The Standup with ThePrimeagen - The Standup - Jira Bought 2 Browsers???
Episode Date: October 28, 2025Thank You! https://blacksmith.sh our #sponsor today! Speed up your GitHub Actions AND pay less! https://twitter.com/terminaldotshop - Want to order coffee over SSH? ssh terminal.shop 📍 Chapters:... 00:00:00 - Intro 00:04:06 - Blacksmith #ad 00:04:44 - Who is Atlassian? 00:15:34 - How Enterprise Software really works 00:22:30 - Switching Browsers 00:24:15 - Terminal Coffee #ad 00:28:55 - Did AI make us lazy 00:42:15 - AI Data centers and their impacts Become Backend Dev: https://boot.dev/prime (plus i make courses for them) This is also the best way to support me is to support yourself becoming a better backend engineer. Great News? Want me to research and create video????: https://www.reddit.com/r/ThePrimeagen Kinesis Advantage 360: https://bit.ly/Prime-Kinesis
Transcript
Discussion (0)
All right. 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 podcast I'm ready for.
Uh, anyway, I'm sorry.
Go.
Go.
All right.
Let's do it.
Welcome to the stand-up live every Wednesday and Friday at 11 a.m.
The Lord's time, 10 p. or 10 a.m. Pacific Standard time.
If you're in the valley.
13.m. crazy time.
13.
Jose, Philippe did go.
or Jose Valemden go, can we meet at 13 a.m.?
And I was like, I got this.
I know, I know what you're talking about.
All right, back on topic.
Today, we are doing.
Jira bought a 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.
They did.
Well, they bought one company, two browsers.
They suck.
Yeah, tell them, try it.
Tell her.
Okay, so this is more.
Okay.
So the reason why I want to bring this up isn't because I think this topic is the single most interesting topic.
I just want to see Casey get confused.
Like that's actually my primary reason for bringing this one up to the standup.
Oh, he's going to do.
All right.
So yesterday, Adlasian.
Casey, are you familiar with Adlazian 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
Git like thing
Trash I think before
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 or Bitwarden
Or BitWorten or BitWitt something
Is it a bit lab or Bitbucket
That was man
No no no
Atlassian owns Bit Bucket
That became Stash
Oh yeah we use that
What was the there
So originally
Way, way back when before all y'alls was born.
The, the kernel team for, like, so Linus, like, started using this.
The Tech Tips guy?
No, yes.
Linus Tech Tips, the guy who wrote the Linux operating system.
I'm not going to lie used to mix those two up all the time.
Bitkeeper.
Casey, it's Bitkeeper.
Okay.
And 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 dropped.
Well, I just meant that's what made Linus right, 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 Git-like thing.
I have no idea, never touched it.
There's going to be a confluence of a feature.
Go for it. No, T.J. has to go.
Dang it. Dang it. Trash!
That ruined it.
It was so good. Trush.
Josh, 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 too mad.
Hey, Casey.
Yeah.
You're going to have a confluence of Trello.
Tickety.
Trauma.
Yes, okay, so Casey,
they make Confluence,
which is like the world's worst wiki.
So if you're not familiar with like wikis,
somehow confluence,
confluence is the only known,
like, product that you cannot search with
that is meant to be for searching for documents,
for teams to use,
except in the search bar auto-complete.
Search bar, auto-complete, anybody,
please type one in chat.
To say it in the YouTube comments right now,
the auto-complete is the best search of all time,
and then everything else about confluits 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 get-up action workflows at a much lower price.
Blacksmith, twice the speed and up to 75% reduction in costs.
Check them out at blacksmith.sh.
Now back to the stand-up.
Everything's going to do.
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.
It is actually worse.
It's objectively worse.
You didn't think it was possible, but it's worse.
They managed to figure out how to make something that shows none of the words you type in.
Casey, 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?
Like even if there's seven different ones, Atlassian 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 worst version of GitHub.
Which I know that is shocking
the same. 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 Bit Bucket, 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?
Just imagine GitHub for corporations.
Private, hosted GitHub for corporations.
Isn't GitHub also for corporations?
This was before, and plus you could own it, so it's not like...
