Subscribe
Share
Share
Embed
The latest Elixir news and general software industry topics
Welcome to another episode of, uh, what do we call it? Macro Mayhem.
It has been a week.
I have my buddy, Gus Workman, with me.
My name is Peter Ulrich.
Um, Yeah.
Thank you for tuning in to another episode.
Gus, how are you doing this week? How's it been for you? It's been a busy week.
It's been a busy one.
It's been.
I feel like a lot of people are on, like, vacation right now, and maybe that's me soon.
Hopefully, I can get into some vacation mode, but, like, this week has been it's been a week.
Exactly.
For me too.
So just to give you a time stamp, today is August the seventh.
We are recording at eleven thirty nine in the morning, Amsterdam time slash Luxembourg time.
Yeah.
This is the second episode of Macro Mayhem.
We're gonna go over the news of the last one and a half weeks.
We are recording a couple of days early because Gus is gone and away next week.
I believe you're on vacation, or are you doing something else? Yep.
I'll be Vacation.
Hiking.
Yeah.
So for the Americans, vacation is these days where you don't work.
Usually, Europeans get, like, twenty to thirty days a year, you know, just to get, you know, be culturally
inclusive here.
So first of all, we have a couple of follow ups from the last episode.
We got comments.
People screamed at us, and they said, you're absolutely wrong.
Deep dish in Chicago is indeed a pizza and not a pie.
Is it, though? Is it though? It's Is it though? I think I said last time it's like a casserole.
It's a casserole.
I'm I'm gonna stand by that.
But Good.
It's been maybe what, four years since I had Chicago deep dish pizza last.
Was in Chicago a couple years ago.
Yes.
And I'll be back for Electric Conf US.
So maybe maybe in September Yeah.
We can report.
I'll report back on on whether I still think it's a casserole or Yes.
Gus is gonna be our reporter on the ground, so he will bring you the latest news of deep dish in Chicago
and whether it's a pizza or a pie or a casserole.
Yeah.
Second of all, a follow-up to the hologram news last week.
Indeed, we were, uh, we confirmed Hologram's written in Elixir, and it can be compiled on JavaScript.
And, uh, there also has been news on Hologram.
Uh, lately, they added regexes that can be cross compiled and so on.
So there's a nice blog post out there.
You are following the hologram project, go and check out that blog post.
But let's get started with the news.
First of all, in the last week or so, OTP has released a couple of emergency patches for OTP twenty seven,
twenty eight, and twenty nine.
Each one of these releases contains six to seven pretty critical CVEs, so that's vulnerabilities.
And if you haven't updated the OTP version, please do so.
There are some very serious vulnerabilities in here that can be exploited in the wild.
One of them that I would like to highlight is, for example, that if you use TLS one point two or earlier,
if you make, you know, SSL or, yeah, SSL connections to another server using the the OTP functionality,
a man in the middle attack is completely possible.
Like, if you use TLS one point two or one point one, a man in the middle attack can happen.
There is an attacker that would put just himself or, like, the attacker would be between you and the
destination.
And usually, the destination server so the the server that connects to another destination, they send
a list of algorithms of TLS algorithms they support, and then the server has to pick an algorithm out
of that list.
And if they don't select such an algorithm out of that list, the connection is rejected and, you know,
there won't be an insecure connection established.
Unfortunately, in OTP, there has been a bug or a vulnerability where the server could select any algorithm
that was not advertised.
And there are some algorithms which are just anonymous.
You know, I don't wanna have any algorithm between us.
So that is what an attack can use to become a man in the middle where they just say, well, I'm I'm just
gonna choose the anonymous algorithm.
I don't wanna I don't have to send you any certifications.
That means also OTP won't verify the domain of the attacker.
So, yeah, it's it's very much just, yeah, sure.
I would like to connect to you anonymously, and please don't check my validity or my identity.
So, yeah, this is now fixed.
It is pretty pretty it's a nine point one, one of the highest CVEs I've ever seen out of ten.
So, yeah, if you haven't updated your OTP version, please do so very fast.
And the new versions that you should check for are twenty nine dot zero dot four, twenty eight dot five
dot zero dot four, twenty seven dot three dot four dot fifteen.
That's a mouthful.
But Yes.
Check your check your versions and get upgraded to those ones there.
Yeah.
And yeah.
Update now.
And also, if you haven't updated in a while because you have a legacy system, if you're not up now, you
are setting yourself up for a security event.
So there's no excuse anymore to not update.
Alright.
Right.
Next item.
Right.
Gleam one dot eighteen was released, and this one looks like a big release focusing on the developer
experience.
A few a number of, uh, updates to the language server, a lot of really cool things, rename and find references
now work properly, uh, on certain types of, uh, fields.
I think that's records and types.
And then renaming a file rewrites on rewrites all the imports.
So a lot of these things that if you're just using Gleam and, you know, making updates with your editor,
you're gonna update and you're gonna have a couple of nice nice features now available to you.
So the Gleam team has been hard at work there and know that they're we're gonna link the blog post with
all of the descriptions of the changes down below.
But if you're a Gleam user, go check it out.
If you're not a Gleam user, go check it out.
Go check out Gleam in general.
It's a small and friendly language as I was told.
It is.
Yeah.
And Next up.
Next.
ElixirConf US twenty twenty five.
