guess, pseudo production. People rely on them, but they're creaky, not maintainable. the thing is because you can always see studies in this.

cognitive abilities are going down. is way before AI, but like our ability to critically think through problems is declining. There's countless studies on this right now.

Hugo Bowne-Anderson (23:32)
I'm concerned about our

ability to read maps. Like we can read local maps, but now thinking more distributively, our map mines are shrinking.

Joe Reis (23:35)
Yeah.

What is? Yeah. But the

consequence of this, though, is if you fast forward what the industry looks like in 10 years, is there a sense of craftsmanship at all? Does anyone know what the hell they're doing? Or are they all just like these old time cold ball programmers who are, you know, about retirement age and have to be dusted off for half a million a year just to go, you know, fix some creaky code that nobody understands, right? Because the assumption is, well, know, AI will just do all this stuff for us in 10 years. I think that's the because AGI will be here.

These are legitimate things that I hear serious people say, by the way. It's not just me saying, yeah, it's crazy.

Hugo Bowne-Anderson (24:11)
Yeah.

I do as well. Yeah. And that definitely

isn't my take. I personally, I don't even think AGI is well defined. I think we probably can like give some dimensions and that's why I find the arc at AGI stuff relatively interesting. But I think the term AGI is a misnomer. There's a religiosity about it as well. It's like, let's recreate ourselves. Let's be gods, right? And fly to fucking Mars and upload my consciousness to like AWS.

Joe Reis (24:26)
Mm-hmm.

Yeah. Oh, it's a cult.

Yeah

I mean, these are all the things I grew up on as a teenager, right? I think I'm a bit older than you. I grew up on all the cyberpunk novels of the 80s and I used to read, I grew up on Kurzweil too. And it was like, I just grew up with this notion that your mind would be uploaded to some internet at some point or something, but I don't know that I want that anymore. Somebody brought up an interesting question though. Like, you drive anywhere without using a navigator? I think that that's one of the, that sort of, I think drives home one of the points that we're talking about.

Hugo Bowne-Anderson (24:41)
Yeah.

Sick.

Amazing.

Yeah.

Mm.

Joe Reis (25:05)
Could you, do you think you could drive, do think you can navigate without a GPS?

Hugo Bowne-Anderson (25:08)
Dude,

I did something really dumb a few years ago. I got a dumb phone. I was so addicted to my smartphone. I got a dumb phone and used it for six months. It was brutal, man. Like it ruined my social life. I was single at the time and trying to date. It was like one of the most intense periods of my life. it made me retrain myself. And this is actually interesting. What I'm about to say I find interesting. Road signs are actually pretty good.

to be honest, at least in Australia where I am. And there's my ability to drive, but also if we all use GPS enough, maybe counties and cities aren't incentivized to keep up public signage as well. So what does that mean in terms of like the infrastructure we have, right?

Joe Reis (25:48)
good example of this. was in Costa Rica in January and they don't as far as you can tell they don't even have like road signs. Everyone uses Waze. Not even Google Maps. I don't know. Yeah I tried that and it wasn't as good as Waze but everything is sort of yeah just lines on an app. If you don't have Waze you're going to get very very lost there.

Hugo Bowne-Anderson (25:55)
Mm-hmm fascinating. Yeah

It's

interesting that it's ways as well. And I think about that in Sydney, because I came back, moved back to Australia during COVID from New York where seamless is used a lot for food delivery, right? Moved back to Sydney and it was all Uber Eats. And I'd never been in a city where Uber Eats was the major. And I think, you you get somewhere, you harness or exploit network effects and suddenly race to the bottom as well is what we're seeing.

Joe Reis (26:16)
Mm.

Interesting.

It's

interesting. Yeah. But I I grew up in an era, you know, I grew up in the mountains in Wyoming. So I mean, I can use a compass and a topo map just fine. Right. But that's like, if you show that to most people these days and just go, you know, go on a hike, you get these tools, you don't get your phone. Good luck. I mean, it'd hard.

Hugo Bowne-Anderson (26:34)
Mmm.

I remember growing up, remember

we had, it's the Gregory's, the phone, it's like the, it's the map in the book, right? Whatever you, the drive in the car. And I remember mum used to direct dad all the time or vice versa. And I remember when I was like five or six, was like, can I, am I allowed to give dad directions now? And I was like, that was, that was super exciting, right?

Joe Reis (26:57)
Yeah.

Hugo Bowne-Anderson (27:04)
I do want to just go back to Simon's post and I apologize for everyone who wanted a bit more spice. Joe and I agree on most things as is quite common but...

Joe Reis (27:04)
It's interesting.

Wrong, wrong. I'm just kidding. mean,

yeah, I don't know. mean, yeah, go back to Simon's Post. I think it's one of those things where I, it's funny because the conversation that Hugo and I were having via text, I think was a lot more heated in the sense where I think a lot of things just come across differently with no verbal cues or context. I can also come across as a bit rough on text. I do apologize, but it's what brought us here today. So it's fun.

Hugo Bowne-Anderson (27:29)
Mmm.

It did

and look, it brought us together. I do want to say, because these, think these are points we've circled around, but from Simon's post explicitly, when he thinks it's okay to divide code, projects should be low stakes. Okay. Think about how much harm the code can do, right? Consider security, watch out for secrets. Think about data privacy, be a good network citizen. Then a very important one that I think is, we haven't talked about yet, is your money on the line. Yo, if you're building, trying to build LLM stuff and you're pinging open AI,

guaranteed your LLM assistant, if it's allowed to execute arbitrary code, you're in YOLO mode will fucking like, you got to put caps on that stuff, right? Or you'll get caned, you'll get caned. But I do, yeah. I suppose, are there any other, cause we've had so much interesting conversation in the chat. Actually, I've,

Joe Reis (28:13)
yeah, yeah, or you're gonna get capped.

big one.

Hugo Bowne-Anderson (28:24)
I'm interested if you were the BDFL, Benevolent Dictator for Life, king of software engineering practices around the world, what are three rules you'd set for how software engineers use AI tools now?

Joe Reis (28:37)
It's an interesting question because as you're saying that I was thinking about getting a driver's license and knowing the rules of the road, right? But then I'm thinking about self-driving cars. And well, you don't need to know any rules of the road. guess it's already been programmed in, right? But you still need to, I guess, have an idea of it. Do you know where you're going if you get a self-driving car? So.

Hugo Bowne-Anderson (28:44)
Mm-mm.

Mmm.

Yeah. And how

do you handle failure modes? Maybe with a self-driving car you actually do need someone who has a license to take over in the 1 % of cases or one in 10 to the minus 5 % of cases that it could actually kill someone or people, right?

Joe Reis (29:12)
Yeah,

exactly. yeah, I mean, it's one the things where I would say at least understand what you're doing. mean, you know, it's this is something I've been talking about a lot for years, really. But I don't think we have any shortage of great tools at this point, great technologies. There's maybe too many, in my opinion. But our ability to use those tools to the fullest potential, think, you know, lives and dies, but our ability to understand how to use them to their fullest potential.

