In this episode I talk with Eleni Arvanitis Konior, returning to the show, about the downsides of AI: people using it to reinvent the wheel, AI slop in code and on social media, the hype, and rising costs. We also discuss why proven practices like TDD still matter, and how to guide AI toward writing better tests. Links: - Eleni's personal website - Martin Fowler's testing resources - Nonsense Monthly
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
SPEAKER_01Hey, it's Jason, host of the Code with Jason podcast. You're a developer. You like to listen to podcasts. You're listening to one right now. Maybe you like to read blogs and subscribe to email newsletters and stuff like that. Keep in touch. Email newsletters are a really nice way to keep on top of what's going on in the programming world. Except they're actually not. I don't know about you, but the last thing that I want to do after a long day of staring at the screen is sit there and stare at the screen some more.
SPEAKER_01That's why I started a different kind of newsletter. It's a snail mail programming newsletter. That's right. I send an actual envelope in the mail containing a paper newsletter that you can hold in your hands. You can read it on your living room couch, at your kitchen table, in your bed, or in someone else's bed. And when they say, What are you doing in my bed? You can say, I'm reading Jason's newsletter. What does it look like?
SPEAKER_01You might wonder what you might find in this snail mail programming newsletter. You can read about all kinds of programming topics like object-oriented programming, testing, DevOps, AI. Most of it's pretty technology agnostic. You can also read about other non-programming topics like philosophy, evolutionary theory, business, marketing, economics, psychology, music, cooking, history, geology, language, culture, robotics, and farming.
SPEAKER_01The name of the newsletter is Nonsense Monthly. Here's what some of my readers are saying about it. Helmut Kobler from Los Angeles says thanks much for sending the newsletter. I got it about a week ago and read it on my sofa. It was a totally different experience than reading it on my computer or iPad. It felt more relaxed, more meaningful, something special and out of the ordinary.
SPEAKER_01I'm sure that's what you were going for, so just wanted to let you know that you succeeded. Looking forward to more. Drew Bragg from Philadelphia says, Nonsense Monthly is the only newsletter I deliberately set aside time to read. I read a lot of great newsletters, but there's just something about receiving a piece of mail, physically opening it, and sitting down to read it on paper that is just so awesome.
SPEAKER_01Feels like a lost luxury. Chris Sonnier from Dickinson, Texas says, just finished reading my first nonsense monthly snail mail newsletter and truly enjoyed it. Something about holding a physical piece of paper that just feels good. Thank you for this. Can't wait for the next one. Dear listener, if you would like to get letters in the mail from yours truly every month, you can go sign up at nonsense monthly dot com.
SPEAKER_01That's nonsensemonthly.com. I'll say it one more time nonsense monthly dot com. And now without further ado, here is today's episode.
SPEAKER_01Eleni, welcome back.
SPEAKER_00Thank you so much, Jason, for having me back. I'm excited.
SPEAKER_01I'm so glad you're here. I'm excited too. Um, for anybody who didn't catch the first episode, do you want to tell us a little bit about yourself?
SPEAKER_00Yeah, for sure. Um, so yeah, I'm Eleni Konyor. I'm a senior staff software engineer at Cisco Meraki. I've worked here for about four years. Um outside of my profession, I am a Greek American, first generation um first person to go to college in my family. And I don't know, I like the outdoors, I like the bass.
SPEAKER_01Um wait, wait, you said you like outdoors and you like what was the other thing?
SPEAKER_00The bass. We talked about the bass guitar. Oh, the bass guitar, of course. Yeah. Yeah, yeah, yeah. Um, and yeah, we talked a little bit about left and right brain. So yeah. Uh I am biased. I would recommend people go hear that recording, but yeah, that's a little bit about it.
SPEAKER_01Yeah, I thought that was a great one. Um, yeah, and dear listener, if you didn't know, I too work at Cisco Meraki. Um I'm a tech lead on a different team. You and I, Lenny, don't cross paths a lot in a work capacity at work, but we do complain in the DMs a lot to each other, which is great. Um, I often feel like, you know, I have a complaint, and I'm like, man, who who would understand this complaint that I need to complain about?
SPEAKER_01Lenny would understand it. She gets it. Uh so I'm I'm glad you're there to. Yeah. Um, anyway, what's on your mind today?
SPEAKER_00Yeah, um, we did talk a little bit about AI last time, and I think you know, AI is great, but there's also a lot of downsides. And, you know, right before we were like gathering here, we would talked a little bit about the reinventing of the wheels that is happening. Um, so maybe we could start complaining about that and then see see what other topics come to mind in terms of complaining.
SPEAKER_01Yeah, and dear listener, I have to share just a little bit of background. Um, kind of in jest, I had the idea. I'm like, what if we start a podcast called I I forget, do you remember what I said for the podcast name idea?
SPEAKER_00It was great. I was gonna try and see it in our DMs.
SPEAKER_01Well, it it was something to do with you and me just like ranting. Uh I'm not gonna start a whole new podcast for that, but we can at least do that here on this podcast. Anyway, yeah, tell me tell me what you're telling me a little bit about pre-show where people will reinvent the wheel with AI.