So last year, the talks are being published on YouTube.
I think they link every video to a forum post on the Elixir forum.
So you can go there and check out all of the videos that are being released.
There's a lot of good talks that are coming out now.
Yeah.
There's about one talk a day, I think, they release now.
The latest one was about Elixir's configuration layers, a visual journey by Stephanie Lane, and many
more.
So, yeah, the the videos of last year's ElixirConf are released now just ahead of the the ElixirConf
US this year, like three, four weeks ahead of time.
So you can already, you know, get up that that feeling of being at a conference than listening to smart
people educators educate you.
And last news item is that the version zero point three point zero of the programming nerves book by
my good friend Alex Kutmos and Yugo Baruna is out.
So, uh, Alex and Yugo, they are actively working on this new book called programming nerves, and it's
a really nice introduction book into getting started with NERFs.
So if you're interested in in, you know, having a couple of projects handed to you with a part list you
can buy and then exactly how to wire everything together and and to learn the the fundamentals of hardware
design, hardware programming, and the nurse framework.
The programming nurse book is an amazing material, like an amazing resource for you.
And Alex and and actively working on this book.
So they release a new chapter every every month or every two months or every couple months.
I don't wanna put pressure on on Alex here.
But yeah.
So it's it's out now.
You can already buy it for a a lower price of forty nine dollars for the beater, and eventually, it's
gonna be seventy nine dollars.
It has four chapters by now, which I think includes three projects already.
So it's already worth a purchase.
And then once you purchase now, going forward, you will you will get the latest chapters as they come
out.
Right.
And Alex and Hugo are authors of other Elixir books.
So they're experienced authors.
They have some really good content out there.
I also know that Alex and Frank, the creator of Nerves, are giving a training based on the book at ElixirConfUS
and I know that there's a you get a free copy of the book with workshop registration.
So Wow.
If you're gonna be in Chicago and you want the book, might as well get the training too.
Right? And just to give you a glimpse, I believe the project that you will work on at the workshops is
gonna be an etch and sketch.
Is that the right word? Yep.
So, yeah, two knobs you can turn, and then you have kinda like a was that a display? Is that a e ink
display? Yeah.
So I made the hardware for that.
So they'll they'll they'll be using the nerve starter kit, is something I'm working on, early version.
Exactly.
But with some rotary encoders for the Etch A Sketch knobs.
It'll be fun.
Yeah.
So you can draw with two knobs on an e ink display, you know, kinda like this old Etch A Sketch where
you can create, yeah, drawings on an e ink display.
Alright.
Well, moving on to the blog post section.
Again, we kinda tried to just highlight the news this week, but there are a bunch of blog posts that
came out.
So I'm gonna start with the first one by Mike Zornick.
He also has been busy during the summer months.
He wrote another blog post called shimming your way off a debt elixir dependency.
And this is a very interesting one because now, especially with the vulnerabilities coming out, you always
have a delay of your libraries adopting new versions of another library.
Right? So in in Mike's issue situation, he used the Timex library, and that library was blocking the
x money library that that he also had in his project because Timex was not yet up hasn't upgraded yet
to GetText, and there was a a version mismatch, basically.
So Mike, he created kinda like a a little module in between that he could put in front of, I believe,
Timex, And that meant that he could kinda give the old GetText API to Timex, but already update GetText
to a new version.
And, you know, this way, he had backwards compatibility to Timex that hasn't yet updated to a new version
of GetText.
So this is a very nice and and small and easy way to kinda keep an old version of a to keep a library
working while you already update another library depends on through a version that has breaking changes.
So this is a very nice and neat trick that I liked.
And thank you, Mike, for writing this blog post.
Yeah.
That was a good one.
Also, next up, Joanna Larson, who was also featured last time for one of her blog posts, published a
new one in her distributed elixir series on building a distributed rate limer with HRW in Elixir.
And I think HRW stands for what was it? Hash random waiting.
No.
Highest random waiting.
And it's a way that if you have a cluster of nodes that you can largely direct the the requests to the
same node based off of a hashed ID or something like that.
But this is another good post, and it's very approachable in her description on how to get started with
distributed elixir.
I think this was the second or is this the third post in her Oh, she she has written so much this summer.
I think she has, like, seven, eight blog posts.
So she's been a busy bee.
Busy bee indeed.
I saw another one recently that that she also made an AT Proto Yes.
A loss library.
So that's another cool one.
So if you're interested in AT Proto, Blue Sky, all that world, check that out too.
But highlighted specifically, if you're interested in the the distributed elixir world, she's got a really
good intro with these series of posts.
So I'm excited to see the next Yeah.
Next couple ones that'll come out too.
Very excited.
The the other one you mentioned is called Latch, that library to add Elixir based OAuth authentication
with AT Proto, so with the Blue Sky protocol.
And actually, the OAuth authentication with Blue Sky is quite tricky because they implemented it by the
book, you know, not by kinda developer experience, but just by specifications.
And that's why there are a lot of tricks and steps involved that you need to get right.
And Johanna, she wrote a a library that makes it easy to just have your Elixir application be able to
authenticate with your Blue Sky account, and that's called Latch.
And she also wrote a blog blog post about that.
A t proto.
A t proto.
A t proto is not just Blue Sky.
That's the big one.
Yes.
There's lots of apps on AT Proto these days.
So Exactly.