Hugo Bowne-Anderson (29:26)
Mmm.

Yeah, agreed.

Hmm.

Joe Reis (29:38)
And right

now, I'd give the industry, whether it's software, data, machine learning, I'd probably give us about a D in that area. mean, having run a consulting company, I always joke that I should just be quiet about this because that's what you pay me to go fix. And if you're a consultant like yourself, there's a lot of money to be made by fixing the mistakes of others.

I truly believe that still it's a knowledge gap problem at the end of the day. And it's only going to get worse actually, not better. So I'd say that if I were a benevolent dictator, I would just make sure that there's a standard body of knowledge that anyone who wants to call themselves a professional in this industry has to learn, I would say, in their respective crafts. So.

That's the one thing I would do then from there. You know, feel free to make your own mistakes as you see fit. But until you know, until you have that body of knowledge and the skills, how would you know? How would you know that anything works? You know, and I think the other part of it is map this back to what the businesses need. Right. I think more than ever, if you don't know what you're building and you don't even know what the business needs, like this is a recipe for disaster. Huge, huge, huge disaster. But this is exactly what's going to happen.

Hugo Bowne-Anderson (30:32)
Mm.

Yeah.

Joe Reis (30:47)
And of course, every CEO in the world is saying, we need to replace all of our workers with AI right now, not even including software, it's everybody at this point. So we're in a very interesting time where think that, know, it's what was interesting. Axios had an interesting article covering a survey that was done by Rider AI on just the disconnect between what C levels think about their AI capabilities and progress so far versus what employees.

Hugo Bowne-Anderson (30:53)
Mm.

Joe Reis (31:11)
Think about it. It's fascinating. I should find this and put it in the chat here. it was, yeah, it's enterprise AI attention workers, execs and stuff. But it is fascinating to read through. Essentially what's happening is it's the C-suite, probably for reasons that they want to look like they're accomplishing a lot of stuff with AI, say, well, we're accomplishing a lot of great stuff with AI. Right? But then you talk to the workers and like,

Hugo Bowne-Anderson (31:14)
Yeah, please do.

Joe Reis (31:36)
No way in hell or we're not even close. The data sucks. Systems are creaky nowhere near doing AI. So that's the that's the tension right now. Or employees actively undermining AI initiatives because, well, I don't know. I think somebody called it. What was the quote here? You are what was it? Just to hear.

Yeah, asking a turkey to vote for Thanksgiving. It's like asking employees to embrace AI. So it's an interesting study and I'm sure everyone wants to look at it, but this is sort of the crux we're in right now where there's a lot of pressures. I did. I put it in the chat for the YouTube. Yeah. Yeah, it's interesting. This is where we are, right? I mean, there's not an executive in the world who's not saying again, I talk to people who are outside of tech, by the way.

Hugo Bowne-Anderson (31:57)
Mm.

Fantastic. You able to share the link at some point? great. I missed it.

Joe Reis (32:19)
I do try and live outside of our bubble, but like white collar workers across the world, knowledge workers are getting pressure to, you know, embrace AI. You know, my wife had this happen to her at her job where the boss pulled her aside and said, Hey, can you help us like understand where we can put AI into all of our processes? And she's like, I see where this is going.

Hugo Bowne-Anderson (32:34)
I, this is the first

time after our quote unquote chat GPT moment, that I've had like family and friends reach out to me and people I chat with all the time, but people actually reach out to me about, about work, like people working in marketing or lawyer, lawyer cousins. And that type of stuff being like, Hey yo, what, like what's happening? actually some of my really sharp teacher friends, like pinging me about it all the time as well. I want to go back to the analogy of

Joe Reis (33:00)
What are they talking

about? are the questions they ask you?

Hugo Bowne-Anderson (33:03)
on the...

So on the legal side, they're very interested in what can be automated and what can't and how trustworthy tools can be on one side. On the other side, they're asking about regulation. So I've got some family members actually who are pretty high up in the legal system here and they're actually on councils and committees to figure out how to regulate AI. so...

Joe Reis (33:25)
interesting.

Hugo Bowne-Anderson (33:26)
been trying to pass some certain things, which will mean that in certain legal situations, no AI can be used. And I was like, well, that, I mean, that seems quite, we need to define certain things because...

If I send an email, right, or if I'm using Grammarly to correct stuff, is that using AI, if it's AI powered and these types of things. So I think having a more nuanced idea. But to your point, there are so many tools, so much software and the movement's so quick that it's actually incredibly tough. The teachers are at schools where use of AI is being banned and students are submitting stuff that's clearly...

Joe Reis (33:46)
Yeah

Yeah

Hugo Bowne-Anderson (34:00)
if not generated by AI has a bunch of AI in it. And of course you and I know that AI can be incredible for ideation and then filtering as well. I think the combinatorial explosion of ideation is one thing, but not enough people use it for then crafting and honing and filtering.

Joe Reis (34:07)
yeah.

Well, they're not incentivized

to it. It's like, me a term paper. It's like, fine, I'll give you a term paper. Don't worry about it.

Hugo Bowne-Anderson (34:20)
Well,

that's the other thing. As I said to my teacher friends, was like, yo, this isn't, this actually isn't, this is exposing a huge problem with the system that why are we even asking students to write essays? What's their incentive? Their incentive from like fucking like middle school to the end of high school is to get the best grade possible. I don't know what a GPA or whatever it is there, or maybe that's college, I don't know. But then it is college actually. Then after that, so they're incentivized to get the highest mark possible.

not incentivized to learn right

Joe Reis (34:49)
Yeah,

that's exactly it. It's interesting. My sister's a lawyer. So I asked her about how she's using LMS there. And it definitely helps with some things, I think, speeding up doc creation and stuff like that, or potentially researching. But still a lot of errors with it. But then I was thinking, what are the similarities and what are the differences? If you were to replace Vibe coding with Vibe accounting or Vibe legal work, what would that look like?

Hugo Bowne-Anderson (34:53)
Hmm

Mm.

Joe Reis (35:14)
Or Vibe anything else, it's not coding. What are the similarities and differences here? That's something I've been thinking a lot about this week is like, what are the boundaries? Where if you were to say, let's just Vibe out. so think part of it is a creative pursuit. And that's a big distinction where Vibe writing is probably something that's interesting. Can't copyright any of it, by the way. Versus maybe Vibe loitering.

Hugo Bowne-Anderson (35:26)
Mm.

Joe Reis (35:35)
And so let's just think of an interesting case here. I guess there's maybe some utility there.

Hugo Bowne-Anderson (35:41)
people are

doing it but I think the good lawyers are humans in the loop and they're able to then like it it supercharges them I think we've got lots of 10x lawyers these days Joe I'm so sorry

Joe Reis (35:51)
10x lawyers.

