The Standup - Jira Bought 2 Browsers???

Alright, are you guys ready for this? Oh, so ready. I am so ready, dude. I'm so ready right now. Okay, you don't even know how much podcasts I'm ready for. Yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah! Anyways, I'm sorry. Go! Go! Welcome to the stand-up live every Wednesday and Friday at 11 a.m. The Lord's time 10 p.m. or 10 a.m. Pacific standard time if you're in the valley.

13 a.m. Crazy time. Third heads. Jose, if I live did go or Jose, if I live did go. Can we meet at 13 a.m.? And I was like, I got this. I know you're talking about it. Alright, back on topic. Today we are doing Jira Bata browser unironically yesterday. Jira, can I say, yeah, I'm going to technically and actually you right now.

They bought two browsers. Oh, true. Mm-hmm. Mm-hmm. Oh, they bought one company. Oh, okay. Two browsers. Oh, yeah, telepathy. Oh! Okay. So this is more. Okay, so the reason why I want to break this up isn't because I think this topic is the single most interesting topic. I just want to see Casey kick confused.

Like that's actually my primary reason for bringing this one up to the stand-up, which is, all right. So yesterday, Adlazion, Casey, are you familiar with Adlazion and some of the wonderful tools they have created? No, I do. I have heard of them vaguely. I am like vaguely aware, right? I know they do some like get-like thing.

Trash, I think, before I had said it was Bitbucket, which if I remember correctly was like a pre... Like that was before get. That's what they used before. They stiffed Linus on it, right? That was like bit, bit enforcer, bit warden, or bit something. Okay. It was something like it. It's what allows, man.

No, no, no. Adlazion owns Bitbucket that became sash. Oh, yeah, we use that. I think so. What was the... So originally, way, way back when before all y'alls was born. The kernel team for... So Linus, like started using this... The tech tips guy? No, yes. The Linus tech tips, the guy who wrote the Linux operating system.

I'm not going to lie. Just to mix those two up. Bitkeeper. Casey, it's Bitkeeper. It wasn't Linus that did that. It's somebody on Linux team broke the terms of service with Bitkeeper and reverse engineered their protocol. And they said they wouldn't. And that's what caused them to have to switch, because then they were getting dry.

Then that's what made Linus get, right? Is they were like, all right, I guess we got to do our own thing now. That was the only thing I remember. Okay, so that was Bitkeeper. So that means I literally have no idea what this company does. I knew they did some kind of, like, get-like thing. I have no idea.

Never touched it. There is going to be a conflict in this feature. Go for it now. Teach it. Take it. Trash. No. Oh, it's so good. Trash. John, run it back for me. I never said anything. Use AI to make this work, Josh. There's going to be a confluence of features like you've never seen before. Something like that.

I don't remember what I was going to say, because I got to mad. Hey, Casey. Yeah. You're going to have a confluence. A Trello. Yes. Okay. So Casey, they made confluence, which is like a little worse wiki. But if you're not familiar with like Wikies, somehow confluence. It confluence is the only known product that you cannot search with that is meant to be for searching for documents for teams that you use, except in the search bar autocomplete search bar autocomplete anybody please type one in chat to say it in the YouTube comments right Now, the autocomplete is the best search of all time, and then everything else about conflicts you cannot use at all.

It's just absolutely garbage. It is crazy. Whether you're a vibe, coder, or a handcrafted free-range NeoVim user, you still need CI to be able to build your binaries, run your tests, or make your releases. That's where Blacksmith comes in. It takes your GitHub actions, and with a one-line change makes them faster and more cost effective.

Not only that, but you can finally get visibility into your CI pipeline right when things go wrong. We're happy to have them as our sponsor for this episode because just like us, this one is truly a no-brainer. Sign up and enjoy faster GitHub action workflows at a much lower price. Blacksmith, twice the speed, and up to 75% reduction in cost.

Check them out at blacksmith.sh. Now back to the stand-up. Everything's going on. I heard after Elon bought X, the entire search team went over to Confluence. That's what I heard. Confluence search is worse than Twitter, like a bit of a worse. It's objectively worse. You didn't think it was possible what it's worse.

They managed to figure out how to make something that shows none of the worse you type in. Well, so you know everything you love about corporate enterprise culture and software. So it turns out you know how like Frito Lay owns every snack you find in the grocery store, even if there's seven different ones.

At last. Asian is the Frito Lay of enterprise software. You're like, oh my goodness, I can't believe that they're taking 17 minutes to load up this custom page with 37 different tags for every single ticket that they have with story points and t-shirt sizes, and everybody has 13 ways to interact with the ticket.

That thing is also made by the same people that can't figure out how to search text in your wiki. Same people. Makes sense. And they also do a stash, which stash is actually the objectively worse version of GitHub, which I know that is shocking to say. That's right, they've been in hate for now, for now.

So since they had nothing to do with BitKeeper, which is the only thing I actually vaguely knew because we looked at it briefly at Rad at the time, way back when before Git existed. So BitBuckit, which I guess is now stash, is what exactly, are they literally just running Git with the front end on top of it?

Or what is it? It's just imagine GitHub for corporations. Private hosted GitHub for corporations. Isn't GitHub also for corporations? This is before and plus you get owned into a complex, like it's like self-hosted. That's one of the big differences that you can self-host, you can commercially self-host your version of it.

So that way you own everything. Whereas GitHub's like, we own everything, you get to say you have, you get your own little slice on our stuff. Right, we won't show it to anyone else except us. Yeah, except for Copa. Copa, we'll love your stuff. Yeah, do. And any other company who wants to look at it.