SPEAKER_00Yeah, I so because AI is so quick at just churning out code and like thousands of lines of code, and you know, it it gives you something demoable very fast. People love creating like bespoke solutions for things that like at least 80% they could just get already out there. So one thing I've noticed is that people who don't necessarily like the Jira interface, which you know, I'm I have never worked at Alassian.
SPEAKER_00I have no skin in the game. I used to be the biggest Jira hater, and they've made so many improvements in the past few years. I don't know who's making those decisions, but please keep doing it. Um, so I don't think the interface is as bad as it used to be by any means, but still people have this mindset that it's frustrating to use. And, you know, software engineers don't like writing tickets. I don't like writing tickets, so I get it. Um, but they I guess managers want to display it in a very specific way that you can probably just get from a board anyway, if you just configure the board differently.
SPEAKER_00So they'll make their own like Jira instance effectively. And I can't even tell you how many managers have done this at Cisco alone. And they'll be like, look at what I built, and it's so pretty, and I'm so happy. Like, okay, so you're just gonna keep maintaining that now? Because that's gonna take away from whatever other responsibilities you have. So that's one thing. And then I think with skill development, like so many people create skills, and then you either put it in like one huge monorepo, or then you have many different scattered repos.
SPEAKER_00And it's like, how does everyone get the information that they can use a skill and blah, blah, blah? And there's like open source packages that create marketplaces that handle automatic updating and all this stuff that you can configure, and it's quite nice. And like I said, open source. It does so much more than anyone could possibly be building on their own, but there's at least five at Cisco where people are like, well, I had a very specific need that I wasn't that wasn't being met somewhere else.
SPEAKER_00And I'm like, okay, so you just wasted all of those tokens to build something that you didn't need. You could have been building something cooler that doesn't exist. Um, so yeah, that's like my biggest bright gripe right now around landscape.
SPEAKER_01Yeah, it's so entertaining. It's it's depressing and entertaining at the same time. Um, you know, I've I've talked about this before. I don't know if I've talked about it on the podcast, and not to get political, but there were kind of three events in the 21st century which made me change my view of humanity a bit. Um one was when Trump got elected in 2016. I'm just like, what the fuck? And it it it caused me to like rewire my brain and like have a new understanding of people and America and the world and stuff like that.
SPEAKER_01Um and that was you know, that was on me. Like my my understanding was distorted, and so this this thing that was a big shock to me caused me to to to rethink everything and gain a better understanding. Um and then the next thing was COVID. There were all sorts of wacky responses to that whole thing on both sides of the political spectrum. Um and now AI the the response to and AI is is less political.
SPEAKER_01There's there's some political stuff mixed in, but mostly it's just it's depressing, and I hate to say it, but like the average person is so much dumber than I previously thought. Um even supposedly smart people. You know, like I used to work at a grocery store in a bad neighborhood, and like I all day, every day I had to deal with not everybody, but I had to deal with a lot of awful people.
SPEAKER_01Just awful people. Like, quick example, one time a guy stuffed a bunch of steaks down his pants, and he tried to walk out and steal these steaks, and we had to fight him. Oh my god. Yeah, me and these this other guy fought him, and he like threw me against uh this like brochure rack, and like I cut my head and I bled. He tried to this is this is one of the best parts. Um the other guy tackled him, and I like jumped on him and put him in a headlock, and we were like wrestling around on the ground.
SPEAKER_01The guy put his thumb in my right eye and tried to fucking go gouge my eye out over stakes. He he didn't manage to gouge out my eye, but he did manage to scoop my contact lens out of my eye. Yeah, so anyway, just that that was like a sample of reality at that point in time where like I grew up in a small town, then I moved to this small city and and had a whole new experience of humanity.
SPEAKER_00Yeah, I'm so sorry you had to deal with that. That's wild.
SPEAKER_01Oh no, it was fun. Um and and now it's like okay, I have been like sorted into a different socioeconomic stratum, and now I don't deal with the kind of people who stuff steaks down their pants. Uh it's it's all these uh college educated, supposedly smart people. But as you have experienced as well as me, a lot of these people are not actually fucking smart. It's crazy.
unknownYeah.
SPEAKER_01And it's like uh it's it's one of those bitter truths that's really hard to accept.
SPEAKER_00Yeah, no, I I hear you on that. It's there's a huge discrepancy in terms of like excuse me, what kind of college education you're getting and how open-minded you are. And like I when I first went to college, I I have lived near in New Jersey, so like I've lived near New York City my whole life. I've been there so many times before I ever lived there. But when you actually are there every single day and you're around people that you didn't grow up with, and it's like, you know, I went to NYU, and so there's a lot of people who come from international places, and so you get immersed into this lovely like hodgepodge of humanity, right?
SPEAKER_00And get exposed to different things. And if you're someone who hasn't made an effort, a conscious effort to do that or to learn about other people or their own experiences, or even try to have empathy for their experiences, then no matter how educated you are, you're still gonna be missing all of that stuff that actually teaches you more about life. So yeah, you can get A's and and graduate um and and have gone to a college that you still didn't really get exposure to different people or different trains of thought.