10x ambulance chasers, thing in Australia.

Hugo Bowne-Anderson (35:57)
Not as much because we don't have the I'll sue him vibe that America has. Or the 47th and 45th president of the US who whenever it's like hey someone said this about you I'll sue him. I'm sorry I didn't mean to degrade the quality of the conversation by bringing up old mate. But actually when I lived in New York one of my favorite subway ads was

Joe Reis (36:03)
Yeah, I guess that's maybe an American thing. kind of like a better call Saul. Yeah, yeah.

Mm.

You

Hugo Bowne-Anderson (36:24)
for a divorce lawyer and it said the phone number was 1-800-DIVORCE and the tagline was for when diamonds aren't forever. Yeah, I'm a huge fan of that. My other favorite of recent times marketing campaign is the city of San Francisco with Alcatraz. They decided to do a marketing campaign a few years ago, escape to Alcatraz to get tourists there.

Joe Reis (36:32)
That's pretty good.

Hugo Bowne-Anderson (36:47)
which is wonderful. am also, it's something we've been talking to, like lawyers I know who can use these tools, you know, can use it to do the types of things that usually have paralegals or interns or students help with or junior assistants. One thing I'm actually very concerned about is what this...

Joe Reis (36:47)
That's good.

Hugo Bowne-Anderson (37:05)
what career paths this carves out. And I think we're seeing it in engineering as well and data science. We've already had an issue with entry level positions in data science and ML from the get-go. But now, how are we even gonna learn things if all the people who've learned things are building and they don't really need to employ people to help them to establish careers? So what happens to entry level positions now?

Joe Reis (37:28)
That's the big, big question right now. That's the big question. I everyone in the world, I don't have an answer. I don't think anyone knows just yet what happens, right? There's a lot of factors at play. I mean, it's a competitive job market as it is, even if you're senior right now. If you're senior, mid-level senior, even whatever, it's like you have a lot of factors working for and against you. I was talking to somebody at one of the big...

Hugo Bowne-Anderson (37:42)
It's tough out there, yeah.

Joe Reis (37:53)
tech companies yesterday and they're going to be doing a massive amount of layoffs really, really soon. You'll see it. I'm not going to name which one. You'll find out. That means a lot of very qualified people are already on the street on top of all the other layoffs they just did. That and other companies, right? So if you're a junior in this market competing against that and AI, I don't know what you do. This is a good question. So I think it's going to redefine the notion and what a junior is and what

this looks like. What does it mean to start your career off? I got two kids who are thinking about this a lot. You know, and I don't, sadly, I don't have an answer for them. Right. And that's not because I'm a crappy parent. It's because I just this is just one of those inflection points that you're in. We've been in here before, but not quite to this level where you have things that can replace knowledge work to the degree it does. That's, you know, so it's actually Jason brings up a good point. What's what's up?

Hugo Bowne-Anderson (38:32)
Mm.

Joe Reis (38:41)
A thought experiment to understand if junior engineers need to be junior AI agent managers to get ahead. Well, it's interesting actually. So one thought experiment I've been doing with respect to the broader labor force is if we assume that agents are these things that we're just going to be using, this is the first year of agents, we're going to just go for it and everybody has agents.

If you kind of zoom out, it sort of makes you wonder what the nature of a firm is to begin with, the nature of a firm being what Robert Coase defined in his economics paper. But I do see this as potentially being as impactful as what happened with the internet where it displaced or definitely changed the dynamics of physically-based businesses.

Hugo Bowne-Anderson (39:06)
Hmm.

Joe Reis (39:22)
So two things could happen. I actually think there's going to be more of a need to go back to the world of Adams, I suppose. I've been telling my kids, why don't you just go find a trade as you can do? For one, I'm disillusioned personally with the idea of white collar work. think a lot of it's pretty silly and doesn't really do anything. just I think given the open disdain that employees and employers have for each other right now.

Hugo Bowne-Anderson (39:32)
Mm.

Joe Reis (39:47)
I don't know why you'd want to go to school for that, depending on what you're doing. Right. So, yeah, I don't know. I don't know what this does to juniors. I think that's the big question. But what is it? think more broadly, what does it do to everybody else? Because it doesn't say anyone's really exempt from what's happening. You just get, you know, mess with in different, different ways. But it is discouraging. I talk to a lot of students around the world. I actually lectured a lot of colleges and filled a lot of questions and people are asking, you know, what can I do?

Hugo Bowne-Anderson (39:50)
Totally.

Joe Reis (40:12)
So I just think that it's, but if you look at what's been sold to students too, it's right, know, go to college, get a job in these hot tech fields and so forth. And I just think that it's, don't know, college, yeah, especially in the US, tons, right? And you can't bankrupt yourself out of that, by the way. So it's, I know, it's interesting. I don't have an answer. What do you think?

Hugo Bowne-Anderson (40:24)
Get college debt as well.

I don't have an answer. I think one way I think about it is difficult to see outside the storm when you're in the eye of the storm. And I think we have, we've been in a very complex society. think, I can't remember what the distinction is here, but I think we've moved to a chaotic society in many ways where I think in a complex society, you can do tests and pressure testings and see results that then you can perhaps.

Joe Reis (40:39)
Yeah, it is.

Hugo Bowne-Anderson (40:53)
thinking vectors from and kind of see the future a bit better. But at the moment, it's become so chaotic that prediction is incredibly challenging. I also totally agree that the relationship between employers and employees is unsustainable. So I think...

generally power structures and people subject to them is unsustainable. Like look what's happening to scientific institutional knowledge. Look what's happening to governance and relationships with governments more generally. One thing I am very interested in as you know and some viewers and listeners may, I went freelance full time for the first time in the past year and it has been such a wonderful experience for me. I love it so much.

Joe Reis (41:33)
That's cool.

Hugo Bowne-Anderson (41:36)
challenging and so like I have an existential crisis once a week. But I get to do so much interesting work. Well, exactly. I didn't so much when I was when I was full time even at know, early stage startups, but I do I think that maybe I'm cut out for being freelance in a way that others

Joe Reis (41:40)
Don't worry, people with jobs have that too, by the way.

Hugo Bowne-Anderson (41:54)
other people I know may not be. I know that you're a huge proponent of everyone considering going and doing their own thing. So I'm wondering if you think there are people who are cut out for it and people who aren't. And if so, if AI can help shift a lot of people who aren't to being more capable in this type of thing.

Joe Reis (42:11)
So I'm just answering a question here.

Hugo Bowne-Anderson (42:12)
Yeah, yeah.

Joe Reis (42:13)
It's interesting. I think that people have the option of going off on their own and increasingly I think it's going to be your only option is to go off on your own. so if you take it, if you figure that everyone has the same capabilities with AI right now, right? It comes under your ability to understand how to use AI to maybe ask you better questions. When I saw that PWC had got a massive subscription to chat GBT for their teams, for the consultants, I was like,

Okay, that just showed me there's gonna be no differentiation at all other than hiring a big name firm, right? You all have the same tools now. So the only thing that's gonna matter at the end of the day is distribution and branding for yourself. That's it.