Like it's like self-hosted.
That's one of the big differences is 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 a 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 co-pilot.
Copilot will love your stuff.
Yeah.
And anyone else.
You can self-host GitHub.
Okay, well, this was at one point you couldn't.
That's, you know, this thing.
But.
What were you saying?
trash?
All right.
We're getting kind of distracted.
Well, Teach brought off Friedelay, so I Google
what Fruit delay owns that they make all my
favorite chips.
Which are?
Sun chips.
Okay.
Yes.
All right.
Sun chips.
Ruffles.
Hey, someone clips.
Smart food.
White cheddar popcorn.
Yep, that's a good one.
Cheetos?
Tostitos.
Salsa.
Were you about to say tortillas?
Trash.
I love me.
It sounded like you were halfway to tortillas.
Some vaginas and tortillas.
Look, if it's, if it's a,
If it's a flotilla, it's a tortillas.
They're spelled the same way.
I'm chicken vagina on a tortilla, please.
Delicious.
Go ahead.
So that is what Atlassian is.
They've always built developer productivity,
very heavy air quotes on the word productivity tools.
Right.
But it's always been enterprise-focused productivity.
And yesterday, the CEO drops a two-minute video
on why they acquired
a browser.
Okay.
And I'm going to give you the best quote I personally think of the video.
Okay.
Which is...
I already know.
Some people use browsers to browse.
And it's the name.
Whereas other people use browsers to work.
Okay.
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 hold on, but, so at first I thought, holy cow, they spent a lot of money, cash on hand
money to be able to buy this.
But then I realize Winsurf, which is a tertiary clone of a VS code, sold for like three
plus billion.
Except it didn't, right?
Wasn't it more like they just took the people and literally left Winsurfed with the company?
Google did that, but then Devin bought the rest.
So Google got the, Google got one part of it and Devin got the other part.
Google took the parents.
They got two Christmases now.
Yeah, they have two Christmases now.
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
text search properly, I'm guessing they did not actually write a browser.
So is this just like we re-skinned, you know,
webcam?
We got a backup, Casey.
We got a backup.
Sorry.
Atlassian is the parent company that purchased the browser company.
So the browser company is the one.
Browser company of New York.
True.
Not to be confused in 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.
Okay.
So I could do Browser Company of California
You could trash unless it's taken
That's kind of Google's thing
But maybe you could
Maybe
I'm thinking
You could be the other browser company
But can you just
It's a chromium fork
So they made a chromium fork first
In this is this
They made ARC
Ark. Ark was this browser
That was like a chromium fork
But they did a lot more stuff than like normal
Forks they wrote like their own whole
like UI kit thing
and all this other so that you could do a bunch of stuff that was not
possible to do before
in a chromium fork. They like
move tabs around and let you do like tab
groups and they took like everybody's favorite
extensions and built it
like faster and better in like a
native way that was supposed to work better
for everybody and people are 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 sat on the top.
I'm so much cooler and better than you.
better. It is objectively better to have tabs on the 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're 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, they did it. 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 a 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 side.
I need to cash in on some AI money.
And like it's coming down on the AI side and they're all running over.
They're 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.
For real.
Hey, and Tras, are you tasting like a wave of nacho flavor over there or anything that you want to support?
Dude, I just finished my bowl of cereal and I was like, I'm still hungry, so.
Does anyone have a connection at Frito Lay in the audience, please?
If you're a YouTube viewer later or you're in Twitch chat right now,
true.
We, at least half of the people in this call would chill Frito lay like none other on this podcast.
Just throwing it out there.
For real.
Okay.
So.
Back to topic.
Yeah.
So literally.
Literally, somebody just spent $610 million to buy a browser that because they put the tabs on the side.
So, well, here's the deal.
They built that browser.
So that was called ARC.
But then, of course, as T.J 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 he was named for.
And apparently it was so heavily AI indifferent that they actually had to roll back 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 Lazian saw this browser
and they said, we'll give you
610.
And you know what? We need an AI
browser. And that was the whole pitch
of the two minute meeting. Oh,
wait, Rob, stop. Trash said he has a joke.