SPEAKER_00And so then how can you question your own reality? You don't, right? So it is fascinating.
SPEAKER_01Yeah, it is. Um it's fascinating and it's frustrating and and all those things. Um, but anyway, back to the AI thing. Um, just the whole response to this wave of AI has been so irrational. We have this incredible tool that is like, you know, right up there with the light bulb and all that stuff in terms of the impact of inventions and what are people doing with it. Some people are doing amazing things, and a lot of people are just squandering this incredible power that we now have.
SPEAKER_00Yeah, I saw like an 18-year-old is using AI to help cure like some specific kind of cancer. And like that to me, it's like, yes, please, let's keep hearing more about that. Um, but yeah, no, I I feel like mostly it's let me just make a copy of some website that already exists.
SPEAKER_01It's like yeah, I don't know if it's like uh Parkinson's law or whatever, like work will expand to fill up the available time or whatever. But you know that thing how like early on in the Industrial Revolution there is a prediction that people would eventually only work like 10 or 15 hours a week and spend the rest of their time on leisure, because it's like, oh, you can make your whole income in just this fraction of the time you used to be able to, and people will spend the balance of their time in all these amazing ways, but people ended up just working the same amount, and now we have more money, and we um we've we filled up that available time.
SPEAKER_01Um and I kind of see the same thing with AI. It's like we have uh uh I don't know, let's let's just throw out a number and say we have 5x the productivity capacity that we did before. Um but I don't feel like people are creating 5x the value we're just we're we're creating the same amount of value and then using that extra capacity to like waste a bunch of time.
SPEAKER_00Yeah, it's so sad.
SPEAKER_01Yeah, that's probably overly cynical, but I I think that's probably like directionally correct.
SPEAKER_00Yeah, no, I'm with you. Um, and I think it's healthy to have cynicism around this because there's so much hype with AI, and there's so much marketing from all of these different tech companies, and you know, they just sneeze out AI and then the stock price goes up, right? So it it is important to talk about the limitations and the reality of how we're using it versus the like capabilities and the pros. But yeah, it's there's so many things that go into it.
SPEAKER_00And I don't know about you, like to bring it back to the political part a bit. Um, I've been seeing so much more on like different social media and actual news outlets, um, just reporting on how much uh damage AI is doing. Like before Chat GPT, we had companies going out and being like, look at how sustainable we are, and we're gonna meet all of our targets by X date, and like even bragging about meeting the targets earlier than originally planned.
SPEAKER_00And now we are like spinning up data centers that kill people's water and just completely destroy the environment, um to a uh at a rate that we have never seen before. And none of these companies are kind of learning from the experiences of the past. Like, why did we have all these sustainable sustainability efforts to begin with? And then it's like, oh, whatever. Like we just need we need the tokens and we need everyone to be using this.
SPEAKER_00And you know, to me, it's like, yeah, I I do want to keep using AI, but I don't want to be seeing people using it for things that, like, to your point are not contributing to the bottom line or increasing value, whether it's for themselves or for other, like for their companies, whatever. Um just a waste of resources, like real resources.
SPEAKER_01Yeah, I'd I'd I'm not up on all that yet. I'd like to understand it better. Um because like I've I've heard about that, the environmental impact of data centers and stuff like that, and the the water usage and that kind of thing. I don't quite understand how. Um because like one one criticism is just like the sheer volume of water that it uses.
SPEAKER_00But then not just water, but fresh water.
SPEAKER_01Oh, okay.
SPEAKER_00They don't recycle it.
SPEAKER_01And then like other uh economic activities like agriculture, of course, use enormous amounts of water, um, like many, many times as much as data centers. So I don't quite understand what's different about the data centers versus these other uses.
SPEAKER_00Yeah, that's a fair point. I haven't looked at that comparison metric either. Um, what I've been seeing reported is a data center can take up, I don't know the exact metric. I'm in my head, I'm saying 300,000 homes, but that might be off. So one data center takes up that many homes of electricity. And excuse me, it's it gets to the point where um some I forget what state it was in, but they were saying that a data center, I think it was Facebook's or Meta's, one of Meta's data centers, had opened up and they like completely tore down all the trees in that area to even build the data center.
SPEAKER_00And this farm that was right next to them to that data center, they started realizing that their water pressure was very, very low. And then they started seeing traces of sediment coming through their water. And um, I don't know, I should know more. You probably know all of these things because you do so much around your house. But no, I mean, like, I is it the water tank or like something else? I have no idea what they were referring to, but they had something, they had to replace a whole thing in their house because the sediment was so bad.
SPEAKER_00And they were like, this usually lasts like X amount of years, and now it's lasting like only five years, something like that. So it's a lot. And to see like the dirt that comes through someone's faucet, and no one is doing anything about it, they're just like, well, we need the energy, that to me is kind of gross because it's just telling you that everyone else is expendable for corporate profits. So yeah, there's a a real big negative side to AI that I wish we talked about more.