You can self-host GitHub. OK, well, this was at one point you couldn't. That's, you know, this thing. But what are you saying, Trash? All right, we're going to get distracted. Well, teach about a free to lay. So I Google what free to lay owns, that they make all my favorite chips. Which are sun chips?

OK, yours. All right, I'll keep an anion of raffles. Hey, someone clips smart food. White cheddar popcorn? Yep, that's a good one. Cheetos? Tostitos? Salso? Where do you want to say tortillas? Try to be honest. Let me sound like you were halfway to tortilla. Some of you are, it isn't tortillas. Look, if it's a, if it's a flotilla, it's a tortillas.

They smell the same way. Oh, Jada on a tortilla, please. Delicious. OK, so that is what Atlassian is. They've always built developer productivity, very heavy air quotes on the word productivity tools. But it's always been enterprise-focused productivity. And yesterday, the CEO drops a two-minute video on why they acquired a browser.

OK. And I'm going to give you the best quote I personally think of in the video, which is I already know. Some people use browsers to browse. And so there is other people use browsers to work. And that's why we spent $610 million in an all-cash offer to acquire this company with one product that people loved.

But they abandoned and a new AI one that all I can find is people who hate. And at first, I thought, holy cow, they spent a lot of money on hand money to be able to buy this. But then I realized, Windsurf, which is a tertiary clone of VS Code, sold for like 3 plus billion. Except it didn't, right? Wasn't it more like they just took the people and literally left Windsurf with the company?

No, Google did that, but then Devon bought the rest. So Google got one part of it. And Devon got the other part. Google took the parent. They got two Chris into the account. Yeah, they took two Chris into the account. Very confusing. So here's what I would like to know if I may ask. In this particular case, based on everything you said about this company who cannot implement tech search properly, I'm guessing they did not actually write a browser.

So is this just like we re-skinned webcast? We got a backup, Casey. We got a backup. Sorry. That Lassian is the parent company that purchased the browser company. So the browser company is the one in New York. Browser company of New York, true. Not to make you the browser company of somewhere else. Can I say?

I do want to say that's a pretty sick name. Yeah. I actually think of all the things that they made. Their name is the coolest. OK. So I could do a browser company of California, and I would be a fish unless it's taken. That's kind of Google's thing. But maybe you could be the other browser company. But can you just say?

It's a Chromium fork. So they made a Chromium fork first in this, they made ARC. ARC was this browser that was like a Chromium fork, but they did a lot more stuff than normal for these Chromium forks. They wrote their own whole UI kit thing and all this other stuff so that you could do a bunch of stuff that was not possible to do before in a Chromium fork.

They moved tabs around and let you do tab groups. And they took everybody's favorite extensions and built it faster and better in a native way that was supposed to work better for everybody. And people were like, wow, this browser is sick. I love tabs on the side. I'm going to make that my whole personality is that I have tabs on the side of my browser and set on the top.

I'm so much cooler and better than you. It is objectively better to have tabs on side. Thank you for proving my point. And so then after they had that for a while, they were not getting a lot of market share. Believe it or not, not a lot of normies thought, hmm, tabs on the side is worth me figuring out what the heck browsers are.

So then they were like, you know what we need to do? Casey, can you guess what they needed to do after they didn't get market share? Yeah, I mean, like you said, they did AI. I'm sure. Yeah, I mean, that's obvious, right? AI pivot is like that, you know, there's literally just the thing where it's like, it's like someone is dropping money out of the top of a building.

And there's all these like dudes with buckets running like, I need to catch some of them. Like it's coming down on the AI side. And they're all running over there hoping to catch some of it before they stop dumping the money out of the top of the building. I need to catch them for real. Hey, and drafts are you tasting like a wave of nacho flavor over there or anything?

Did you want to buy it? Dude, like I just finished my bowl of cereal and I was like, I'm still hungry. Does anyone have a connection at FritoLay in the audience? Please, if you're a YouTube viewer later or in Twitch chat right now, we at least half of the people in this call would shill FritoLay like none other on this podcast.

Just throw it out there. For real? Okay. So literally somebody just spent $610 million to buy a browser that because they put the tabs on the side. So here's the deal is they built that browser. So that was called ARC. But then of course, as TJ said, no one used it. So then they actually built a second one called DIA.

Awful. And DIA is supposed to be all, yeah, after D's nuts, that's what it was named for. And apparently it was so heavily AI indifferent that they actually had the rollback some of it and kind of go a bit more traditional than they were originally going for on it because people just couldn't understand it at all.

And they had, they saw this browser and they said, we'll give you $610. And you know what? We need AI browser. And that was the whole page of the two-minute meeting. Oh, just wait, Brian, stop. Trash said he hasn't got it. I got a joke. Play Trash. They had no idea what they were doing. Ah! Ah! Ah! I don't even get it.

I'm actually missing the joke. I have no idea what the joke is at all, but I don't know what the software is either. So the browser's called DIA. So I said they had no idea what they're doing. Oh, man. That's it. Are you kidding me? Did you just, did that just fly over everyone's head except for D? I couldn't help because I was thinking of idea as an intelligent.

I was like, Trash, it's a very different set of currently close. No, chat got it. Trash, chat is spamming it. They were spam, they get it. Man. At least someone gets me. Dude, I'm getting that into a short just for me and you. I don't care. No one else watches that. What, or I guess I should say, Trash, are you sure they didn't have a idea what they were doing?