I got a joke. Play a trash. They
had no idea
what they were doing.
I don't even get it. I'm
actually missing the joke. I have no idea of the
is at all, but I don't know what the software is either.
The browser is called Dia, so I said they had no idea what they're doing.
Are you kidding me?
Did that just fly over everyone's head except for...
No, dude, I couldn't help because I was thinking of idea as an intelligent.
I was like, trash, it's a very different set of productivity tools.
No, chat got it. Trash, chat is spamming.
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 if 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.
Yeah, they definitely had a couple ideas.
I can't, I don't even know.
I'm gone.
Trash, finish the sentence.
You got to think of something.
Don't leave it hanging.
Figure it out.
Dude, a idea?
That's pretty good.
Okay, 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.
Okay.
From before.
I have no idea.
This is just pure speculation.
Yeah.
If investors were like, oh, we can get a nice chunk of Atlassian now, they're,
would have been like cool.
But here they're like, nope, we do not want that.
Like, don't, 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 Chira.
Yeah, trash.
They are 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 and everything.
Like, it just, it's like, it's like, it's like, it's like Microsoft.
Microsoft again?
Dude, we're talking about
Alexa.
Last time.
Yeah, yeah.
But dude,
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,
like a magic enterprise.
That's what enterprise is about.
So the way enterprise software...
Managers love it.
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 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 awful.
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 had 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 jira 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 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.
I made a billion dollars, whatever.
It's trying to solve, like, Jira in particular,
is trying to solve the problem of when our project's going to be finished.
Right.
Well, can't you just return, just return,
Rand
Maud
That would be a more
effective strategy
Yeah
Returns
RAND mod
Two years or
something
And whatever comes back
It's like
That there you go
This is our new
AI prediction model
A bunch of
If 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
Knowledge worker
Yeah
Knowledge Work
We have a tab
One billion of them
Too
Trash
There's one billion
Knowledge Workers
Worldwide
Yeah
I don't even know
where that number came from, but sure. I mean, I saw it. I saw it. I don't know how we derived that.
How many knowledge twerkers are there exactly?
Exactly. One and that's me. The remaining billion. Trash, do it. No, I can't do it right now.
You cannot come up with that and not.
I get trash. T.J. This is a family friend they show. You will not get trash.
We'll save that for Cron Digital.
Anyways, go ahead, Trash. Sorry. You were asking about knowledge workers.
Yeah, but yeah, so there, obviously sounds like
their sales pitch was like you have a lot of tabs
and like Confluence Trello
God forbid you use Trello
I don't know anyone that uses Trello
Trello
What is Trello? Can you guys tell me what the hell
Trello? Is this by the same company?
Trello is Jira but with
Color Cranes.
But he's in the same company as a different company.
It's owned by Atlassian 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, okay, so are you supposed to
if you are a proper enterprise
doing serious development.
Are you supposed to have Jira and Trello,
or you just have one or the other?
Trello is for side projects and startups,
but not anymore because they got bought by Atlassian,
so now it's like lame.
Isn't it linear or whatever is like the new?
That's the alternative to Jira, right?
Oh, my God, guys.
Yeah, it's been a computer.
Yes.
Okay. It's, 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
liked 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 good.
To be fair, when they did, so,
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 like looked at some leaves.
And a little mouse came up.
It just clicked twice.
And it just said, like, coming soon or something like that.
That's what I'm.
saying all of their videos are like so when we did terminal dot shop launch week we parodied like what they
did intentionally as like a meme and tried to take ourselves as seriously as they could 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 atlasian is going to go over with their audience base it would be
like your favorite local coffee shop gets bought up by starbucks and we're like we're still going to be
independent.
Yeah.
Guys, we're still just a small coffee shop fighting against the man.
I did take a three-in-law.
The Browser Company of New York is now wearing like a full, like three-piece suit, like kind of thing from Wall Street.
And they're sitting in Greenwich Village.
And you're just like, guys, you're not passing.
They used to have the curly mustache and the bow tie.
Now they have the suit on and all that there.
They're not the hipsters anymore.
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 Ark before, that.
A lot of them left when the D's Nuts scenario happened.