Hugo Bowne-Anderson (42:41)
Mm.

And data as well, think. I do, I mean, the conversations exactly. And if you know how to integrate it into your modeling workflows as well, right? I do think, you your models and so much of the conversation is.

Joe Reis (42:52)
If you have data that's different than others, sure, yeah.

Hugo Bowne-Anderson (43:04)
model focus once again. always rant about this, but it'll be very interesting for this conversation, I think, and I'd love your perspective on this. Even when ML was the hot thing and Kaggle and all of that, it was all about what are you using XGBoost? Are you using random forests or ensembling methods as opposed to what are you doing with your data?

Andrew Ng, who you've created a course with that everyone should definitely check out. He was a big proponent and lots of us were, but I'm mentioning him because he's such a wonderful superstar and so thoughtful. But he came out like, we've got to think more about data centric modeling. And this was pre the big LLM chat GBT.

Joe Reis (43:26)
Yes.

2021 is when you came up

with that, yeah.

Hugo Bowne-Anderson (43:41)
Exactly right. And the idea then was have a competition, for example, where you hold the model constant and you work on the data, which is so, so beautiful. And then of course, with the rise of LLMs, we've become model focused again. But I think if practitioners become more data focused and it's nowhere near as sexy, but as you're, know, a huge force in data engineering and robust data engineering, I'd love your thoughts on how, like how could PwC start thinking about

the data aspects of things and integrating them into their pipelines and models.

Joe Reis (44:11)
That's a big one. That's a big one. think it's the only thing that's going to give you a moat at end of the day, right? Like what's the institutional data that you have that you can leverage? And this comes back to, know, somebody's asking, do I need to learn anything these days or in comments here? I think more than ever is a time to learn things. It gives you such a huge competitive advantage. mean, otherwise you just, you can not learn things, I suppose, but most people don't anyway.

Hugo Bowne-Anderson (44:15)
Yeah.

It's essential.

to my silly

analogy about it being a chaotic environment or being a tornado, you literally that's all you can do. You can learn experiment, iterate quickly, you can't even necessarily plan for the long term. And one example, I saw one way to think about who you hire for certain things at early stage startups, we say you want to hire explorers, right, who will go like bush bash, jump in the river, camp for three hours, build a fire.

kill a kangaroo, put it on the whatever it is, right? And figure out the lay of the land. Then as you start to scale, once you've got a few things figured out, you want villagers or builders to come and be a bit sedentary, start building things and be chill. And they've actually got specific tasks that each of them are good at and they do these things, right? I think that's no longer the case anymore. Like a lot of like series B, series C startups now,

Yeah, and Alan Nichol, he's the one who he's in the chat. I spoke about this analogy with for the first time some years ago. But startups now, even series B, series C, where you'd think you'd need a lot of people to build and be sedentary, you actually need far more exploration, explorers and

and people who are learning all the time. And there's an argument that all of us now should be spending 40 % of our time doing R &D of what's possible right now and then building MVPs, things at work, iterating on them and so forth. Yeah.

Joe Reis (45:47)
And that's where vibe coding is great. That's where vibe coding

is awesome. But you have to have a sense of creativity too. mean, that's what I think is missing in this industry entirely, which is like probably one of the reasons for having the conversation is just there's.

I think finally, if you go back to John Maynard Keynes' argument that people would be working a 15-hour work week because he envisioned a world where computers and machines are going to do everything for people. That was back in the 30s, I believe, when he said that. It's around about 100 years. investing in yourself, think, is still the greatest thing you could be doing. mean, looking through the comments here, somebody's asking if data engineering is going to be replaced. I don't think so.

Hugo Bowne-Anderson (46:10)
Mm.

Joe Reis (46:24)
I'm writing a book on data modeling right now. Why would I be writing a book, first of all, if learning goes away on data modeling, which is like the most unsexy topic in the world, right? Because I think this data modeling is one of the things where you mentioned data as a moat, perhaps, at least you're kind of indicating that. But, you know, by most accounts, data is an absolute horror show. And most companies, it's just true. Data quality is a big, big problem.

Hugo Bowne-Anderson (46:32)
Mm.

Yeah, yeah.

Mm-hmm.

Joe Reis (46:47)
And it's becoming a very big problem that companies are investing money into precisely because of AI. Data quality in governance and management used to be a very unsexy thing until AI. Now all of sudden, everybody knows getting a lot of attention to this. At the same time, so that's modeling, I'm writing a book because I think I want to write books where, again, we help fill the knowledge gap. Of course, it's going to be harder. People aren't going to want to put in the time.

If you have the capability of learning and if you're interested and you're curious, like you can imagine you're going to be that much further ahead than everybody. If you're the kind of person who's an explorer, who's curious, who's creative, is going out there building stuff, and you're using AI versus a person who just doesn't think at all and just relies on AI, who do you think is going to To me, it's not even a question. Like, I don't even know why it would be up for debate. But I see this once in a while, people are like, well, we can just stop learning.

Hugo Bowne-Anderson (47:30)
Mm.

Joe Reis (47:35)
Stop thinking AI just does everything. I'm like, you can do that. Go for it. I encourage you to do that. That just means one less person I got to compete with in this world. And one less person my kids have to. So go for it.

Hugo Bowne-Anderson (47:42)
Yeah,

and once again, it's about what we're encouraging people to do. So I do think generationally, mean, like your kids and their generation.

Everyone should be incentivized to be curious, not to get 90%, to get a high GPA. Everyone should, you know, we are beaten down as we grow up to, you know, meet these goals, which don't incentivize learning and curiosity. They may incentivize rote learning, but they incentivize performance, not curiosity. So that I think there will need, hopefully be a huge shift there as well. I am interested in thinking about data quality. It becomes far more challenging when we're working with AI systems, but...

Do you think AI can help us with data quality as well?

Joe Reis (48:24)
I think if it knows what to look for, right? So it's interesting. just have my every afternoon I bike for about 30 minutes to 60 minutes of my spin bike that I just read on. was reading this paper on data smells. So you've heard of code smells, I'm sure. Yeah, so this is the same thing with data. But the paper looked at, I think it was published in 2022, but it looked at the impacts of bad data on AI models. Right. And it kind of goes through the categorization and it's

Hugo Bowne-Anderson (48:26)
Mm.

Hmm.

Mmm. Yeah, yeah.

Joe Reis (48:51)
It's true. think that data more than ever is going to be one those things where we have to understand what it looks like. And I think AI could certainly help. I've been working on some stuff with AI and data modeling. And I think there's some potential there for simple stuff, for complex stuff. Forget about it right now. The models are not in a state where they can handle any sort of enterprise complexity. just not. Anyone who tells you differently is like, just bullshit. Or they're to sell you a subscription, something.