Oh, man. They definitely had a couple idea. I can't, I don't know. No, I'm gone. Trash, finish the sentence. You've got to think of something, don't leave it hanging. Figure it out. Dude, A idea, that's pretty good. OK, look, I just, here's my first question among many. Yes. My first question is, what does it say about your company that the people you're buying are like, we're going to need an all cash offer for that.

They're like, we don't want stock in whatever this thing is that you've got here because it looks dire. Like, what does that say that it was an all cash offer? My guess is, so actually, I just assume that literally the browser company is just cash in the bag. Like, that's what it says to me when you get an all cash offer is that you are just taking the money.

You have probably some contractual period that some amount of people have to work for for a heightened salary for some amount of time. And the other people are just going to be like, and then do the disappearing thing. But still, it's like, I kind of figured they had a bunch of investors they needed to pay back from before.

I have no idea. This is just pure speculation. But even if investors were like, oh, we can get a nice chunk of Atlassian now, they would have been like, cool. But here they're like, nope, we do not want that. Don't be doing that. To be fair, Atlassian is the only company I know that has built a product that is universally like hated and they're a billion dollar company.

Like people hate Tira. Yeah, trash for a billion dollar company. I'm so impressed that a company can make so many things that are so bad, but be like number one, everything. Like, it just, it's like, it comes off again. Did we talk about Atlassian last time? Yeah, but it's kind of crazy. Like, when you just talk to the whole user base and they're just collectively against this company.

Yeah, but you know, that's a tremendous surprise. That's what enterprise is about. I know. So the way enterprise software, the way enterprise software works is you make absolute dog poop. So think Salesforce or something, right? And then you have like a giant apparatus that goes out to someone's company and convinces upper level management that it's really useful and then they buy it.

And then all of your employees hate it. That's enterprise software. That is what it is, right? The people who are using it don't actually get a say in the decision, which is why it's so terrible. Yeah, my buddy who works at like a food distribution place used to work in sales. They had their own thing and then they moved to SAP.

He literally was like, I have to get out of here because all I do now is answer customer calls of where did our huge pallets of food go? Because they can't find them anymore. They like lot, they had lost so much stuff because they switched their inventory system because big companies use SAP, we're a big company now.

And then it's sad. And this is the same with like all the like, you know, the HR software and the health insurance management and all that stuff, it's just all like complete garbage. But that's because that is not how it works. Like the people who actually use it have like no say, right? It's somebody who's like looking at like budgetary numbers and compliance issues and legal and all that stuff.

And they're just signing off on it and then everyone gets to suffer. That's like what it is. So it's not surprising because enterprise software just isn't about delivering quality software and never was as far as I can tell. So to be fair, I feel like if I were to try to build Gira feature for feature, I might end in a very similar place.

Oh, it's the same, it's the same result for sure. Yeah, right. Like if I really had to build something that made every manager happy, but it, but tried to make every employee also happy, I think I'd end up in the exact same spot where every employee hates it. Half the managers like it. And I'm just like, well, I tried.

Well, I made a billion, I made a billion dollars, whatever. It's, it's trying to solve the like Gira in particular is trying to solve the problem of when our project's going to be finished. Right. Yeah, just return, just return, Rand. Now it would be a more effective strategy. Yeah, return me a more active mod, two years or something.

And whatever comes back, it's like, there you go. This is her new AI prediction model. A bunch of good statements. So what do we think this browser is going to look like? So I think like in the advertisement, they say it's for like the worker, right? So we have a technology worker. Yeah, we acknowledge or agree with the tab.

One billion of them too. Trash, there's one billion knowledge workers worldwide. Yeah. I don't know where that number came from, but sure. I mean, I saw it. I saw it. I don't know how we derived that. But I mean, knowledge workers are there. Exactly. That's exactly what we want. And that's me. They were remaining billion.

They were remaining billions. No, I can't do it right now. You can hook up with that. I can't. I can't do it. It's a family friendly show. Oh, you will not get trash. Well, we'll save that for crime digital. Anyways, go ahead trash. Sorry, you were asking about knowledge workers. Yeah, but yeah. So obviously, it sounds like their sales pitch was like, you have a lot of tabs like Confluence, Trello.

I have a bit. You can't forbid you to use Trello. I know anyone I use as Trello. I'm off to where? Trello is Trello. Can you guys tell me what the hell Trello? Is this by the same company? Trello is Jira, but with color grants. What is the same company as a different company? It's owned by Atlassians as well.

And it's a con bond board. That's it. It's just a drag and drop tasks between places. So what do you? So are you supposed to, if you are a proper enterprise, doing serious development? No, are you supposed to have Jira and Trello? Or you know, one or the other? Trello is, Trello is for side projects and startups, but not anymore because they got bought by Atlassians.

So now it's like lame. Is it linear or whatever is like the new? That's the alternative to Jira. Oh my god, Peter. Yes. So this is the, for me, I feel like this is kind of the funniest thing because everybody, I know who liked browsers made by the browser company. Like it? I'm going to just say it. Here's my hot take.

97.3% of the reason they liked it was because they had cool promotional videos that made you feel like a cool person. And now they're owned by Atlassian. Your artsy hipster browser is owned by the universally hated enterprise company. How can you get a good product? They're going to get chewed up and spat out by the Jira machine.

Like it doesn't make any sense. Just go get that. To be fair, when they did so, Jira did do a two minute-ish video kind of saying why they bought it and how they're going to use AI and we have too many tabs and all that. The browser company did an 18-second video where it just looked at some leaves. And then a little mouse came up and just clicked twice.