Right.
Like, they were all like, my artsy browser, it's gone.
Serious question.
Do you all actually switch browsers ever?
I did switch to Zen eventually just to try it out.
The side tabs.
I wanted to see where the side tabs any good.
Is it good?
Did you find a like way?
My actually favorite part is 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.
Okay.
It's funny because 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 barely runs anymore.
It's kind of questionable.
Firefox is actually awful.
Firefox.
I'm using Zen and it's Firefox base.
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 begrudgingly, 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 would 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 of 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.
Could be bad.
Treyf, where is your bag?
I don't have it.
Do I need to order coffee via WSL?
Do you need to do WSL and then SSAH and then order coffee?
Yes, you can't.
I saved this to open.
I don't have one.
I don't have one.
That was someone else's picture.
Oh, what?
You still haven't gotten yours?
No.
Was I supposed to get one?
Yes.
Unless you lied to me and you said you signed up for Kron for no reason and then you
didn't do it.
I actually did not.
I completely forgot.
Okay.
I completely forgot.
I will do it.
Casey, check this out.
Ah, yes.
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.
We're making fun of trash because he did
on hashtag ads, ad.
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 editing 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 dry.
fruit, milk, chocolate, and floral peach.
Oh.
And we're actually going to add this bag.
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 Kron members
can get them. Because I know there's a lot of trash
supporters out there. Yeah.
I can't drink caffeinated coffee
anymore because I get too freaked. I get
too wired. I see we got a sick
decaf blend. 404.
Oh, really?
Yeah. And
and I have not
had this and I don't have
a bag of this for what reason again?
Well, when you go to that? Because you haven't gone on
to SSH, terminal dot 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 SFTP
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 in SSH. You just lied
to me. What?
There you go. 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 in SSH, why can't you just transfer files through that?
It does 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, SCP.
I actually do SCP dash-h help.
You just, it's literally
just SSH for files. R-Sink, true.
R-Sink is even better. It was crazy
because, like, they made R-Sync, and then
after that they made Dropbox, and Dropbox really
just R-Sink rather.
No, no, I want to get this straight. I would get this straight
because you guys might be lying to me here.
I want to know you're telling me the truth.
No, CP is real.
SCP is real.
So if you are in an SSH...
Just open up your terminal right now.
Use Weasel, too.
Okay.
And just type in
SCP.
dash-tash help. 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. That's what I'm talking about. Got you. Got you. I don't know if that's possible.
See? See? This is what I think. Like you're right there. You're on the machine. You have a tunnel
set up. You just want the thing to transfer a freaking file. And it's like, Z upload. Is that a
thing? Nah, bro. That sounds like a site where you download NX.
Yes, it does. Okay.
I'm just throwing it up there, guys.
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 Jirup, so we're just going to have to call that one.
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 ruin 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 looksy at it.
Yeah, yeah, yeah, yeah.
I said, 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.
Is 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 N Coinbase hacks of which there will be many.
They already leaked all the data to Moonbite's research.
10 million lines of tailwind.
Like it's not going to change anything.
They're doing that that is all tailwind changes.
There is no way they're letting it have hit the.
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 diffs are all plus 30,000 lines.
You know what I mean?
Like that.
Please be the case.
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 how someone even.
I wouldn't believe it.
I mean, like, a lot of, 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,
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 spitting it.
It 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 gold in it anymore, Casey
Yeah, Casey
Yeah, man
Actually?
Where's the gold audit? Where's the Fort Knox gold audit?
Trash
You can look it up
And the Bill Gates microchips
Have all the gold in them
They needed the gold for the chips
For the microchips
For the 5G 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 a Vengeance.
And that movie was awesome.
Didn't see it.
I didn't see that one.
The idea, though, is, have we hit, is this too much AI?
Have we finally hit it where people just don't even, they're like, because it's not like LLMs 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 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 if what does that mean
so like um
I don't know my job
no but like
it's time to show up to the podcast
like I'll have it like
it's doing my job right now
why do you think I could be here
dude I just I just ship like 30 new features just
now, okay? I'm getting paid
and I'm shipping features.