But yeah, I think that it certainly can. Where I'm also seeing some good moves is in data and systems migrations, right? Where if you had to migrate thousands or tens of thousands of legacy jobs to a new programming language or new framework, it's like, why not try and leverage all of them and AI as much as possible? You should. You should. But you got to know what you're looking for. That's just it. It all comes back to, do you know what you're doing with these tools?

Hugo Bowne-Anderson (49:22)
Mm.

Totally.

Joe Reis (49:41)
It's kind of like if Elon Musk is out wielding a chainsaw a few weeks ago on stage, right? I grew up using a chainsaw. I know how to use them. If I gave you a chainsaw and you just say to go vibe chainsawing, like you'd probably lose your leg in about 10 seconds, right? And I think it's the same thing. You have a very powerful tool, but you got to know how to use it. Just like when using it for writing. Like if you didn't know.

Hugo Bowne-Anderson (49:45)
I saw that.

Yeah.

Joe Reis (50:00)
what good writing looks like and using LLM to do it. Well, you can just go on LinkedIn and look at most posts these days and see what that looks like. but there's just, yeah, at the end of the day, it's just, my whole thing with vibe coding and the whole pushback was, there was this notion that you can substitute critical thinking and knowledge and skills with, you know, just blindly accepting whatever AI gives you. And I think that's a catastrophic mistake. But like I said, you know, the blast radius has got to be pretty small.

If you want to go for that, go for it. Just understand the game you're playing. I ain't going to stop you. I'll just might call you stupid, that's all, but you probably won't care about that either.

Hugo Bowne-Anderson (50:31)
Well, as Mr. Me 77 has written in the chat, imagine doing vibe surgery. I definitely wouldn't want a surgeon. Vibe surgery meets vibe chainsawing together. How about that?

Joe Reis (50:36)
Fuck it.

That, exactly.

That's actually a good image, right? Well, vibe surgery is an interesting one, right? I mean, I can imagine, you know, because anyone on the street just do surgery. Go for it, right?

Hugo Bowne-Anderson (50:44)
Yeah, that's what it would be like.

But on the other side, I would, I don't know really what I'm talking about, but my understanding is that surgery is kind of vibe surgery in a certain sense, right? That these surgeons have internalized their processes and muscle memory so much that they are in a flow state where they're not actively rationally going, okay, now I'm gonna do this now all the time. There's a deep learned intuition behind it. So it's a different type of vibes, but it is a deeply intuitive process as well with guard rails and checks and balances.

Joe Reis (51:15)
Oh yeah, have a relative who's a brain surgeon.

have a relative who's a brain surgeon. You how hard that is? But this person's been doing it for a long time and it's just second nature like you say. It's like, okay, I'm gonna go pick your brain apart right now. Not a big deal. But yeah, it just comes through practice. Like anything, right? Like you vibe, you know, ML modeling or whatever, right? Like what you would do. You have a sixth sense about how this stuff works.

Hugo Bowne-Anderson (51:20)
Yeah.

So.

Yeah.

Yep.

Joe Reis (51:38)
And also if you were to use AI for this stuff, you would know almost immediately because this makes sense or not, right? Yeah, because you're a master at this. So, yeah. Yeah.

Hugo Bowne-Anderson (51:42)
Absolutely, yeah. Yeah, you can see what the stop is immediately.

I do like the chainsaw reminded me of the car analogy earlier and...

I do I am interested in how we can build AI tools and infrastructure to put guardrails around certain things so people don't necessarily need to know as much. So one example is when cars first came out, need to know you needed to only specialists only mechanics could actually drive at the start, right? Because you needed to know how to fix everything, how to improve everything, failure modes, internal combustion stuff. Then we have the emergence of stick vehicles, right? And you need to know a certain amount there. Then we have automatic

cars where you need to know how to change your oil, change, put some gas in, whatever. But the skill set to do this safely and successfully has actually changed. So do you see a similar thing happening with with software?

Joe Reis (52:34)
It's interesting.

do. And I actually started rereading Zen and the Art of Motorcycle Maintenance, which is written in 74, a good book. I read it when I was a teenager and I actually had the original copy from when I was a kid. It's a rereading that. And I think there's a lot of arguments that are similar to what you described. Right. You know, the classic thing is, I trust the mechanic to fix my bike or do I know how to fix it? I think it just depends on your circumstances. But we do see this happening in programming. Right. I mean, even before

Hugo Bowne-Anderson (52:39)
Awesome.

Mm.

Joe Reis (53:02)
you know, LMS and code gen, was, you know, IntelliSense. So, you know, which was depending on the programming language, probably pretty dang good. but you know, I know some people who turned it off. They're just like, I don't need this. She gets my way. you know, as, I'll never forget, I was hanging out with a Hannes Mollhaisen, guy, the guy who created DuckDB for having lunch. he was just like passively just coding like an extension for DuckDB. think it was for Avro. Unsublime text.

Hugo Bowne-Anderson (53:08)
Mm.

Joe Reis (53:25)
which is like very punk. I know he doesn't use AI, I'm pretty sure he doesn't. But that was an example of just like somebody who's like a true expert at the system that he's building and the language, in this case C++. And he's like, this is what Masferi looks like, right? You don't need anything. But I feel like that's going to be a relic of the past, for sure.

Hugo Bowne-Anderson (53:26)
Mm, totally.

Mm.

Joe Reis (53:49)
That's like watching just a maestro or a harness in this case, just doing their thing. And it was, I just, think that's, that's not going to be as common anymore. I mean, it's, it's a dying breed. It's a dying breed. you, if you look at, I just don't know where you're to learn this stuff, you know, building a system from scratch, right? By hand, every bit of it. It's just, that's, that's cool. But yeah.

Hugo Bowne-Anderson (54:09)
What do you think about the space of, at the moment, um...

all of these technologies and coding assistance and agent mode on cursor or whatever. It's all single player. Right. And you mentioned you were chatting with Claude about this beforehand. I was doing the same. How much fun would it be if we could chat with Claude together, have a multiplayer conversation, right? So is there like, in terms of us all working together with an agent that we're delegating things to checking it together, that type of stuff, product managers, UX people, executives, you don't want to get too crowded, no middle management because where AI is removing all middle management, which essentially

Joe Reis (54:25)
That'd be dope. I'd love that.

Hugo Bowne-Anderson (54:41)
I like to refer to as bounces for the executive But yeah, is there a future where we can have multiplayer mode with these systems and an Excel there? Yeah

Joe Reis (54:48)
I hope so. I hope so. I mean, that'd be pretty cool. I'd love to collaborate

on this kind of stuff. you know, I have a lot of friends like yourself. Yeah, I think it'd be awesome. I don't see any, yeah.