And it just said, like, coming soon or something. Some crap or something like that. That's what I'm saying. All of their videos are like, so when we did terminal.shop launch week, we paraded like what they did intentionally as like a meme and tried to take ourselves as seriously as they could. And I think we still didn't reach that level.

Like I still couldn't take myself as seriously in a parody as they did. So it's like, I don't know how getting acquired by Atlassian is going to go over with their audience base. It would be like, your favorite local coffee shop gets bought up by Starbucks. I don't know. We're still going to be independent.

Guys, we're still just a small coffee shop fighting against a man. I did take the browser company of New York as now wearing like a full like three-piece suit like kind of things from Wall Street and they're sitting in Greenwich Village. And you're just like, guys, you're not currently must-ass in the bow tie.

Now they have the suit on and all that there. They're not the hips or something more. Yeah, okay. Right. And so that part is, I don't know how that's going to go over. As for everybody's reaction who I knew used ARC before that a lot of them left when the DSNUT scenario happened. Like they were all like, my artsy browser, it's gone.

Seriously question, do you all actually switch browsers? I did switch to Zen eventually just to try it out. The side tabs. I wanted to see where the side tabs any get. Is it good? Did you find it really? My actually favorite parts of the essentials. You actually have like three different levels of things you can say as like super important, sort of important, and then just everything else.

And that's been pretty useful. That's funny because like I just been using the same setup for years and I just see all these people like switching to this and then they ultimately come back to where I was. And I'm just like, Casey, what browser are you using? That's what I want to know. Yeah, Casey, Microsoft Edge.

Currently brave, but not for any real reason. Other than I just wanted a Chromium based browser because Firefox kind of tanked. Like it doesn't really, Firefox like barely runs anymore. It's like, it's kind of questionable. So, Firefox is actually awful. Firefox, Zen is Firefox based, it is awful. It's pretty bad now.

And I don't expect it to get much better. So I just, I don't know, like I use browsers for grudgingly obviously. And I don't really have to use them for much actual like work. Except that now that I do have a website that I have to maintain, it's more work. So I actually do unfortunately spend more time in browsers than I would like.

So if there was like some really amazing browser, that'd be great, but the problem with browsers is usually not the browser. It's the web architecture. So I don't only see how switching a web browser is really going to help much, you know. I don't see it being something that's going to be useful long term.

Putting like some kind of random, you know, sprinkling a fairy dust over the top of the same underlying engine is not going to solve any problems that I have, really. Yeah, it's almost as if I wish I could order something through SSH, right? It's true, if only there was an alternative way in which the internet could be that.

Where is your bag? I don't want it. Do I need to order coffee via WSL? You didn't get it yet? No, he did need to do WSL and then SSH and then order coffee? I saved this. I don't have one. I don't have one. Oh, someone else's picture. Oh, what do you still haven't gotten yours? No. Was I supposed to get one?

Yes. A hundred percent. Unless you lied to me and you said you signed up for Cron for no reason and then you didn't do it. I actually did not. Correct. Okay. I completely forgot. I completely forgot. I will do it. Casey, check this out. Ah, yes. That is the camera. When I get mine, I'm not even going to open it.

I'm just going to put it in my office. Trash, I'll send you some more. Here's a question for you. It's a big day tower. We are making fun of trash because he did un-hashtag ads, ad ads. Just in case, you know, I don't want to be mistaken about any of this. Could you tell me something about like the flavor and quality of this coffee since I can't myself experience it right now?

Just to be clear, this is for entertainment purposes only and does not constitute an advertisement. I just want that to be clear before we do this because somebody on here got themselves in a little hot water on x.com, the everything application. This one is a medium blend from Papua New Guinea. Okay.

It's not just any coffee, not just your hero type, type hero reference. Oh. Yep. Hashtag ad is a harmony of dried fruit, milk chocolate, and floral peach. Oh. And the color actually is going to add this back. That's good. We're going to add this bag to the shop for a little bit as a limited time. So that some people who weren't crown members can get done.

Because I know there's a lot of trash supporters out there. Yeah. Yeah. I can't drink caffeinated coffee anymore because I get too freaked. I get too wired. I can see we got a sick decaf blend, 404. Oh, really? And I have not had this and I don't have a bag of this for what reason again? Well, because you haven't gone on to SSH, terminal.shop.

Do I look like someone who knows how to use SSH to you? I'm the person who still does not understand and will never understand why I have to use SFDP to transfer files instead of just, when I'm in SSH, just saying put the thing. Like how is that not a thing? You just said you're an SSH, you just lied to me.

What? There you go. Do you like this, Casey? 404, caffeine not found. Yes. I'll send you some. Talk to me after. You guys know what I'm saying, right? Isn't that confusing? I don't understand it. How come you have to, like if you're an SSH, why can't you just transfer files to that? That's seem like you should be able to.

SCP, secure. Copy. Copy. You can just type SCP now and copy something from the local machine to the remote machine without running. The 80s were wild. Can you actually do that? Yeah. Yeah, SCP is. It's something to help you just. It's literally just SSH for. R-sync true. R-sync is even better. It was crazy because like they made R-sync and then after that they made Dropbox and Dropbox really just R-sync right.

No, no. I want to get the straight, I want to get the straight because you guys might be lying to me here. I want to know you're telling me the truth. No, SCP is real. SCP is real. So if you are in SSH, let your terminal right now. Use Weasel 2. Okay. And just type in SCP dash dash help. No, no, no, no.

I mean, in an SSH, you want to just be like, I'm on this machine, I want to copy something. That's right. Oh, yeah. Got you. Got you. I don't know if that's possible. See? See? This is something like you're right there. You're on the machine. You have a tunnel set up. You just want the thing to transfer a frickin' file.