SPEAKER_00And it's not gonna be the individual consumers that recycle and do all these things that are gonna make the biggest impact on the environment. It's gonna be the corporations and how do they find ways to be more sustainable so that we can still benefit from AI, but we don't destroy people's lives for it.
SPEAKER_01Yeah. Yeah, that's a good point. Um, I really want to learn more about it because I don't understand how it works. Um what this is expanding the scope of the conversation wildly, but I really wish that we would get back to nuclear power because that's such a safer and cleaner form of energy than coal. Um, and so that could at least help. Help address some of this. Obviously, that doesn't do anything about the water issue, but at least the energy production part.
SPEAKER_00Yeah, yeah. No, I I've heard for years that nuclear is cleaner, but it's riskier. So if you if something goes wrong, then it goes really wrong. Um, but I haven't read up on the latest and that in years. So I don't know if that's outdated.
SPEAKER_01Yeah. Well, the way I look at it is like a per capita point of view. Like if if you if you total up all of the cost and human life and health and stuff like that from coal versus nuclear, um like proportionally to the amount that it's been used, um I understand that nuclear is so much safer. Um because like coal, you know, we're we're putting out all this uh all these emissions from the coal plants and um all this mercury is getting into the ocean and then the uh the fish the the fish get contaminated with mercury and that has all these and then goes into us.
SPEAKER_00Yeah.
SPEAKER_01Yeah, and we're not having any anything from that from uh anything like that from nuclear power plants. The trouble is like when you have something like Chernobyl or Three Mile Island, that's a big like news headline event rather than coal, which is just doing the same kind of damage every day. Like it's not newsworthy that, like, hey, some more tuna were contaminated by coal today. There's no um PR about that.
SPEAKER_01So it was like really unfortunate that we had these bad accidents early on in the lifetime of nuclear that soured everybody to nuclear. And it's kind of a um, you know, most people are uh not very numerate, not very statistically literate. And so it was very easy to convince people that nuclear is so dangerous when statistically it's safer, but to our like uh caveman brains, that's that's not so easy to for that lesson to sink in.
SPEAKER_00Yeah, that makes a lot of sense. It goes back to I feel like everything is marketing and PR. Everything is an ad, you know. So that's another thing to complain about. It gets tiresome.
SPEAKER_01Yeah. Yeah, I can agree with that. Yeah, that's that's another annoying thing about the AI, is is all the hype around it. And the AI generated social media posts.
SPEAKER_00I can't. Even on LinkedIn, right?
SPEAKER_01Mm-hmm.
SPEAKER_00Oh, I'm so happy you brought this up. I see, like sometimes I see like really, really awesome posts, and I'm like, okay, you actually cared. And then I see AI generated ones where I'm like, I understand that you wanted to make content, but I'm now just getting something from a model that I could have gotten myself. Like, you didn't add value to my life by me reading your post because you didn't really write it. You know, you had an inanimate object write it. And the way that it writes prose is nonsensical.
SPEAKER_00Like it, it could look right, right? Um, but when you're reading things, the biggest thing to me is I had had a professor my freshman year who said, like, read every line and like try to see it as an arrow. And so every sentence should be an arrow to the next sentence. So they should all connect. And then every paragraph should, you know, somewhat relate to the next paragraph. You can't just go from A to Z. Um, and I see AI, like it doesn't do that.
SPEAKER_00It could write a couple sentences and then like the next thing has nothing to do with the previous three. And maybe those are improving with like more, we just had what we're on Opus 4.8 now. Um, so there's improvements, but it still doesn't, the editing is not there. And so it's still obvious to me when I read a post that's AI generated. And I'm like, I don't, do I unfollow you at this point? Because I just want to see what like humans are actually saying.
SPEAKER_00Um, like I want to hear about AI and I want to hear how people are using it, but I don't want to read a LinkedIn post or even like a whole, I don't know, what what are those things called now? The where people stack something. Um Substack. I don't want to read a whole Substack article that is AI generated either. Cause it's like I could have just put that, fed that into my own prompts and gotten some result back, and it would have effectively been the same.
SPEAKER_01So you know, I think there's a profitable lesson here. Um, and I have to think through what exactly it is, but like, okay, so I, as you know, am somebody who I I I hate the term content, um, but I'm somebody who has created content for many years, podcast episodes, blog posts, and stuff like that. Um and the advent of AI has made many things obsolete.
SPEAKER_01Like, for example, Stack Overflow. I don't remember the last time I've even visited Stack Overflow. It's kind of just um relegated to the dustbin of history at this point. Um and then certain kinds of blog posts, like a blog post about like uh I don't know, documenting some function in Ruby or something like that. Like nobody needs that anymore because you can just ask AI. And so a big question for me is like, what is still worth writing and what are people still gonna read?
SPEAKER_01And it it the things that that you said just now um I think shed some light on the answer to that question. Because people don't want I I don't remember your exact words, but like if your social media post is something that I could have conjured up myself with a prompt, then your post is worthless. Like I want something that only a person could have come up with.