And if you wanna build them with Elixir, now it's a little bit easier.
Exactly.
The next blog post we have is by Paul Wicks, and that I found really interesting.
It's called profiling Rust Nifts in Elixir with Hot Path.
That addition that that's an addition by mine.
But the issue that Paul explains here is that he wanted to, yeah, profile or benchmark a Elixir application
that also uses a Rust NIF.
And he used the Benchy benchmarking library written by my good friend, Toby.
But Benchy only benchmarks your Elixir code, and it cannot, for example, measure the memory consumption
in your entire system.
So if you have very memory high or a high memory consumption in your Rust code, Benchy isn't able to
measure that right now.
So it might actually just say, oh, zero bytes memory use, which you think is kinda weird, but it's great,
so you move on.
But the issue here is that the Rust code, you know, is not measured.
And Paul wrote a really nice blog post about how you use the hot path Rust library to also extend the
the benchmarking and the profiling to the Rust side of things.
And I gotta say, I use a lot of Rust in in my applications, and I also ran benchmarks on them, but only
focused on speed.
And then I take the the kinda the system memory as a snapshot, and I kinda try to estimate what is the
memory consumption of this function.
So this blog post actually helped me also ready to benchmark my Rust code and the connection to the Elixir
code much better.
So thank you, Paul, for writing this blog post.
It's always fun, though, when you look at the Benchy output when you're looking at a NIF and it says
zero memory consumption.
You just gotta move on.
It's like, oh, it's perfect.
Another What do you realize? Another, uh, caveat here is that also Benchy doesn't measure other processes
memories.
I think it's only the process that is under, yeah, the the the supervision tree.
So, yeah, if you do some async work between processes in your benchmark, it also won't measure that kind
of stuff.
So, yeah, just be aware that there are some limitations to this.
Right.
Next up, uh, blog post by Jacob Swanner, um, batching phoenix live view updates.
This was a short but sweet one, uh, really interesting based on some issues that Jacob had at work with
live views that were having a lot of events, like I think hundreds of events per second and sending those
down the WebSocket and rendering them on the client was really slow for them and so he implemented a
way of batching those updates so that you can not have a stream of DOM patches but really just one one
big one that happens every every periodically, every time out.
The way that he did that was using the private, uh, key in the socket, which I don't even know like,
before reading this blog post, I didn't know that there was a private key in the It's Jose does Jose
want you to know about it? It's private.
It's private.
Maybe they didn't want you to know about it, so that's why it's not documented.
Well, I think I think this is probably the use for it.
So it's not included in the change tracking, so you can assign assign.
I don't know if there's an assign private function.
I think he uses assign.
Yeah.
Was it we'll have to check that.
Yeah.
But you you you stash some data in that private, uh, that private key, in that map under the private
key, and then he sets up a timer with, um, with a timeout.
And every two hundred milliseconds or one second or however long the period that you're interested in,
it would fire that event and then copy all of that data from the private key into your assigns where
then it gets change tracked and sent down the wire to the client.
So pretty cool.
And I really appreciate Jacob that he had a a single file phoenix playground snippet that you could run
and and see this with just, like, thirty lines of code right at the end of his blog post.
So if you're curious, you can just paste that into IDX and watch it run.
Very nice.
And it's actually true.
You can just do update in socket dot private dot count, for example, and then update the the assignment
there.
And, yeah, usually, you would do socket dot assigns dot count, and those are the change tracked assigns.
But, yeah, the only thing you need to change is socket dot assigns dot count.
You need to change to socket dot private dot count.
And out of a sudden, you have, like, a private state that has not changed tracked.
Very smart.
Very smart.
Indeed.
Thank you, Jacob.
So the next blog post is by my good friend Michael from Parexial.
He wrote a blog post about how other security vendors, AIs, they you know, when they look at Elixir code,
they often either don't find vulnerabilities because they don't understand the language or they over
report every single thing.
That is actually a cornerstone of the language.
For example, cookies between Erlang notes.
Like, that is something I've seen quite a bit.
So Michael just he wrote a blog post about just be careful when you use these security vendors tools
and then point them at Elixir applications because most likely they have no clue what Elixir is or what
it does and what makes it special.
And he took one particular example where there was a security vendor and they wrote a blog post about
an an Erlang function, the binary to term function, which we all know is very insecure and you should
not be using it, especially on user controlled input.
But some people think, well, if I use binary to term and then I just add the save option, well, it makes
it safe, doesn't it? That's what it says.
And that is in Obviously.
Obviously.
Right? It's just, you know, safe version of this function.
That that doesn't exist really.
It only if you add the safe option, the only thing that binary to term does not do is create new atoms,
which is good.
But it still, for example, creates a function based on your input.
So I can still give you a binary input that that defines a function.
And when you do binary to term, even with the save option on it, it will create a function in memory.
And then if I you know, if you if in the code you use that the output of binary to term, for example,
just do an enum map on it.
Enum map can actually execute a function.
It might realize, oh, that seems to be the mapping function.
Right? So I'm gonna execute it and poof.
Out of a sudden, you have a ticker controlled function execution, which is a remote control execution,
an RCE, in your application.
So although you use binaries to term with a safe option, there's still an option, yeah, to create functions,
modules, overrides, configurations, all the good like, bad stuff.
Good stuff.
Good stuff.