And it's like Z upload, is that a thing? Nabro. That sounds like a site where you download NX. And yes, it does. Okay, I'm just throwing it up there. Yes, it does. Are you ready to move to the next thing here? Yes. At the end of the day, nobody knows what happened with Cheer Up, so we're just going to have to call that one and do it.

Well, we're going to have to wait and find out. I know, but every single person I've talked to just said, I have no idea. I have never understood. I have no idea what they're going to do with this, other than rolling it. No one's going to use it, 100%. Yeah, it just seems confusing. Anyways, yeah, go ahead, go ahead, go ahead.

All right, so this next section, which I'm calling too much AI, I saw this little tweet earlier, and I thought that maybe you would like to take a little look at it. It wasn't me. Yeah, yeah, yeah. I tried manually coding a feature, and I just can't anymore. It's like I know how to do it, but my body simply refuses to do this manually again, no matter how wrong the LLM agent will be.

That too much AI, have we finally hit the peak of too much AI, but hold on, there's more. This little beautiful one right here is everybody's favorite company Coinbase, which is 40% of daily code written at Coinbase is AI generated, and they're trying to get it to 50%. Which honestly, when I hear of a company that is based on irreversible financial transactions, you know what I hope for?

High percentage of AI coding. I was just going to say like, so I am looking forward to the next end Coinbase hacks of which there will be many. There are just generating 10 million dollars. It's not going to change anything, but. They're doing, that is all tailwind changes. There's no way they're letting it have, he literally says in the tweet on that one, some code should not be AI generated.

It's surely that they have like a million vibe coders pushing forward random features on the front end, and the dips are all plus 30,000 lines, you know what I mean? Like that, please be the case, Coinbase, please be the case. They're probably also potentially somehow considering like tab auto completion.

Maybe they're like hand waving, like, ah, 20%, which is just, you know, like there's just saying, oh, that also counts. I don't even know what someone needed. I mean, like a lot of code, so. I can look at a project, you know, on GitHub or something that was made before AI, and I'm like, yeah, I could believe this was all AI generated.

Like the coding quality is like super low, right? It's just like piles and piles and piles and piles and piles of stuff. Clearly Swiss cheese, like if you were a security researcher, you'd just be like, well, I'm getting a lot of papers out of this, right? So I would believe it. I don't think we weren't rolling, we did not roll into the AI era with high coding standards.

We rolled in with rock bottom coding standards. So I believe people when they say 40% because I'm like, I can't tell the difference between the code you're writing before and the code that this AI is putting out. Looks the same. It's just complete garbage where you just pile crap on top of each other and duplicate code 50 million times.

It doesn't look different to me. So I believe people when they say that, maybe they're lying, but I don't think they have to be because I could believe it, you know? I don't think Coinbase is a Fort Knox of security, right? Fort Knox doesn't even have golden in anymore, Casey. Yeah, man. Actually, we're all the, we're all the gold.

Where's the gold on it? Where's the Fort Knox gold on it? Trash, you can look at Bill Gates microchips have all the gold in him. They needed the gold for the chips, for the microchips, for the five dream receivers. Trash, look up when was the last time we got to go see in the inside of Fort Knox. The only thing I remember from Fort Knox is in die hard with the vengeance and that movie was awesome.

Didn't see it, I didn't see that one. Love the first one though. The idea is, have we hit two, is this too much AI? Have we finally hit it where people just don't even, they're like, ah, because it's not like LLM's can just do everything. So we're kind of in this weird world where you do have to manually code quite a few things.

But if everyone's just like, I don't even want to do it. Like what, what, what does that mean? What does that say for us? I mean, I use AI to do like the stuff that's just really easy to do, but what does that mean? So like, I don't know his word. Oh, it's my job. They'll be like, say it, say it by just seconds.

It's time to show up to the podcast. Like, oh, I haven't like, just like, this is my job right now. Why do you think I can be here? Dude, I just, I just ship like 30 new features just now, okay? I'm getting paid and I'm shipping features. Don't be like, if it's just like, hey, test is for me and like these different scenarios are like, yesterday, I had to like test some cashing stuff and I was like, I gave it some like, uh, to do is of like, test this scenario for the cash and make sure the cash isn't validated.

It just did all the tests for me from what I could tell properly or like, or just like, small, refactor stuff that are just kind of like grunt work. I will always delegate that to that. I like that. Yeah. For what I can tell, probably. I will say it. You're like, I told you the test of thing and it tested the whole thing.

I'm like, I mean, it did, I was worked for like an hour. It must have done something. I don't know. Like literally like 80% of the time I'm like, that's not right. And it's just like, you're right. I should. So it's kind of annoying and that's some, at some point, I just stop and I'm just, I do it myself.

But, you know, we'll say the trap. Yeah. Like I'm quite a bit more willing to change like the shape of something if like, let's say we have like 10 instances or like 10 to 30 or something like, because I'm working on more Dory right now for our game and like sometimes we like make enemies one way to start and I realize we need to change it.

I'm like, a lot more willing to change that now than I used to be, because I can just say like, hey, here, I'd be like, you know, whatever, cursor or opa code or whatever. Like, here's the file that looks right now. Here's the directory of the rest of them. Like, make them all look like this one. That's literally what I'm trying to do.

Yeah, yeah. That's literally what I do. Like, here's a work in example. Just copy it. Yes. Don't do anything else. Goal fix the other ones. There's 20 of them that are the wrong way. We had to like move this thing into this trial thing and now we're passing it here. Boom. Even, like, even, you know, sometimes sure, you could do a macro and you could do some other stuff like that.