SPEAKER_00So I think that's don't you think there's value in like that documentation example that you gave? You know, like I recently I've seen posts that have switched, and it's like, oh, now that all these AI clients are charging, or AI companies are charging by token usage and not whatever subscription model they had before, the bills are way more expensive expensive. Um, and so people are like, oh yeah, we just hired an intern because the intern was cheaper than AI. Or we I think GitHub Copilot or no Microsoft in general, they canceled their CLUD subscription for all of their uh employees because it was too much of a cost for them.
SPEAKER_00Um, they probably just want to use Copilot anyway. And so we're like not hiring people who aren't given the chance to actually make the mistakes themselves. And so if they're leaning only on AI, the AI is just gonna be very confident and spit something back that might not make sense or be helpful. I find a lot of the times, too, if you're iterating, it'll try to give the context of the iteration. And I'm like, no one cares about that. No one is going to read that part. Like, if I'm so if I'm using AI to create a pull request, right?
SPEAKER_00And I have like a ticket that's pretty well scoped, and I as an engineer understand what I want to accomplish out of it, I can refine the code with the prompts in a conversation with the agent to get to a certain place. But what I find is that agents really love to give historical context in a way that doesn't help. So it's there's different, like when you get historical context about an application to understand how you came from A to B and C and D, that is helpful.
SPEAKER_00But if I'm writing a Jira ticket with an agent and it's tell and I'm refining it and it's telling me, oh, a previous iteration, blah, blah, blah. It's like, if I give that to a fresh agent, why does the fresh agent care?
SPEAKER_01Yeah, right.
SPEAKER_00If I'm writing a pull request and there's code or a doc string or something in Python, and it's telling me, like, oh, you know, we didn't need this anymore because we decided to do this instead. It's like, but this is a new pull request. You know, I don't need to give the context of our conversation in the pull request.
SPEAKER_01Right. I've experienced that exact thing when writing tickets and PR descriptions and stuff like that. It seems to be no good, at least not yet, at like putting itself in the shoes of somebody who um doesn't have any context on the thing you're talking about, and imagining what they would need to know from that perspective. Which makes sense because like that's a fairly that's kind of a specific and unique human skill.
SPEAKER_01Uh and like I think the AI can kind of do that, but it doesn't do it without being prompted specifically to do that.
SPEAKER_00Exactly. Yeah, and so to my point of these kinds of posts that you said, maybe they're not helpful to people anymore, but how would we know? Right? Because if these the younger generation of software engineers, if we're even giving them the opportunity to enter the workforce with this profession, how would they be able to learn from those mistakes? Especially as we keep moving towards like, oh, now AI can I can generate like 25 PRs in a day. And so now I'm gonna send that all to the team.
SPEAKER_00That's a lot. And so now it becomes, oh, let's use AI to review the code. It's like the inmates in the asylum are just like reviewing themselves. Right. Right. So it becomes too much for a human to then it's overwhelming, right? Because the sheer volume of the code. And then, you know, if you're not someone who is used to has like, you know, early on in my career, I created a PR that was like 3,000 lines of code, and I no one had told me that that was a bad idea until I asked for review.
SPEAKER_00And I was being stubborn instead of splitting up the PR. So it took like two months for that change to actually go through all of the review iterations and land. It was very thoroughly tested, like by the end of it and reviewed, but you know, we still needed to fix some issues after that, understandably. And that to me was like a huge landmark in my career of like, this is not something to do. I don't want to be sitting in review for something for two months that sucks.
SPEAKER_00Um, so to me, like I try to keep my PRs as close to 200 absolute lines, AI generated or not, so that it's easier for a human to actually review it or even an agent, right? That's fine. I'm not saying don't use those tools. But how do the younger people actually learn if no one has capacity to teach them because we're all overwhelmed with work and the stream of PR is coming through to review, and we're all just talking to agents all the time, right?
SPEAKER_00So to me, it's like there's still value in teaching people that, even if like two people read that and it made a difference in their career. I feel like to me, that would be a win and worth it.
SPEAKER_01Yeah. Yeah, like the um this this wave of AI is revealing so many interesting things. Um like there are these fundamental and very old programming principles that for some reason don't they they don't seem to gain um acceptance or familiarity or whatever over time.
SPEAKER_01It's like five percent of the programmer population is aware of these principles and it's a static five percent. We we don't as an industry we don't mature and get wiser, it's just uh new people are coming in and not learning these things, and then they get old and retire, and more new people come in and don't learn these things, and blah blah blah. Uh I I wish it were the case that our knowledge as an industry would accumulate and we get wiser over time, but sadly that seems not to be the the case.
SPEAKER_01Um, but but more specifically, um things like uh feedback loops, for example, um, and that short feedback loops tend to work out a lot better than long feedback loops, like your example with a 3,000-line PR that took two months to get merged. You learned from that painful experience, and now you write PRs that are no more than 200 lines. Well, I try.
SPEAKER_00Let's not give me all the credit there. Okay, I tried as best as I can.
SPEAKER_01And these things are true whether AI is in the picture or not, but it's like AI is shining a bright spotlight on the fact that a lot of people they're either not aware of these fundamental principles, or they like think for some reason that these principles no longer apply because of AI.
unknownYeah.