Yeah.
In your application.
So just don't use binary to term.
And if you use it, use it with a save option.
But even then, be aware that it's actually not safe.
It just doesn't create atoms, but it can still create modules, functions, and and configurations, and
that kind of stuff.
So yeah.
And Michael wrote this blog post as a warning to say two things.
One of them, binary to term is not safe.
Although there is a safe error alternative, which is Pluck Crypto dot non executable binary to term,
you need to know about that.
But even then, just don't use it.
You know, if you reach to binary to term full binary to term on user controlled input, just don't.
Just don't.
Yeah.
So I I like this blog post because, also, I didn't know that binary to term with the safe option is still
unsafe.
So thank you, Michael, for writing this blog post.
Yeah.
Security is a big one these days, and and Parac's still doing good work in the Elixir community.
Yes.
Michael is is definitely the smartest person on that topic, I would say, in in in our industry.
And the next blog post is by also another friend of mine, Herman Velasco.
He wrote a very nice blog post about diverge then converge with LLMs.
And what he wrote about a little bit is his workflow when he works on big projects.
And, you know, when we've all got started with LLMs maybe a year or two ago, our approach was kind of
a sequential serial approach where you do one thing and you wait for the LLM to finish, and then you
do the next thing based on that output and then the next thing.
So if you do research, you know, first one LLM and then another research task, another research task,
and then you would, you know, condense it or verify it, and then you would write an execution plan based
on that.
Now that obviously changed in the last, what, six months since we now got agents or sub agents to do
to fan out the work, basically.
So instead of having one LLM agent doing research, you can fan out to thirty, fifty, a hundred agents.
And, you know, that's what websites complain about when out of a sudden they get hammered by agent HTTP
requests, you know, that all wanna have the information.
So, yeah, there's a downside to this, but the structural change now is that you can fan out a big task
into smaller task one per sub agent.
And, yeah, German, he writes about his approach where he deliberately diverges, you know, fans out, and
then eventually he converges the results back into a single document or plan, and then he moves from
there, um, so on.
So he has a really nice, uh, example here, yeah, where he splits a review of a pull request into different
agents that all focus on particular parts of the review.
And, for example, a couple of agents focus on correctness, other was focused on security, and then, again,
others focus on simplicity of the code and also bug fixes, you know, simple, thorough, and creative bug
fixes.
So, like, he he and this is actually what Claude does.
If you just do slash code review, it does exactly the same thing.
It also spans out, like, I had a hundred agents looking at my code at the same time, all focused on a
different aspect of the code and kind of the topic they look at.
So, yeah, thank you, Herman, for writing this blog post.
And the last item that I would like to just announce, it's a little public service announcement.
I also saw a blog post and I still see blog post written on Medium.
And I gotta say guys, please don't write on Medium.
I'm at a motorcycle outside my window.
Sorry about that.
But the problem with Medium is that now it's just the private company trying to make as much money as
possible.
And they kinda hide your content behind all these pop ups that ask you to subscribe, and then they only
show half of the blog post.
And for the other blog post, you need to start a subscription.
And, I mean, there are so many hosted services for writing blogs.
I yeah.
Ghost dot org, for example, is a good one.
People know about Substack, but there's always political problems with Substack, you know, that support,
like, right wing very right wing extremist writers as well.
So just from a personal standpoint, I would just say start your own blog.
Like, start your own website.
You know, it takes you an afternoon.
It doesn't have to be Elixir.
It can just be like a Yugo or or Eleventy is also another good framework that people use these days,
and you can just throw that at Netlify, and you have a website running in thirty minutes.
And especially with LLMs these days, just point it at the problem, and it will post the website, and
then start writing on your own website under your own domain.
Like, I I think nobody should be a medium anymore.
If you wanna have a managed system or, like, a managed hosting experience, you know, Ghost is a good
one.
And there are also a bunch on the Blue Sky ecosystem like Leaflet, Overleaf no, Overleaf something else.
Leaflet, PocketBlog Yeah.
And a third one that I always forget.
But, yeah, they're a bunch.
So There's a bunch these days.
Yeah.
And if you wanna do it in Elixir, Nimble Publisher is from DashBit is the The Nimble the standard way
in the ecosystem for writing blog post.
You can use Nimble publisher.
Down markdown files, and it just post converts it straight into HTML on your on your homepage.
Yeah.
All all you need with live use.
All you need is a markdown file and m dex and Nimble Publisher if you want to.
That is from DashBit.
So it's very easy to spin up your own thing.
And it's also kinda like a rite of passage to build your own blog.
So just don't use Medium.
And sorry, but we will also not mention blog posts on Medium.
Like, I don't wanna support that.
Too many paywalls.
Personal pay Yeah.
Beef.
And just you write those tags to be available to the world.
And then Medium goes in between and says, no.
Actually, we don't want that.
You know? And then you can only like, people can only read three Medium articles per month, and then
they have to start paying.
It's just that's not the idea of blog posts.
So yeah.
Anyways, that has been the blog section.
Last up, the meetups and conferences.
Yes.
So we're probably gonna rehash these until they happen, but this month we have next week on Wednesday
the Nerves meetup online via Zoom.
Come join us for a good time discussions on Nerves and Elixir.
Yeah.
We always have a good time.
Around thirty people tend to show up.
Oh, it's me speaking.