And like, I could get through quite a few of them sometimes, like, pretty quickly. But sometimes it's not quite so obvious, right? It's like, I'm changing the name of something to be slightly different. And it's not really like macroable, really, or whatever. That is an easy way to generate 5,000 lines of code.

Exactly. That is essentially zero in a way, right? Because it's like, I changed a bunch of stuff. Sure, it's AI generated. So it's like technically it's true, but it's really like, I just ran a macro on this thing. So it's more like, you know, the change number of lines. Like, for some reason, we think measuring how much code AI writes is good by lines of code.

But we would never decide whether software developer is a good software developer to company by measuring how many lines of code they did this year. There's some stuff like, did they write any? Like, okay, well, they got to write something. Do something good. So there's some cases where it's obvious, but you wouldn't be like, ah, Tom over there.

He wrote 100,000 lines last week in JD's all. And so he's really the guy that we love, Tom's a genius. You'd be like, no. In general, it's usually the other way around. So I would usually measure productivity by features and how few lines of code it took. Right? If you tell me you can implement this feature and 100 lines of code, that's way better than someone's like, oh, man, I implemented that feature.

And I wrote 30,000 lines of code. I'm like, because I see that a lot online. And I'm like, do you guys have any idea what you can do in C in 30 lines, that thousand lines of code? Let alone like something like Python where you got a whole game inside your application. You literally can do an entire game engine in around 10,000 lines of code if you want to, right?

And so you did 30,000 lines of code and they're like, I moved this button and you're like, what the hell is wrong with you? Like, how is this possible? So there is a lot of that where I look at the things that people are stating as, like, supposed evidence that things went well and I'm like, that sounds terrible.

But again, I mean, I think when you roll in with such rock bottom standards, I mean, that apparently skates as something that is, quote unquote, evidence that this is working well, but sounds like it's working terribly. I would love to see, we have 90% of a code that's getting deleted is by AI. Yeah, that's what you want, going through.

That's what you want. And it's five rows, that's what it should be. It should be like, I ran this thing on this 500,000 line GitHub repo and it reduced it to 50,000 lines and it still does all the same stuff. That's the AI you actually want, right? Yes. Maybe. Can it reduce entropy? That's what I want.

I had so far, I've seen it add a lot of entropy to my systems, which sometimes it needs, that's okay, it could still be productive and helpful or you're in JavaScript land and everything's entropy. There's no such thing as, just, I got installed 37 tools to get through my day. But yeah, I would love to see more, like hey, 30% of our PRs are code reduction PRs by AI.

It starts up stuff and finds patterns that we wrote the same 37 times throughout the code base and wraps it up into one. Which it feels like it should be able to do, but for some reason is not doing. So I don't know, like I don't really know what's going on there if it's just a case of like the people who are training these things, don't train them properly or what?

I don't know, or maybe it's just really hard, probably. It's a lame demo compared to one shoting something. I think that's one of the big problems is like a lot of stuff. It's based on the demo. It's zero to one and not, you know, 38 to 42. Right, you know, they just don't really care about 38 to 42 as much.

Are these football terms? Because I don't know. I don't see like zero to one, like the touchdown. And yes, no, I said zero to one, like I don't have any code to like a minimum, like an MVP. I knew that one, but I know the 38 to 42. I just think two random numbers of like farther in the future. All right.

I still don't get it. Yeah. Sorry, I'm just doing a scale of zero to 60. 60 being here. I was just making a random number. Oh, God. And hoping that they weren't so out of code to get somewhat better. Like I just want things to improve. And right now, AI seems to be doing the opposite. That's the thing, right?

So I don't think they can do even what you're saying with the whole like fine patterns and reduce it. I'm not sure if AI can do that because it might, because I'm positive if you took tree sitter and you just said, how many like similar patterns do I have in my code base? Just like structurally speaking, you could find patterns all over the place throughout your code base structurally.

And it's just like, what does it mean to an AI to find things of similar pattern or shape? You might, I'm not sure if this, this might be actually a really hard problem. That is not very obvious to us because for us, it's very easy to see what should be, or I know I say that so flippantly, but a lot of people get it wrong, but you get the idea that it's easy to look at things and be like, oh, I can make this more simple by moving it to here, but that's also one of those amazing parts about humans is it takes 100 neurons out of something to cat and it takes trillions of operations for an LLM to see if something's cat.

Right. Well, I mean, but the vast difference is incredible. I just mean like, all right, I mean, that could be true in which case, AI is just doomed, right? But I'm like, if all the people out there who are AI optimists, I'm like, you should be showing me this. Like, show me the thing that actually does something good instead of the thing that always poops out poop, right?

Like, encoding it feels like we're still at the thing where it generates 17 fingers on the guy's hand in the photo, right? Like, that's where it feels like we're at in code. And so like, I want to see someone that's like, no, it can actually draw a human fairly reliably now. Like, get us there at some point.

Please. I mean, because people are using these tools and they're very bad, in my opinion. Yeah, I've had good success if like the problem is well-scoped. It's in a like code base that has similar patterns and I can say like the one we were saying before and trash, I think you've said you've done this too of just like, hey, I changed this one.

I updated it. Now I need you to go do this in like 40 files. Just go spend some tokens. Burn some electricity and solve this for me. Like, that actually has worked really well for me and it is nice because it frees me up to fix my previous mistakes, right? In the sense that like sometimes you'd be like, ah, it's just not really worth it to like go and change the structure of this because you're gonna go find all the places and do all this stuff.