Hugo Bowne-Anderson (54:57)
be so much fun. You, me

and KJam, riffing with an LLM agent to see what's up. That'd be so cool. Exactly. Well, just imagine it embedded in YouTube right now. Whoever's watching, mean, YouTube, whoever's watching, build this feature or I'll build it and then you can acquire me. That's not a bad idea, actually.

Joe Reis (55:01)
That'd be dope. Or you already said everyone in this chat too. I mean, think that'd be fun as hell.

No, I think

that's where it's all going. I think it's just going to be one of the things where eventually all our agents are collaborating with each other. And we've got models.

Hugo Bowne-Anderson (55:26)
Well, one of the biggest fucking issues

is memory with this shit, right? Like even most of the basic LLMs at the moment do such a horrible job with memory and overindex on my most recent comment and like actually getting them to...

Joe Reis (55:30)
Yeah.

Hugo Bowne-Anderson (55:38)
We should have insight in products into memory and be able to shift that ourselves. think, look, I just want to say there are so many people in the chat saying how much your book has impacted their man. So big props. I am interested. This is a slightly tangential converse question, but I think it's related. I'm interested in why you write books and what my real, what the question under the question is. I love books, right? But as someone who wants, who's a wonderful educator, who wants to educate more, why is the book the choice as opposed to

Joe Reis (55:41)
Hmm. Yeah.

cool. Thank you.

Hugo Bowne-Anderson (56:08)
a GitHub repository with executable code and all of these things. Why work with publishers? Why want something in hard copy?

Joe Reis (56:14)
Oh man, that's a question I ask myself a lot. Like when Fundamentals of Data Engineering was published, all of a I started seeing my name with the word expert next to it. Like expert data engineering. I'm like, well, I'm literally the same person that was before the book was published. Right. But now that the book's published, you know, I think it's just a different, you know, what I've come to realize a few years in is there's a certain gravitas that comes with publishing a book. Right. I think publishing a popular book, right. You can go publish a book.

Hugo Bowne-Anderson (56:37)
Mmm

Joe Reis (56:40)
sight unseen just using chat GBT and Kindle direct publishing if you want to get done like 15 minutes. But I think publishing a book that you know is widely read that's hard to do when I think that's something I'm proud of me and Matt for doing that. But think to the broader point of why I write books and why are they useful these days. I don't know why people read books anymore. Like there's there's every reason not to read a book. There's every reason there's but people do.

Right? a book is out, the only thing I can think of is it signals a sense of encapsulation of ideas that are impossible to replicate in the ways you described on GitHub, on a YouTube course, on something else. I was actually just chatting with Mark Freeman about this because him and Chad Sanderson are wrapping up their book on data contracts. And it was interesting where I remember when Mark and Chad are starting their book, or before actually, when they were just thinking about it.

I was like, you sure you want to do this when you're starting a company? You're going to be, you know, but I think they realized it was hard work, but it also meant like they had to think about a topic on a level that you just aren't going to think about it otherwise. And I think that's the one thing where writing, writing forces you to think just at a fundamentally different level than anything else. You've seen this.

Hugo Bowne-Anderson (57:42)
love that raisin.

Yeah, and some well, I actually want to get back to that. And I don't have a very, no, it was a profound experience for me recently with what you just said, but the idea of writing a full book as opposed to a long series of blog posts series, a long series of long blog posts with accompanying code and that type of stuff. It forces you to think what this entire framework of thinking is. And I love that. But what I so I write a lot about as of late, I've been writing more and more with

with AI and basically around conversations I have putting transcripts in then co-writing and that type of stuff. But what I did recently was I sat down for the first time in a long time and wrote a 3000 word draft on the software development life cycle for LLMs which LLM powered apps and anyone wants to see that it'll be on O'Reilly radar next week so that's a shameless bit of self promotion. But I said, I.

Joe Reis (58:20)
Yeah.

Yeah, you worked hard at it. Don't feel bad about

promoting it.

Hugo Bowne-Anderson (58:42)
Well, I did and I,

but dude, I sat down and like wrote a 3000 word draft without any AI assistance. And it was, it was such a beautiful and challenging experience. And it's something I used to do a lot of that was second nature to me, but in six months of not doing that so much, I had lost touch of that. And I, know, when you like nearly feel your brain move in different ways and that, that experience really did.

Joe Reis (58:49)
It's awesome.

Hugo Bowne-Anderson (59:10)
kind of push me in a way that I think it's really important, just like we're chatting about. If you sit down and try to read a map these days and get somewhere from A to B, that's something I appreciate about certain parts of the US is you don't have cell reception, you got to go to like the local gas station and get the local map, right? But yeah, what it forces you to think and tools that force us to think while they automate the boring stuff. mean, yo, like the amount of

Joe Reis (59:25)
Get the map.

Hugo Bowne-Anderson (59:35)
One of the biggest problems of schooling, think, is the amount of long division you have to do at elementary school and then at college the amount of matrix, factorization and like Jordan canonical form bullshit. I mean, do it once or twice, but we've got tools that do that. We've got calculators now and it does, it's a paradigm of the rote learning where you don't learn anything. You're just implementing a fucking recipe over and over again, right?

Joe Reis (59:43)
Ugh, yeah.

Yeah.

Yeah, but the thing is something we said for that and writing is just writing in general, whether you're writing an article or a book is I just don't think there's any substitute for it. I don't think there will be a substitute because you're using language to express ideas and using ideas to express language. And that's just it literally is thinking like, you know, this book I'm working on right now, it should have been out a long time ago, but I've been thinking really hard about it and.

you know, want it to be really, really good. I could just have it in LLM, just generate everything for me, but I'm not going to cheat myself out of the experience of thinking about this topic and writing about it. I'm not going to cheat the audience and the readers out of that. Right. And so the books, you know, I started a publishing company too, and the books I'm publishing or the books I feel that are going to help change the thinking of the industry and push us forward. that doesn't go away, right? New ideas are the things that help promote us and, and push us. If we lose that, like, what do we have?

Right. So that's the kind of books I want to write. I'm not interested in pushing publishing books that are, you know, how to do X, Y, or Z in Python, for example, like that to me actually is a good use for LLM's at this point. Like you have a question, it'll probably do a good enough job helping you. Right. And so I don't think that those need to be books these days. I haven't. In fact, I I I haven't felt like those need to be books for ages. Those should be courses on YouTube or now just prompts. But the idea is like, you know,

Like my friend, Jim Actigani came up with the idea of data mesh, right? Revolutionary idea. Took the world by storm. Now, LLM wouldn't come up with that. Right? It just wouldn't. And so these are the types of books I'm interested in really helping people bring to life or the books that are really going to push our industry forward. Because more than ever, we need that type of thinking. Otherwise, we become stagnant. So.

Hugo Bowne-Anderson (1:01:34)
Yeah.

And to that point, do think, you know, I do a lot of education as well for data science and ML and AI stuff. until LLMs, sadly, when I teach, a lot of it, I have been needing to teach people APIs, right?