SPEAKER_01Um, and it's again, it's one of those things like Trump, COVID, AI. They're like lifting the veil so that we can see these things that were always true, but I never realized them before. It's like, whoa, like so many people just don't know what the fuck they're doing at all, and they're behaving in these crazy ways. It was always true, but now because of AI, it's it's just so much more clear.
SPEAKER_00Yeah. To me, it's so obvious that the folks who let's stick with that static 5%, those folks who are using AI, they're using AI in a way that maintains what they would have done beforehand, but doing it faster. And so the quality output is roughly the same as it would have been if they were just handwriting the code. And to me, those are the artists in software engineering. They are the people that actually care not just about the science, but the artistic flair.
SPEAKER_00And it doesn't have to be complicated to read. Like in fact, I would I would be willing to argue that that static 5% care very much that it's simple code and easy to read. Um, but everybody else is just, oh, I don't have to write the code anymore. Awesome. I'm gonna use AI and just chug this along and get it shipped. I don't care. Right. And that to me is like the most depressing part of AI is that so the people who created all of these, you know, agentic workflows and whatever, and these like CEOs that go and they're like, coding is no longer the bottleneck.
SPEAKER_00It's like if you had actually followed practices that are statistically proven, like studies have been made, you can read about them um online and all of that. Even Cisco has done this where the lower amount of lines of code and like the single responsibility principles and good design patterns, you're and testing, very important, having high coverage, you're actually gonna ship things faster. And that's always been a thing that people have proven over and over and over again.
SPEAKER_00And now we're like not, we're still not listening to it because none of these models have been trained on those practices, unfortunately. And I feel like those people who created all this, they just never cared. It to them, it was a burden to write the code. Whereas to the 5%, it was the thrill.
SPEAKER_01Yeah, that's one of the things that I found that has been a bit surprising. Like how many people uh had like a like a uh I I don't know the right word, like an antipathy that that feels a little bit too strong, but like a a they they didn't relish the work that they do um and that feels too strong the other direction, but like it this has revealed that a lot of people view a lot of this work as a chore, which you know I think makes sense.
SPEAKER_01Okay, I'm just thinking in real time here for the first time about some of the stuff. Um you know, okay, so I always say that I'm a testing chauvinist, like I I think that that is the one right way to do it. Um because of reasons. Like that's not just my opinion, but there are arguments behind this because when you practice uh test-driven development, you're deciding what to do, and then you do it rather than the other way around, which which sounds crazy, but like okay, when I used to do and sorry for like jumping all over, but um when I used to do consulting, people would bring me in often for help with testing.
SPEAKER_01And so I would come in and I would see like, okay, I agree, you you have some room for improvement with testing, but let's look into this a bit and see what's happening, and I would sit with programmers and uh observe their work and stuff like that. And in all the cases, um in addition to a problem with testing, they their their programmers and managers didn't decide what to do before they did it.
SPEAKER_01They would just have a vague idea and just start coding. It's like, wait, wait, how the fuck do you think you can accomplish anything if you don't know what you're trying to accomplish? Like it it sounds so crazy, like it couldn't possibly be true what I'm describing. But that's that's how almost everybody works. Um they just get a vague idea and start coding. And so, you know, I've I've I've sat with hundreds of different programmers over many, many years and watched people work.
SPEAKER_01And quite a lot of people, most of the time they're working, are in a state of confusion and fog and uh frustration because their way of working is so disorganized. And often I will stop people and say, Hang on, what are you trying to do right now? And they don't have an answer. And it's like, what the fuck are you doing? Like, you should know what you're trying to do.
SPEAKER_01And if you don't know what you're trying to do, stop. Stop working because you're not gonna not gonna get anywhere that way. Um, anyway, I don't even remember what my point was. I think maybe my point was that's why I'm a TDD chauvinist, because it forces you to clearly state what you're going for. Um and then you have to encode your intentions into a test, and then you write the code to fulfill um the expectations that your tests, your specifications describe.
SPEAKER_01Um, I'll pause there because now I'm just starting to ramble.
SPEAKER_00No, this is amazing. I see this all the time, uh, and it's only gotten worse with AI because you just prompt whatever and it goes off and does its thing. But the way that I've been using AI for so many years is I still spend the longest amount of time in quote unquote the planning phase, right? So I'm iterating with like what is it that I want to build? What is the point? And the code part should take the least amount of time, right? And that has been true for years.
SPEAKER_00Is like by the time you start writing code, you should have an idea of what it is you're doing. And that part should be really fast, and then hopefully, you know, like I said, single responsibility, pull requests, low lines of code diff if you can, and ask for review and keep chugging along. Like that stuff is tried and true. But for some reason, people get so angry over the testing of it all that they're like, well, we don't have time to test.
SPEAKER_00It's like you don't have time to not test. That's you. Because then you're gonna have bugs and then you're delaying shipping this product that you're supposedly caring to show to your customers. Like, I don't understand it, but there's always been that discrepancy. It actually reminds me of I I had um, and back to like the documentation point a little bit too. I had a manager in my previous company that would say, like, we're we're not documenting anything, we don't have time, I don't care about tests, all this stuff.