If you can just be like, go find all the places and fix those for me, it's nice. God, I've also, I've used it like for a wide range of stuff but I do think people would be like better served by showing real examples of like, I pointed it at this code base and it reduced 2000 lines of code. That, for me, I would be like, that's awesome.

I would be way more excited for your project. Yeah, it feels like it would be very useful and what is more needed like in a lot of these projects that I see, like they're just, when I look at them, I'm like, okay, you guys literally took like 10,000 lines of code to do something that's like literally 100 lines of code just because no one really did what's going on.

And I, maybe that's something AIs can't do in the near term but it kinda doesn't feel, it doesn't feel like that should be the case. I just feel like they probably just don't have good training sets to like, they probably just don't have a way to really train it to do that right now but, you know. That was one of the things we didn't get to at last, last episode.

But like, I think honestly, the not, one of the things people are underestimating about the training data on GitHub is not just that you have code there, but that you can actually see the changes over time and like training on diffs and other stuff like that. Or recently I saw an article where open AI, maybe in Microsoft or someone else, I don't know, was thinking about licensing like cursor edits or something for people who have like that turned on or whatever, like.

Yeah, because it needs to see that, but of course the problem is if everyone is doing the thing that I'm complaining about, which is, hey, I just spammed out 10,000 lines of code to move the button over here or whatever, then that means the AI learns that way and that's a problem, right? So, and I don't know how you fix that problem other than you have to shift the way that you're doing your AI to one that's more of an adversarial training process like more like an AlphaGo kind of a thing, which I guess they just haven't really quite gotten enough of that into the process yet or whatever.

I'm gonna be, I'm still gonna have a job writing code later because they're gonna just be paying me to write code to train AI off me. See, so that's my job security, Jason. Do you think it's gonna be like a boutique thing in the future where it's like your, you kind of like have an officially licensed, like kind of like, you know, you have officially licensed sport, like this is Michael Jordan's shoe.

It's like, this is TJ's like, you know, AI that you use and it programs like him or whatever or you can be like him when you program. We should use that. I want trash. I want trash is my coding buddy the whole time. He's like, great work. Let's eat a Dorito. You know what I mean? I'm like, yeah, I want to be the modern day clippy.

All right. Can you imagine me, can you imagine me just in the bottom right here in your corner? It's like, dude, that's sick. Trash, we need to make that. We actually need to make that. It looks like you're trying to be incredibly baller. Go for it. Oh, you know what? You deserve a Dorito. Let's go. All right.

I gotta go to Jitsu. I'll be right back. And then it disappears for like two hours. Oh, no, after me for like a couple of hours. I gotta go work out. And there's a little silhouette of it, like going, like doing all this stuff and like throwing and doing the crosser dancing avatars on windows when they have those, like you could have the dancing avatars.

Yeah. Oh, man. Well, that stuff is evolved now. You know that, right? Like there's an entire market of idle games that run down in the task bar, basically, like on the bottom of your windows desktop. So, you know, that could be where that could be where Trash is ultimately headed. Yeah, I like that trash.

We can make that happen. We can immortalize you. Trash Idlers, a new genre of Trash Idlers. Yeah, it's got to go. All right, so I'm going to attempt to steal man some of the AI, I think, things that has been flowing through a lot of, like the, the normies thought process on this all, which is that computers are fast enough to process a lot of code.

And we call these things like tech that or poorly written code. But if an AI can fix the poorly written code and it doesn't matter how many times it's copied, it doesn't matter about all that. If it just keeps on abstracting you away and it's just able to take care of all of those headaches, it doesn't matter if the diff was 10 lines or 1,000 lines.

These at the end of the day, computers are fast enough and we should be able to produce out the exact, like feature set we want, regardless of the quality of the code. And when I need to refactor, I don't think about refactor. I'm just like, hey, update this and it goes and it finds all the places and it does either the manually updating or maybe God forbid it, it reduces entropy and does some sort of refactoring.

But at the end of the day, it's just like, no, that's not a problem. That's not the way we think about things. Well, there's several problems with it, right? The very first one and most important one for this domain is security. Because the more lines of code you have, the more exploits you have, almost always, right?

So that's problem A, because we're talking about, a lot of times we're talking about web development stuff. So it's not like a video game or something where you're just running it in some situation where maybe you just don't care because it's like completely, it's on the PlayStation and it's completely isolated process or whatever.

Who knows? The thing too is performance is already abysmal. I mean, I'm already sitting here, I click a button and I wait, you know, sometimes is a good thought, by the way. This is just me trying to steal manliness as much as possible. The problem is computers aren't fast enough. They just aren't. We've still managed to slow them down and the problem is the way in which they slow down right now is very, very, very tied to the structure of what you're doing.

So if you have your AIs generating all these like server round trips or whatever because that implements the feature, well, your customers will be waiting longer and longer and longer to click the button or to open up the app or all the other things, which people claim is fine, but whenever anyone does a study, they find that like every additional 100 milliseconds cost them money, like cost them actual revenue, right?

And so it's not okay, right? It's not okay to sacrifice performance because it actually, not only does it ruin everything for everybody who actually uses your software, but it actually does end up hurting you in the long run too. So that's a problem. Thing number three is, where's the electricity coming from?

Someone still needs to explain where the electricity is coming from, right? Liking? Okay, come on. I was going to say Zeus, what are you talking about? Dude, that's how we get our electricity. We just get lightning. I really still at some point want to hear where the electricity is coming. It's going to be building more nuclear power plants, I guess.