Right? Yes.
And it's you speaking.
I it's me speaking.
Listening to this podcast and you like the sound of Peter's voice, you can come listen to him some more
next week Yeah.
Live.
I might actually also have content that you might enjoy, you know, other than my voice and my my German
accent.
So What? Really? I I yeah.
I won't just read a book to you.
I will actually talk about how to not get pound pawned by mythos.
I never know how to say the word.
But, yeah, basically, in these days of AI and LLMs finding all the vulnerabilities, hacking all the systems,
you know, where do we stand right now, and what can you do to protect yourself and contribute to the
ecosystem? So it's gonna be a fun one.
Yeah.
It's next week, August twelfth.
The next well, it's actually an unconference.
So it's not officially a conference, but it is not a meetup either.
So it's an unconference.
Unconference.
Yeah.
It's literally called unconference.
So it's a single day, single track, sometimes two tracks, where people just come together in the morning.
Everyone presents a topic they would like to talk about.
Well, not everyone who has a topic.
Not everyone has to present a topic, you know, but everyone who has a topic presents it, then people
vote on the topics that they would like to hear about.
And then, yeah, people give these talks and and, you know, demos and that kind of stuff throughout the
day.
It's a very social event.
It's a very low key, no stress event.
It's very much just a social meetup gathering with a couple of talks.
So, yeah, it's gonna happen before the ElixirConf US on September ninth.
So that's one day before the on September tenth, the ElixirConf begins.
And it will also happen in October before the CodeBeam EU that happens on October twenty first and twenty
second in Harlem, the Netherlands.
Yeah.
There's gonna be another Unconference the day before on October twentieth.
And you can sign up to these unconference and get all the information and the location and whatnot if
you go to luma dot com slash earl at e f.
So Erling, e r l e f, and then you will see both unconference and the the dates and locations there.
Yeah.
And if it's not clear from the Luma link there, it's put on by the the Erling Ecosystem Foundation.
This is some of the work done to support local events.
You don't have to register for elixirconf u s or code being EU.
You can just go to the UN Conference if you if you wish.
They're independent, but still a lot of really fun content.
You did you go to you went to the Unconference in Malaga.
Malaga.
Yeah.
It was very nice.
It was a very nice, you know, pre conference conference where you just meet everyone and you can chat
to people and yeah.
Because on the conference day, everyone just starts going to these meetups and then it's like, you know,
game on kinda for two days straight.
So the unconference before, it's nice to already kinda warm up, meet the people that you wanna talk to
and yeah.
I really enjoyed it.
Feels a little bit lower lower stress, I would say.
Definitely.
It's a smaller group of people and it's like a nice warm up intro to Yeah.
Conference mode.
Exactly.
And then Plus, it's fun.
It's fun.
Yeah.
So the last conference, Gus? Yes.
Gothmeier happening in Warburg, Sweden, September twenty eighth to October third.
Basically, the whole week of elixir festivities.
There's workshops.
There's talks.
Talks are three days.
It's single track.
Workshops are two days before that.
So, yeah, I there's not gonna be a GoatMire next year, and jury's out on twenty twenty eight.
So if you want your GoatMire experience, if you missed out last year, you gotta get a ticket.
It's not your ticket.
Gotta get a ticket.
And there's also gonna be the AshConf on the October third, the Saturday, and we will also do a Hackathon
on October third.
So you can stay one day longer this Saturday, then you have either AshConf or Hackathon.
They will be in the same building, you know, so you can also go in between these two events, but, you
know, that's just gonna be an extra day of stuff.
Right.
And there's gonna be a lot of fun stuff there.
Yep.
I can't confirm or deny anything, but, um, we we have given out hardware in the past.
So if you're interested in in that, maybe that's another reason to Exactly.
They might or might not be hardware handed out to you.
So Exclusive.
Exclusive hardware.
It any other way than going to Govire.
That's true.
That's true.
Alright.
Well, that kinda concludes our news, blog, and meetup section.
So if you're only interested in that, please feel to hop off.
But if you wanna stay around for the discussion, we have three topics, and we're gonna keep a little
bit shorter this time around.
Last year last week was, you know, a little bit too long.
We got carried away.
We got carried away.
Yeah.
And, you know, like, had too many topics, so we cut it down a little bit more.
Yeah.
So let's start with discussion.
Gus, you had the first topic about deployment strategies in twenty twenty six, the Fly.
Io shifting focus.
What is this about? Right.
So I don't know if you saw this one, Peter, but Fly.
Io has published recently.
Their CEO is stepping down and is gonna be replaced by I think it was the previous Docker CEO.
Yep.
He's stepping down.
I mean, this is all kinda normal startup stuff.
It's maturing into a bigger company.
He wants to focus on other things.
Mhmm.
That all makes sense.
But what's interesting about the blog post is that he said that Fly.
Io shifting their focus, shifting their priorities to their Sprites platform, which, um, if you have
heard of Fly.
Io in the context of Elixir, typically it's a solution that people like to use for hosting.
They have a very nice CLI.
You run Fly Launch or fly publish or whatever, fly deploy, something like that, and boom, your app is
on a hosted website, and it kinda takes care of a lot of things for you.
It's been very elixir friendly.
Chris McCord does work at Fly.
I think he still is there to this day, but he's been there for the past couple of years.