SPEAKER_00And I think about him so much now because now it's the inverse. It's like, oh, well, the agents need to know why this existed and it has to be really well documented for them. Like, wow, this like documentation has shifted. Like, people are on the bandwagon for documentation now, um, because it serves AI, not because it mattered for humans. They haven't come around to the testing fully though, which is unfortunate. Maybe next year.
SPEAKER_01Yeah, maybe. Um, yeah, it it is interesting, the like good practices that AI is prompting people to adopt. That's a very positive thing. It's it's amusing to me that this is how this is happening. I I didn't expect it to go this way, but it's good. It's it's causing people to uh be more interested in testing and apparently documentation. I've I don't I I haven't been in places that um that that is relevant.
SPEAKER_01Um but the testing stuff I have noticed because like people are like, well, have the AI write the tests and then write the code and then it can uh refactor without breaking the old stuff and stuff like that. So that is is good to see. Unfortunately, right now, AI is not, in my experience, is not very good at writing tests without a lot of guidance. But I have found that with the right guidance, it it can write very good tests.
SPEAKER_00Yeah. Now I find myself pointing it to Martin Fowler's uh testing. He has like a whole testing site on his website um with different resources. I'm like, go look at this and do it. And then I routinely will have like, if I'm doing a proof of concept and I still want a high level of coverage, I might not care as much as the quality of the test. But when I'm actually going back and making sure it's productionized for a customer to see, I will tell the agent, like, look at these resources, including the Martin Fowler ones, scan the entire testing code base for um low-value tests, things that if you like deleted this, it still would pass the test and delete those or like refactor the test so that they're actually high value.
SPEAKER_00And every single time it has found multiple tests where it's like, oh, this is a stupid test. And it's actually kind of funny to see the agent evaluate a previous agent's work when you give it this kind of context, because then it's like, wow, what a what a terrible test. Like, well, you're the same model, but a different agent. So I don't know.
SPEAKER_01Yeah, it's funny. Well, it's interesting how the model like kind of knows everything, um, but just because it knows everything doesn't mean it's always gonna behave according to what it knows. And so, like, if you if you ask it to write, for example, some small program, it might write a piece of garbage, and then you're like, no, no, no, like remember, like use the single responsibility program, uh single responsibility principle, and it's like, oh yeah, of course.
SPEAKER_01And then it'll refactor it and it's great. It's like, well, why did you do it so bad the first time?
SPEAKER_00In the beginning, yeah. No, there's you know, Cursor, um, I can't say enough good things about that company. I love Cursor so much. Um, and I really hope that SpaceX doesn't buy it. Um, but they actually exposed their like team toolkit to people so you can install it. And they have a DSlop skill. It's one of my favorite skills that exists out there. And um, some some folks at the company we've been building like different skills for people. And so we'll talk about this toolkit.
SPEAKER_00And it's like you you do the iteration with the agent and it can do your tasks on your source code, and then you run the dslop and also the code review agents that exist in these clients now, and it will always come back and be like, this was terrible. We should fix this, or this, yeah, this was so sloppy. Like, let's delete that. And I recently, like yesterday, just learned about the slop grenade. It's used for messages, but I think it applies to code too.
SPEAKER_01I I saw what you wrote on LinkedIn about that, yeah.
SPEAKER_00Yeah, I love it so much. Um, and what was I trying to make the point? Oh, yeah, and we've said to each other, like, why not why do we even need the skill? Why can't the model just like deslop itself in the beginning? Um, and I guess you know, it's all back to the tokens. You you spend more tokens when you have to go through more iterations, but it would be so much nicer if these models just did tried and true programming tactics that have been proven to actually improve the code from the beginning instead of going through these multiple iterations.
SPEAKER_01Well, okay. So the the the tough thing about the nature of the AI we have, like LLMs, um, is that they're not opinionated. They have all this well, okay, and like okay, so they have all this knowledge, but like we should put knowledge in quotes because because okay, it it has all this all these beliefs, um and they are not all compatible with each other and a lot of them are bullshit.
SPEAKER_01Like so much so much programming education is just so wrong, um, and the LMs just read all this stuff and they believe it, and if you say like uh make sure to organize everything in service objects, then it will happily do it, even though I think that that's a a really bad idea. Um it has read all these blog posts that talk about how good of an idea that is.
SPEAKER_01And so I I think we're probably I I don't see how we cannot be stuck with that situation until we have a different kind of intelligence that can be opinionated. Otherwise it'll just believe whatever you whatever version of the truth you want it to believe. Yeah.
SPEAKER_00Yeah. Yeah, for sure. They can be guided in any conversation. They they won't normally like sometimes it'll push back when I ask it questions, and I'll but I have to preface it with like, I don't want you to just say yes to my question. I'm literally asking you to see, you know, and I tried to like test it almost be like, can you give me a valid response? And I think I'm at the point where it's more trustworthy because of the way that I prompt it. But in the beginning, I certainly would go Google because I was like, I don't trust you, you're wrong.