I'm not sure. But obviously, we've had this whole push of like everyone in Silicon Valley is falling all over themselves to tell us what great people they are and how wonderful they are and how they're on the right side of all these social causes. And then like also, by the way, we're going to use roughly the power consumption of Germany for all of our data center things added up, right?

I mean, I think Germany's like 60 gigawatts and one AI data center is one gigawatt. Can I give you a fun fact? Yeah. I did a little bit of research on this. And so to fly a fully loaded jet from LA to New York, yeah, happens about 22 to 26 times a day. Okay. It would take approximately 183 days of full flights back and forth to equal one day of CHEDGPT processing power in GPT-5.

It's not good. Right? This is not good. Every two days is a year worth of burning that much stuff. When I walked back to the future, though, he just did it like in his car. He made a lot of jiggle-wats casey. So like, I don't understand what the concern is. Well, the concern is just like, where's it come from?

And maybe the answer is like, we build more nuclear power plants or something, which might be fine, but that is not happening currently. So I don't know. Is it just we're burning coal or something, which is not going to be a good outcome, right? The thing of long mass or juice clean out of the ground, buddy.

This is the way people want dinosaurs back. So if you put it all back in the air, we can have dinosaurs again. We can make the planet warm enough for dinosaurs again. This is the whole point of the whole thing. I see the movie, the dinosaurs die at the end. And then Scarletoe Hanson has to go to an island and save people.

Scarletoe Hanson too, and like in CHEDGPT. Do you remember? Same guy. Got in a little bit of trouble with that. What happened? He tried to clothe. He asked Scarletoe Hanson want to be the voice of CHEDGPT. She said, now we've got a copycat question right there. And that was like, oh, we didn't get you.

We got someone else. That was like the beginning of the whole AI. Yeah, crazy. OK, my serious thing that I want to say about what you were saying, Casey, too, is that even if you think AI is still going to write out your code, it's actually better if AI can do it more concisely for regular people because you're going to pay by the token.

But that actually leads a perverse incentive for AI companies who at least for people training models. I think like Cursor would prefer you to take less tokens because they don't want to have to charge you more money. But like open AI, they like all the code takes a lot of tokens and it turns a lot all the time.

Because then they say excellent. Your entire business depends on you sending this back to me and you don't know what to do with it. So they're like, they have a very perverse incentive, actually, to like, well, we'll just keep pushing the context window bigger. That's really what you need. You really need two million tokens of context.

And load that bad boy up. Load it up, you know, like. Which also means that they have age and problems, too, right? Oh, you know what, we need the, our agents are going to get extra smart. They're going to double check every file. Right. Yeah. I would also point out that unlike, so one of the nice things about a current data center, like if you were just look at like a data center that you built for, I don't know, AWS or something, or a Google data center.

For the most part, they're pretty multipurpose. Like you can do lots of different stuff with it. You know, you can, you could run some really slick, very custom, very efficient application on it that was written to bare metal or whatever you want. Right, you could do all kinds of stuff. Or you can run like 12 layers of bloated Python.

You could do whatever you want, right? And the data center will just, they're pretty general purpose. And there might be certain use cases that they're maybe not optimal for. But it's, it, we can, it can adapt, right? And so all of us out there can use it for various reasons. And a identifier can only do one thing.

And one thing only. And that is low precision matrix multiplies. That's basically it. Like it's not going to do anything else well. It's not going to be power efficient doing anything else. It's going to be very power hungry if you were to run anything else on it because most of the equipment in it and the entire structure and like it's liquid cooled, right?

Is it's not going to be air cooled like a normal data center, blah, blah, blah. I mean, there's some, there are some data centers that are liquid cooled, but either way. I'm saying like they're designed totally differently and they have way more power being pushed through them because that's how AI, you know, AI is a very sustained high load kind of a thing.

So we're also building out a tremendous amount of infrastructure that's completely useless for anything else but this. And that also sucks in my opinion because what if this turns out to not be that great? What if we find there's a much better way to be doing this stuff? Like what happens to all of this build out?

So that's not very encouraging either in my opinion, but I guess their answer is just, well, we'll just build more data. So they're like, all right, I mean, I don't know. It's just, I don't love this kind of stuff. I like general purpose computing. I don't like this hyper specialized kind of thing, but that's where we're at.

Confirmed KC likes Python, 12 layers of things over Kuda. Got it. Confirmed. I mean, I hate to say it, but actually yes. Like, that's kind of true. Like I love GPUs and I love GPU programming. I think that's very interesting. If you're a gamer and you bought one, what's that better software conference?

You're not inviting KC back this year. Yeah, but I love the fact that the data center can do that. I love that. I absolutely hate the fact that I have to use code that's 16 layers of Python deep or whatever, right? But I absolutely love the fact that that worked. And an AI data center literally cannot run that.

It would just roll over and die, right? So, you know, ask the AI to evaluate it and guess what the hell is that? I mean, I shouldn't say it can't run it. I mean, obviously it can run it. It can run it. It just costs so much more power to run it. Is the right way to say it, right? It's like, it's not like it can't.

It's just it's so much less efficient. Yeah. So, well, that was pretty neat. Okay. It's a cool story. It's a cool story. I agree with you. But hey, guess what? I think that means we're done for the day, huh? Sign it off for Friday. Sign it off for Friday. See you later.

The Standup - Jira Bought 2 Browsers???

Kapitelklicken springt direkt hin

Shownotes

Transkriptvom Podcast bereitgestellt · Zeitmarken springen hin