No, but like, if it's just like, hey, test
this for me in like these different scenarios or like
yesterday, I had to like test some caching stuff
and I was like, I gave it some like
to-does 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 small refactor stuff
that are just kind of like grunt work.
I will always delegate that
to the AI.
I like that little caveat.
For what I could tell.
Probably.
I will say it.
I told him to test the thing, and it tested the whole thing.
I mean, it did.
It worked for like an hour.
It must have done something.
I don't know.
But 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 at some point, I just stop.
And I'm just, I do it myself.
But, you know, I will say, though, trash.
Yeah.
Like, I'm quite a bit more willing to change, like, the shape of something.
If, like, 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 can be like you know whatever cursor or open 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 did yeah yeah literally what I do I'm like here's a working example just copy it
Yes.
Don't do anything else.
Go fix the other ones.
There's 20 of them that are the wrong way.
We had to move this thing into this child thing and now we're passing it here.
Boom.
It's like 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 macrobable really or whatever.
That is an easy way to generate.
5,000 lines of code.
Exactly.
That is,
that is essentially zero in a way, right?
Because it's like,
it's,
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,
like on,
on this thing.
So it's more like,
you know,
it's the change number of lines.
Like,
for some reason,
we think measuring how much code AI rights
is good by lines of code.
But we would never decide
whether a 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 to 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 Jay Diesel, 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, but,
features and how few lines of code it took.
Right? So like if you tell me you can implement this feature in a hundred lines of code,
that's way better than someone's like, oh man, I implemented that feature and it, yeah, 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 30,000 lines of code, let alone like something like
Python where you got a whole game inside your application.
Like 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 ago, 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
That's what you want
AI bros, that's what it should be
It should be like I ran this thing
On this 500 million
500,000 line
GitRB 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.
So far, I have seen
add a lot of entropy to my systems,
which sometimes it needs, that's okay,
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 to install 37 tools to get through my day.
But yeah, it's,
I would love to see more like,
hey, 30, you know, 30% of our PRs are code reduction PRs by AI.
It starts up stuff and like 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 a really hard problem.
It's a lame demo compared to one shotting
something. I think that's one of the big problems.
It's like a lot of stuff.
It's based on the demo.
Is 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'm saying like zero to one.
Like the touchdown and...
Yes.
No, I was just saying zero to one, like, I don't have any code to like a minimum, like an MVP.
I knew that one.
But I didn't know the 38 to 42.
I'd expect two random numbers of like farther in the future.
All right.
I still don't get it.
Yeah.
That's right.
I was doing a scale of zero to 60.
60 being your...
I was just making a random numbers.
Oh, God.
And hoping that they weren't a dog whistle.
I just want the 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 could do even what you're saying with the whole like find 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 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 into here.
But that's also one of those amazing parts about humans is it takes 100 neurons, hell if something's a cat, and it takes trillions of operations for an LLM to see if something's cat.
Right. It's just like the vast difference is incredible.
I just mean, 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, in coding, it feels like coding it feels like we're still at the thing where it like 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've had 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 got I changed this one I updated it now I need you to go do
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 like frees me up to fix my previous mistakes, right?
In the sense that like sometimes you'd be like, it's just not really worth it to like go and change the structure of this.
Because you're going to 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.
So 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 at this code base and it reduced 2,000 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,
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 knew what was going on.
And I, maybe that's something AIs can't do in the near term,
but it kind of 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 it 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.
Right.
Or recently I saw an article where OpenAI maybe in Microsoft or someone else,
I don't know,
I was thinking about licensing like cursor edits or something for people who have like that turned on or whatever.
Yeah, because it needs to see that.
But of course, the problem is if everyone is doing,
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 alpha go kind of a thing.
Yeah.
Which I guess they just haven't really quite gotten enough of that into the process yet or whatever.
I'm going to be, I'm still going to have a job writing code later because.
because they're going to just be paying me to write code to train AI off me.
See?
So that's my job security, Casey.
Do you think it's going to be like a boutique thing in the future where it's like,
you kind of like have an officially licensed,
like kind of like, you know, you have an officially licensed sport,
like a 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 do that.