And I don't want to teach APIs. I want to teach systems thinking and building systems, right, that incorporate data and the entropy of the real world into them. And now as an educator, I'm liberated, Joe, to teach the things that really keep me up at night and keep me excited and get me out of bed in the morning. They keep me up at night and get me out of bed in the morning. So God knows when I sleep, which I don't really at the moment. We are going to have to wrap up in a second, sadly, but I'm wondering...

Joe Reis (1:02:06)
Mm-hmm. Yeah.

Eh.

Hugo Bowne-Anderson (1:02:15)
As an active prediction, what? Let's fast forward one to five years, whatever it is, whatever you're comfortable with and tell me the future that you don't want to see and the future you do want to see with respect to how how people build software and who gets to build software.

Joe Reis (1:02:27)
I was about to say

five years from now, I don't even want to my prediction what I think the world looks like. But as far as software goes, assuming we haven't completely blown each other away in five years, I think we're just going to have a, I don't know, I don't know what happens. I think things are moving too fast. But what I can see is if I look at what doesn't go away, right? If I look at least what doesn't go away today,

Hugo Bowne-Anderson (1:02:31)
Haha

Joe Reis (1:02:52)
The need for robust systems, the need for systems that work, the need for systems that are low in technical debt as much as they can be. The systems where data is of high quality and systems where AI is helping the business. Like, I don't see that world going away. And so as long as that world exists, I think that we're going to do everything we can to build things that complement that future. Right? that's the only... So when he asked me for predictions, the only thing I can look at is like, well, what's probably not going to change as much.

Right. If I try to extrapolate and say, well, 2025 will be the year of, you know, this type of a model or whatever. It's like nobody can nobody can say that. Right. So but but again, commercial needs, assuming, you know, they're still the same. know, you know, I think that that's that's where we'll still be going as an industry, because that's where we've been. That's what we've been doing as an industry for many decades. So, you know, but that's Frederick.

Hugo Bowne-Anderson (1:03:24)
Hmm. Nope.

Joe Reis (1:03:43)
Brooks said, you know, in one of his articles in the, I think it was the seventies, that there's no silver bullet, right? So we're always trying to look for the silver bullet in our industry that's just going to magically solve all of our problems. And I don't think we found it yet. I don't think that AI is that bullet either. the end of the day, it's...

Hugo Bowne-Anderson (1:03:58)
Yeah, and it's not fair

on AI or us to even think it's a silver bullet, right?

Joe Reis (1:04:03)
No, no, but you know, it's being sold as that, right? So the other thing I don't think is gonna change is like, know, vendors are gonna vendor, they're gonna try and push stuff that is a silver bullet, because that's what their incentivize to do. That's like, I'm not gonna blame them. That's the game. So.

Hugo Bowne-Anderson (1:04:05)
Yeah.

a

really important point because at least when I kind of came up in, you know, a decade ago, let's say, you know, starting in 2011 or so in like data science and ML and AI stuff, the tools that we were all using were kind of the extended PIDATA and SciPy stack, right? So we had the genesis of IPython notebooks had just been folded into an IPython terminal had been folded into

Project Jupiter, had map plot lib, had Wes McKinney not only created pandas, but PD read CSV, we could all import CSVs into into our Python environments in notebooks, scikit learn all of these things. So it was a really like vibrant community of open source tools that we're all using. And it still was kind of fragmented and fractured. The R community, of course, had all of their stuff going on, which was less less fraction and fragmented. But now, with like the venture backed fucking tragedy of the commons that we're seeing in the tooling landscape where every marketing page

Joe Reis (1:04:39)
Yeah.

Yeah.

Hugo Bowne-Anderson (1:05:07)
looks the same and to even figure out what a tool is. I need to get on a call with a BDR who, you know, then messages me afterwards. is... How can we even step back? I mean somebody made a point that we haven't really evolved to use all these tools. I haven't... I don't think our civilization has evolved to deal with all of, you know, these sales calls I have to get on as well.

Joe Reis (1:05:28)
The counter

argument to what I just said where we need to increase our skills to leverage the tools we have, I actually think that one scenario of the future is it's aligned with what Satya Nadella was talking about, the application landscape gets inverted and the tool landscape gets inverted where you just invent, use AI to create the solution for you and it's very bespoke, right? And so that is another possibility. I don't discount things by that.

Hugo Bowne-Anderson (1:05:45)
Mm.

Joe Reis (1:05:50)
You'll never see me say, well, that's absolutely never going to happen. It could happen, actually. And I think that there's a rational use. Like you said with the JSON viewer, you build a tool, and maybe that's it. And like I said, it's the 3D printed comment earlier on in our discussion earlier. this could be the case where you just invent tools as you need them. then the AI agents, the companies become a toll road for different agents and models. That's absolutely a situation that could happen.

Hugo Bowne-Anderson (1:05:54)
Yeah.

Yeah.

Joe Reis (1:06:16)
And I would kind of like that actually, because I know the kind of tools that I want and they're sometimes not the tools like I know I subscribe to a tool like notion, right? Notion is a great tool for a lot of things. Do I need everything that it provides? No, you don't like it. OK, I think it's.

Hugo Bowne-Anderson (1:06:27)
disagree.

Yeah, no, actually, I

do love but just the amount I had to learn about like setting up a database to build a basic Kanban board where I can switch rows like bent my brain. Yeah, it's incredibly powerful. But yeah, what I

Joe Reis (1:06:35)
It's a good yeah. It's it's a it's insane. Yeah, I do agree with that like but. Well, you can do

anything you want it, but as you point out, it's a learning curve, right? But if I just need like a certain thing and I just want to build that cool, make it easy for me, right? So they have AI notion which works 20 % of the time. But the whole point is like if I could just have an agent, I can just tell I just need this thing. Build it for me. I don't need to pay a subscription anybody. I just want this functionality.

Hugo Bowne-Anderson (1:07:02)
Yeah.

Joe Reis (1:07:02)
for this

use, that's awesome, right? And so, and it made me think about this article I was reading on the flight home last week where it was talking about just the gap in robotic expertise between China and the rest of the world. Like they're building general purpose robots. So if you think of AGI, but the robotic version of it, this is what that is. It's like robots that can just do anything. And then they have factories that robots just make other robots. And there's no humans working in these factories anymore. They're just.

Hugo Bowne-Anderson (1:07:18)
wild.

Mm.

Joe Reis (1:07:29)
all robot powered with the lights off. Like this is a world we're heading into. But if I think about that, where I have a robot that can just make anything I want and do anything I want, okay, that's pretty powerful. And that completely inverts the entire manufacturing paradigm of today. yeah, so this is where I'd like, this is the other alternative where I'm like, okay, knowledge is great. Plus this stuff, my God, it's a cool universe. It's frightening, it's awesome. It's everything in between.