So they've been done a lot of work to be very elixir and phoenix friendly, and it kinda come came as
a little bit of a surprise when this blog post came out because they basically said we're shifting focus.
We're shifting focus and working on these sprites.
Uh, sprites is kind of their product for disposable computers.
A lot of people apparently point their agents at them.
It's not disposable computers.
Right? It's not disposable hardware.
It's like a single use computer they throw away after, you know, it was tainted by an LLM or so.
No.
It's it's more of you they spin up, uh, really quickly instances of Ubuntu or whatever it is to let your
agent roam free in its own little sandbox.
It's an interesting product.
I haven't personally used it yet, but but that's where they've seen a lot of their growth recently according
to this blog post.
And that has the Elixir community wondering what's gonna happen to their hosting services if they're
not focusing on it so much.
Now there was an Elixir forum post by Alex Slade, I believe it was Mhmm.
That prompted this kinda discussion.
And I did see some later links that that the machine platform and how they host stuff and how you host
your apps is actually a lower level to their sprites.
Mhmm.
So they need that machine and app platform to do the the stuff that they're doing with sprites.
But this also kinda got me thinking and also related to my work this week because I had set up a a new
deployment of an app that I'm working on for a client.
And deployment strategies in twenty twenty six with Fyde.
Io shifting focus, like, how do you think this changes things? What do you use for hosting, Peter? And
yeah.
Yeah.
I gotta be honest with you.
I I love Fly and I wish them to succeed, but I stopped using them two two years ago, maybe.
For I I for anything other than the smaller side projects.
You know, anything that needs to run, well, twenty four seven, you know, I do not put on Fly.
And I I I don't know, but maybe going to sprites that are more short term running, you know, that, like,
they run for maybe minutes until the LLM has completed the the the task, and then they spin down the
sprite and start up a new one.
Maybe that is the right goal for Fly.
You know? Maybe maybe that's actually a good strategy shift because I I'm sorry, but, you know, I I heard
many people saying that they migrated away for their production servers from Fly because they had outages
that are not always visible.
So it's not that your web that your server is down.
You don't get any notification, but maybe their routing is down, which happens.
Or then whatever.
Like, they can't start new machines because the machines API responds with a five hundred.
I I had that a few times where you just run fly deploy on a new project and everything looks good and
then just, yeah, five hundred.
We can't start your machine.
Sorry.
And you're like, well, great.
You know, I guess.
It's the one thing I wanted to do.
Yeah.
It's like, just wanted to take this project and put it online, you know, and I have a talk in thirty
minutes.
So it needs to be online.
Wow.
That never happened before, did it? No.
Almost never.
But Yeah.
I mean, I've I've heard other people talking about reliability issues.
I've used them a lot for I I throw my side projects on there usually because it's just so easy to get
started.
I haven't I mean, I haven't seriously been monitoring those side projects, so I haven't really noticed
Mhmm.
Reliability issues.
But and I think they've gotten better in past years.
But what did you migrate to? When what prompted that decision, and what do you recommend? Would you keep
going with that today? Yeah.
So I Are you on AWS, the Behemoth, or No.
I didn't I didn't go to the big ones.
I I moved to DigitalOcean for a while because they just have a nice developer experience, and it's you
know, they're stable.
Like, I never had any issues with them.
And I also run at least five or six production servers on DigitalOcean still for American businesses.
So, like, they're great.
You know? They have pivoted a little bit too hard on the AI thing, I think, now.
So they they have a new data center in Atlanta, and I was excited because they had new machines and a
better connection to all the corners of the states.
But then a couple months after I added my service there, they just completely pivoted that entire data
center to GPUs.
And then you couldn't Oh, wow.
You couldn't spin up a second CPU machine anymore, and you had to migrate to New York or, you know, San
Francisco.
So now I have that little migration issue.
But, generally, DigitalOcean was always great.
But then, I think beginning of this year, because of all the political tension, I decided to move all
my stuff to European servers.
And what I just did is I rented a Hetzner machine, and I installed Kudify on it, you know, and you can
just click together things.
So it's So what is Cooldify? Cooldify is kinda like Heroku, but self hosted.
So if you loved Heroku and just like the click, you know, your architecture together experience, Koolify
is very much the same and better and more features.
And, yeah, you can just install it, and then you say, hey, this is my Docker container that I have in
this registry.
You know, I now use the GitHub registry because it's, like, connected to my GitHub actions.
And then I point Qualify at the GitHub registry and just click deploy.
And it fetches the Docker image.
It spins up the Docker image.
It has a switch over, you know, like blue green where you first set the the new one, you make sure it's
healthy, and then you switch over the traffic to the to the new one, and you spin down the old one.
It has that.
But the the coolest thing about Coolify is also if you wanna deploy any systems like Grafana, Clickhouse,
what Postgres, obviously, any s three object storage that you wanna host yourself like Minio.
Man, there are, like, hundreds and hundreds of kind of recipes that you can just select.
You just select servers, and then you say, I wanna have a new service.
And you just one click deploy, you know, just like spin up this thing.
And I have Grafana running, and I never got Grafana to work anywhere else, to be honest.
It's just It's a prickly one.
Yeah.
I mean, you can use the manage, but even then, I I don't I don't know what the connection string is to
the managed hosting for I could and they have so many websites.
I don't understand any of it.