I want trash.
I want trash as 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 just 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 me.
You know what?
You deserve a Dorito.
Let's go.
I got to go to Jiu-Soo.
I'll be right back.
it disappears for like two hours.
Don't ask for me for like a couple hours.
I got to go work out.
And there's a little silhouette of it like going, like doing all this stuff and like throw in a dude across your desktop.
All like when you could have the dancing avatars on windows.
When they had 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.
That could be where trash is ultimately headed.
Yeah, I like that.
Trash.
We can make that happen.
We can immortalize you.
True.
Trash idlers.
A new genre of trash idlers.
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 Normie's 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 debt or poorly written code.
But if an AI can fix the poorly written code.
fix the poorly written code and it doesn't matter how many times it's copied it doesn't matter
about all of 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 a thousand lines because at the end of the day
computers are fast enough and we should be able to produce out the exact uh 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 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, like, 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, you know,
it's on the PlayStation and it's completely isolated process
or whatever, who knows.
Thing two is performance is already abysmal.
I mean, I'm already sitting here.
I click a button and I wait, you know.
I'm not saying this is a good thought, by the way.
This is just me trying to steal man them as much as possible.
The problem is computers aren't fast enough.
They just aren't.
We've still managed to,
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, wait, where's the electricity
coming from? Someone still needs to explain where the electricity is coming from, right? Lightning?
I was going to say Zeus.
What are you talking about?
That's how we get our electricity.
We just get lightning.
I really still at some point want to hear where the lectures come.
It's going to be building more nuclear power plants, I guess.
I'm not sure.
But obviously, like, 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 they're like, also, by the way, we're going to use roughly the power consumption of Germany for all.
all of our data center things added up, right?
I mean, I think Germany is like 60 gigawatts,
and AI data center is one gigawatt.
Can I give you a fun fact?
Yeah.
I did a little bit of resource on this.
So to fly a fully loaded jet from L.A. 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 Chad GPT.
Yeah.
Processing power in GPT5.
It's not good.
Right?
This is not good.
Yeah, so every two days is a year worth of burning that much stuff.
When I watched back to the future, though, he just did it in his car.
He made a lot of Jacob Watts, Casey.
So, like, I don't understand what the concern is.
Well, the concern is just, like, where it's going to 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 burning coal or something, which is not going to be a good outcome, right?
the thing along that
dinosaur juice clean out of the ground buddy
just suck that dinosaur juice
this is the way
people people want dinosaurs back
so if we 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've seen the movie the dinosaurs die at the end
and then Scarlett Johansson has to go to an island and save people
I would love to see a dinosaur like in chat GPT
you remember
same got in a little bit of trouble
with that.
What happened?
He tried to close,
he asked if Scarlet Johansson wanted to be the voice of Chachy Petitia said,
nah, so he got a copy.
Oh, yeah.
And then it was like, oh, we didn't get you.
We got someone else.
That was like the beginning of the whole AI craze.
Okay, my serious thing I want to say about what you were saying,
Casey, too, is that, uh,
like,
it even if you think AI is still going to write all 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 leaves
a perverse incentive for
AI companies who
at least for people
training models. I think like Curser would
prefer you to take less tokens
because they don't want to have to charge you
more money. Right. Right. But like
Open AI, they like
if all the code takes a lot of
tokens and it churns 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 agent 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.
I would also point out that unlike...
So one of the nice things about a current date...
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 multi-purpose.
Like, you can't 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 can adapt, right?
And so all of us out there can use it for various reasons.
An AI Identicator 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?
It's not going to be air cooled like a normal data center, blah, blah, blah.
I mean, 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 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 Casey likes Python 12 layers of things over Kuda. Got it.
Confirmed.
I mean, but I hate to say it, but actually, yes.
Oh, we got it.
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.
Was that better software conference?
You're not inviting Casey back this year?
Yeah, exactly.
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 6.5.
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
help would be. 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.
It's a cool story.
It was a cool story.
I agree with you, but hey, guess what?
I think that means we're done for the day, huh?
Signing off for Friday.
Signing out for Friday.
See you later.
Great Friday, everyone.