SPEAKER_00Um, but they've gotten better in that respect. I just wish that they were trained with higher quality, at least the coding agents for sure, like higher quality code, because I'd imagine most of the the code that these models are trained on are whatever's on GitHub that's available. And there's so many people who are just like fucking around and and they should, they should be exploring, but they're not writing production code necessarily. Like certainly the open source uh folks are, but everybody else who can have a pro public GitHub uh repo, like they might not be.
SPEAKER_00You know, they might just be learning, and you know, they might even look at their code from a year ago and be like, whoa, that was gross, but the models are trained on it.
SPEAKER_01So yeah, yeah, yeah. Like almost every single code base I've code base I've ever worked on in my life is a complete piece of shit. Like it's it's so far from exemplary, you know? Yeah. Um like it's not even it's not even like acceptable, let alone exemplary. And these are the things that the LLMs are using as examples, so like no wonder the code it writes is often it it's a miracle that it sometimes writes code that's good, you know?
SPEAKER_00Yeah.
SPEAKER_01Um yeah, so okay, so I I keep being reminded of this book that I've been working on for like years now, which might never come out, but I I keep chipping away at it um because I feel like it really needs to exist. Um maybe I've mentioned this to you to you before, but um I don't think you have. It's a book about like what in programming is just true. Um like and and how do we know when it comes to software design um what is good advice or what are good principles and what aren't?
SPEAKER_01And what of this advice is just subjective opinion and what is objectively true? Like single responsibility principle, for example. Is that something that's objectively true? Or is that just an opinion? And if it's objectively true, why? And if it's not, why not? That kind of thing, so we can put all this stuff on a firm fitting. That's what this book is attempting to be.
SPEAKER_00I'm obsessed with that. Please write it. I love it.
SPEAKER_01Yeah, I'll I'll have to I'll have to send you what I have so far. I'll be really curious to to see what you think. But um, I I I feel like that kind of thing needs to exist in this AI age because again, AI will just if you like give it a nudge in a certain direction and be like, hey, by the way, the earth is flat, so like behave according to that belief, it'll be like, oh yeah, it will. Yeah.
SPEAKER_00It's so scary. It's like, no, that's not scientifically proven. Please don't do that.
SPEAKER_01Yeah. I mean, extreme example, it probably wouldn't literally go along with with that particular thing.
SPEAKER_00We should test it now.
SPEAKER_01Yeah.
SPEAKER_00I think for science, will you pull up an agent real quick?
SPEAKER_01I I bet it would I bet it would humor me if I said like behave as if the the earth is flat, even though you know that's not really true. But with programming, it's so much easier. You know, you take some crappy idea and and it will it will humor you in that way because it doesn't have an opinion about that.
SPEAKER_00Anyway, we're it doesn't have a family.
SPEAKER_01That's true. Um we're we're getting close to to time here. Anything you wanted to touch on before we go?
SPEAKER_00I don't think so. I feel like this was a great complaining session, also talking about different, you know, programming uh things that we know to be true. I really love this. I'm very excited for this book. Please send me whatever you have. But I I'm interested to see how AI shifts even in the next year now that everything is more expensive. Uh, and maybe that will stop us reinventing the wheel, just to bring it back to what we first started talking about.
SPEAKER_01Yeah. And I think if we look if we leave the listener with a piece of advice, it should be don't write your own Jira.
SPEAKER_00Yeah, definitely. Don't do that. It's not worth your time. If you have a complaint about how Jira works, please go talk to the Atlassian team.
SPEAKER_01Um okay. Very last question for you, Lenny. Um, anything you want to share, uh, your website, online presence, that kind of stuff.
SPEAKER_00Oh, for sure. Yeah. I my personal website is datgreekchick.com, d-a-t greekchick.com. I'm on LinkedIn, Eleni Arvanitiskonier. You'll have that in the notes because I'm sure people will butcher that if they try to spell it on their own. Um, and yeah, I'm definitely every time we meet, I'm thinking I really need to get more of my presence out and write these things. So it's always inspiring. So I'm I'm gonna intentionally promise to the ether that I actually start doing that again.
SPEAKER_01All right. Well, we will put that stuff in the show notes. And Lenny, thanks so much for coming on the show.
SPEAKER_00Thank you so much, Jason, for having me.
In this episode I talk with Joel Hawksley, Staff Software Engineer at GitHub and creator of ViewComponent, about whether AI is making us dumber, why blocking code reviews struggle under a flood of AI-generated code, and...
In this episode I talk with John Athayde, designer-developer and co-author of The Rails View, about what architecture school taught him about software design. John tells the story of a user test at LivingSocial that...
In this episode I talk with Riccardo Carlesso, Google developer advocate and co-founder of Rubycon, a Ruby conference held in Rimini, Italy in May 2026. Riccardo explains how a joke about PyCon became the conference's...
In this episode I talk with Travis Dockter, organizer of Blastoff Rails in Albuquerque. When Sin City Ruby and RailsConf both ended, Travis decided to start his own conference. We discuss how cheaply you can run a...
In this episode I talk with Julien Danjou, CEO of Mergify, about merge queues: why PRs that pass CI separately can still break main together, and how testing changes in a queue keeps main green. We discuss how Mergify...