So with Coolify, I was just like, yeah.
Sure.
I wanna have the whole Grafana, Loki, Prometheus stack.
Click.
Done.
And then and then it gives you a basic setup, and I copy the Docker Compose to to Claude.
And I said, I have these applications running here, like, in Docker as well, and they are named this.
Please add them to whatever.
I don't care.
Just let me look at the metrics in Grafana.
And, yeah, it edited my Docker Compose.
It pointed to the right applications, applications, and then I added Promax to all my Lixi applications
to expose the telemetry data.
And then I I guess that's Prometheus then goes and fetches the metrics, puts it in Loki.
I don't know.
There's so many.
And then Grafana is the the UI part of things where it pulls the telemetry and shows it.
And all of this was literally one click deployment, and then I had the stack running, and then it was
a couple of configurations that Claude did for me.
And within an hour at most, I have a Grafana dashboard that shows all the telemetry of all my applications
running on the same host.
And, you know, it's like the Grafana as a dashboard, those are JSON files, so you can give those to Claude
and just say add me another graph that shows me x y.
And you can run it in your application code base, so it knows, oh, these telemetries are sent.
So it immediately knows the the structure of the telemetry event, and it can just add the dashboard.
Anyway, long story short, like, I have a Hetzner machine running.
I'm unhappy that they had to increase their prices by, like, almost fifty percent in the last couple
of months.
But it's great.
It's super fast.
I have, like, twenty CPUs that I barely use, and I can run everything on it.
So you run all your apps? Because you have you have quite a few, don't you? I have, like, five or so.
Yeah.
Five apps running on one machine.
Yeah.
But it's still at like less than one percent usage.
Right.
But I Wait.
That's I I mean, that's really great for this type of deployment is that you can throw everything.
Like, I mean, long as you're not doing ginormous resource intensive apps, which might want to be distributed
and clustered, which I think you can also do on Hetzer.
You can set up the Yeah.
Private network and stuff like that.
But if you're running this and you have small, medium sized apps, you can throw them all on the same
machine.
And Yeah.
And that's that's really nice.
Yeah.
I have so I have my own Hetzner machine for my private project.
So my blog is there.
Letter to yourself, like a small thing that I built five years ago is on there.
That is all on Hetzner.
But then I also use that Hetzner machine to build local docker images.
So different story.
For more production ready applications, I use Scaleway.
Scaleway is the European hosting provider.
It's one of the two biggest.
The other one is OVH, and they're both in France.
But OVH is, you know, kinda like the Amazon that has it all in big data centers.
And the second one is Scaleway, which also has data centers in Paris, Amsterdam, and Warsaw.
So for Scaleway, it's very much you you just rent a bare metal machine, or they also have managed servers,
but I use bare metal, you get an IP and you can add an SSH key to it.
And then I use Kamal, the Mhmm.
The one from Basecamp.
Right? I'm not happy.
I I I it's fine.
Like, you know, I I kinda I kinda wish there was something better that is not built by DHH.
But for now, I'm okay with Kamal because it makes Have you you seen the Xamal? The one that someone Oh.
No.
Made an elixir? I have not.
Maybe I should look.
What's it called? You have to it's like Kamal but with an x instead of a k.
I wrote that thing.
That's mine.
You wrote that.
I wrote that.
Yeah.
But, like, I I didn't even stop start like, finish it.
Did I send it to you? I don't No.
That's not from me, XML.
I okay.
Sorry.
I had exactly the same library with exactly the same name.
Yeah.
Okay.
This one is great.
This is exactly what I wanted to do.
It's a port of Kamal to Elixir.
Okay.
I gotta look into this.
It's amazing.
You gotta look into it.
Apparently, this this is one of those, what, great minds think alike moments.
Yeah.
Definitely.
It is even the same name.
I I literally took x a a a m a l.
So That's funny.
Oh, that's funny.
I don't remember you sending me anything, at least not recently.
So No.
I started building it locally, then I abandoned it because, yeah, it had other projects.
But, yeah, what I wanted so what what's great about Kamal, and I hope that Chamal or Samal is gonna reproduce
or maybe another project's gonna reproduce, is with Kamal, you have all the configuration locally and
gets version control.
And the only thing you need to change is the IP of your servers.
And then you say, come out deploy, and it's gonna install Docker.
It's gonna install everything it needs.
It's gonna pull the Docker image, and it's gonna deploy it on every machine.
It has different deployment strategies.
And, yeah, you don't need to touch your server other than just having an SSH connection to it.
That that's it.
Right.
So spinning up a new server is literally adding one extra IP to your configuration file and then saying,
come on, deploy.
So Right.
That's really that's really sleek and nice how that that works like that.
I think that this is really interesting that I feel like with these options now, there is it's like it's
a feasible push to have kind of that managed platform as a service experience, but still self hosted
and not on some big cloud providers Yeah.
Platform.
Right? Yeah.
And so, I mean, with Camel or XAML or whatever, you can you can you can port it to not just not just
AWS Mhmm.
Or or DigitalOcean or Hetzner or Scaleway or whatever.
Like, you can go you can put a server on all of those or none of them or your own server in in your basement.
Right? Exactly.
Yeah.
You can point it at the Raspberry Pi, and it will try to deploy a Docker image on it.
I'm not sure whether it works or not.
Yeah.