Hugo Bowne-Anderson (1:07:42)
Yeah, totally wild. And-

Joe Reis (1:07:54)
So.

Hugo Bowne-Anderson (1:07:54)
without a doubt. I do. One thing I'm hearing there is that maybe we just have less centralization as well, which I'm a big fan of. Yeah.

Joe Reis (1:08:00)
I'm a big fan of that too.

Absolutely. mean, yeah, you mentioned K-Jam. I mean, I think we're all kind of somewhat anarchist in some way. So yeah, I mean, I do agree with that. I don't like centralization.

Hugo Bowne-Anderson (1:08:06)
Yeah.

Yeah. I also am interested. mean, you know, there are 8.2 billion people on earth or something like that. And we've got 27, 28 million developers supposedly. And that's, you know, slowly increasing, but kind of plateauing. I'm just interested if this type of tooling can just, how it increases the surface area of who not necessarily can create robust software, but who can contribute to it.

Joe Reis (1:08:34)
think again, if there's guardrails on things. Somebody described it. I can't remember the article. I need to find it. was a really good article, but it just talked about the amount of technical debt in the world. And technical debt, meaning the software that's yet to be written. And that's a lot. We are constrained by the 70-something million developers. if we increase the amount of code and things that software can solve, mean, it goes back to David Deuchess.

Hugo Bowne-Anderson (1:08:37)
Hmm.

Mm.

Yeah.

Joe Reis (1:08:58)
you know, writings, you know, and so forth. you know, this is where you start getting into the realm of like science fiction, you know, but I think this is potentially the world we're getting into. And this comes back to the notion though of craftsmanship, right? Because this is the essence of what I was talking about earlier with you and in my talk, Koikso in Austria, but it's like, if we can build things that are robust and not disposable,

right? And things that work and serve a real purpose, along with AI, this is the benefit of all of humanity. If we don't, if we live in this disposable culture where we just have endless systems producing endless amounts of garbage, it doesn't serve anybody, right? And then you have, you know, it makes the Pacific garbage patch look downright quaint. But this is a world, you know, that we're, I think we're at a juncture to make that decision right now. Well, I don't know that we collectively won't make a decision. It'll just

Hugo Bowne-Anderson (1:09:28)
Mmm.

Joe Reis (1:09:48)
be the consequence of a bunch of decisions. But right now, I'm not hopeful. But there will be lot of software produced. don't think that's up for debate. There's going to be a lot of code, a lot of software produced.

Hugo Bowne-Anderson (1:09:59)
It's gonna be a Cambrian explosion, Joe. I'm sorry. I love that you mentioned we're verging on science fiction as well, because this kind of takes us back to the start of the conversation where you talked about coming up in the 80s. I came up in the 90s. So I was more like Teenage Mutant Ninja Turtles and Wu-Tang Clan and Nirvana. I like...

Joe Reis (1:10:14)
Yeah.

Hugo Bowne-Anderson (1:10:16)
The sci-fi of the 80s, I'm thinking like, Neuromancer Man. I mean, like, this type of stuff, Gibson, then a bit later, Snow Crash, Neil Stephenson. Like, you look at this science fiction and how it relates to the world we live in today, and it's no joke, right?

Joe Reis (1:10:30)
No, it's no joke at all. mean, I have the original books on my bookshelf. These are the books I grew up on. And it's interesting now seeing a lot of stuff becoming reality and in some ways becoming those books are almost quaint in some ways. Right. So

Hugo Bowne-Anderson (1:10:40)
This is super weird.

So in Snow Crash, like the virtual reality place, as you know, called, and for people who don't, it's called the Metaverse. And it's literally in Snow Crash, a place which was meant to be a free place, virtual reality, anybody could do anything that's taken over by corporate interests. That's literally what happens in the novel. And it's called the Metaverse. And I couldn't figure out when Zach rebranded as the Metaverse, whether he like knew that or...

Joe Reis (1:11:05)
he knew it. Yeah, yeah, I think it's a, yeah, it's, yeah, but that's, it's a really good, what hero protagonist, think was the, the, the hero of the book there. But yeah, it's, but that was an interesting one, right? Where you had, he was living in a storage locker, I think in the middle of the outskirts of LA. Yeah, and had to go fight some criminals.

Hugo Bowne-Anderson (1:11:22)
Exactly.

Yeah,

and it actually the the novel opens with like a high speed chase by a delivery a pizza delivery driver being chased by the people who own the company that is essentially the Uber Eats of whatever it is, and they're going to shoot him if he doesn't get it there on time or something like that.

Joe Reis (1:11:34)
Mm-hmm. Yeah.

Yep, yep. Yeah, if you don't

hit you on the 30 minute delivery, you're dead.

Hugo Bowne-Anderson (1:11:43)
Exactly. Look, so that's not not the future we're imagining. Hopefully it isn't the future we get. Exactly. Exactly. But Joe, thank you for your time and wisdom and and I always enjoy conversations with you. I'd just like to thank everyone for joining. We've had over 100 people here collectively throughout throughout the chat. And we'll put out the podcast next week. So keep your eyes open for it. You probably know but

Joe Reis (1:11:47)
Somebody else imagined it for us, we don't need to do it.

Of course. Thank you.

Of course.

Hugo Bowne-Anderson (1:12:05)
follow Joe Reese on on linkedin is there anywhere else people you've got a sub stack newsletter people can

Joe Reis (1:12:09)
I got a couple of

sub stacks. I got a podcast. Not an OnlyFans yet. kidding. who knows? No, no, I don't. Not that you know of. I'm kidding.

Hugo Bowne-Anderson (1:12:16)
You mean you do have an early fans? Weirdly, like I did, well, know,

data after dark with Joe Reese. When you mentioned like, you can write Python code drunk, I've always wanted to do it like, you know, like drunk history, but with like a conversation like this where we just, or like a hot one, is it hot ones? The ones where they eat hot wings or whatever. And it'd be fun to do a podcast like that or a drunk history one sometime, I feel.

Joe Reis (1:12:25)
Hey.

Yeah, they are hot wings, yeah.

That'd be fun. Only bits. Nice one, Travis, like that. Well, it's, yeah, but yeah, I'd be down to do that. Maybe in person we can get some mics and stuff like that. yeah, drunk podcasting is pretty fun. So, yeah.

Hugo Bowne-Anderson (1:12:44)
Yeah. no. Only bites.

That'd be fun. So I'll be in the

US for most of summer and I'll be in Europe before that as well. and we'll I also yeah, I also want to get you and we've mentioned K jam. So this is Catherine Jamal, close friend of both myself and Joe. Check her out does a lot of incredible work on on privacy, data privacy, security, all of these, these types of things. Joe, thanks once again, so much fun. Thank you all for joining and check out the podcast when it comes out next week.

Joe Reis (1:13:02)
hit me up.

Yeah.

Of course

Okay, all right. See y'all.