But the That's cool.
The only downside or the the few downsides, I mean, Kamal, the original Kamal, it's man, the configuration
of it is so bare that it it's just, you know, it they stopped Basecamp stopped developing it after what
they needed.
And then they just stopped.
Right.
Yeah.
And then, you know, when people said, please add this, they're like, well, you can always fork it and
build your own.
You know? Like, they didn't take any ownership of it really as a project, as a community project.
So if you have any most for example, in Elixir, you have the EPMD ports, like the six nine something.
Just to open that freaking port on your proxy that you have running on your machine to another machine
so that you can cluster your Elixir nodes, you have to spin up a second HTTP proxy next to the traffic
one, uh, instead of just configuring the damn one that's already running, but only for port eighty and
four four three.
You know? Okay.
And so, yeah, you have to spin up another Docker container that is also an proxy, but only for this one
port so that your Elixir nodes can cluster through it.
It's just why.
Is that a Kamal, or is that a coolifier? No.
It's it's Kamal.
Okay.
Interesting.
That's one of these things where it's just like, yeah.
We don't care.
Just yeah.
We don't care.
Yeah.
Well, hopefully, maybe you'll have to look at the the Kamal Yeah.
Project and see if they've I mean, with Elixir Native I will.
Keeping that in mind.
I will.
They they might have thrown that as an option.
Yeah.
But So to to kinda wrap this up, so I ended up also doing a similar deployment as you did this past week
with my client project, and I got a Hetzner server and used Docploit Mhmm.
Instead of instead of Coolify, which is, I think, a similar product.
Maybe the UI is a little sleeker, but, generally, it works the same way.
And that was a suggestion, I think, from one of Johanna Larsen's blog posts that I had read a while back,
and I was like, maybe I should give that a try for this.
Nice.
So I did.
And, yeah, it's it's impressive how far these tools have kinda come because it really feels like like
a like a managed solution Yeah.
That I don't have to worry about too much.
I really appreciate the fact that you spin up the Postgres next to the app, and then you can configure
the Postgres to have automatic backups Yeah.
And and all that good stuff to to s three bucket somewhere.
Automatic.
You can pull those back down and restore.
Very cool.
I love to see these projects.
I still think this space is not yet solved.
Maybe it will never be.
No.
But the the downside with Coolify is that it really only works on one machine.
So if you wanna have two machines running your apps, you shouldn't use Coolify.
The problem with Camaro that if you wanna dynamically upscale, you know, add another machine dynamically
and spin it down again.
That's just not possible.
You need to add and delete the IP from your configuration file.
So it's very much, you know yeah.
It's not that hard to spin up another machine.
I use Ansible for configuring it and then run Kamal on it to deploy the application.
But that takes, what, half an hour until you're there.
Right? So if you have a spike in traffic, that's why people use still the managed, you know, AWS and
whatnot services because they can spin up a machine in in, like, seconds.
So, yeah, they're all different angles to this, and I don't think there's one solution that does it all
well.
Yeah.
Definitely no silver bullet, but but this one that that that from your recommendation and also from Yana's
Mhmm.
Blog post, it's it's it works well for me at the stage that we're Yeah.
Developing at.
So Yeah.
Might be interesting for other people.
And if anyone who's listening has other alternative platforms or methods that the user deployed, they
wanna shout out, leave us a message because, yeah, this is it's interesting.
It there's a I feel like it's always evolving.
It is.
Right? Yeah.
And I'm Did you see Mhmm.
The new one, potions dot I o? No.
Isn't that from from Michael? No.
That's both potion shop.
From I don't know who would the name of the person behind it, but ElixirCasts.
Okay.
Potions dot I o.
Deploy phoenix to your own VPS without the work.
That's cool.
Dedicated resources, no cold starts, no surprise builds.
The ease of a prop platform as a service on infrastructure you own.
That sounds very much like Kamal.
Yeah.
Yeah.
It's definitely Elixir oriented.
I think it is a paid product.
I've not tried it.
I'd be curious to.
But Alright.
Kinda interesting.
There's still a lot of development in this deployment.
I'm very happy there's a a healthy competition going on here.
Yeah.
Yeah.
Alright.
Let's there are two more discussion topics, giving the time, I'm gonna super quickly do the next one.
Maybe no discussion.
Just kinda like a if you if you, listener, wanna read more on this, there's a nice blog post out there.
The question that the author I think his name is Nor.
Well, Norpinion is the blog, but that's probably not his name.
But his name is Nor Smith.
And the blog post is called what will more intelligence actually do for us? And the thing here is the
argument here is that people when they think about LLMs and AI, they think about the intelligence of
an AI in terms of a linear scale.
So it will always become smarter, And it's kinda unlimited or infinity is the end, you know, how smart
it can be.
But the author, Noah, he kinda dis argues against this and says that if you look at human intelligence,
you know, we don't have humans who have an IQ of a thousand.
Well, most like, the chance for that is so small that it probably won't be there.
And, you know, and the scale for IQ is also it has an end to it.
Maybe you reach that end and you might be a little bit smarter than the the upper bound of that scale,
but it eventually you you won't have an IQ of three hundred.
Right? You know, so so human intelligence already is not linear and also is is bounded.
There's an upper boundary.
And the the nor the author here, he argues that the same principle holds also for artificial intelligence.
And the the reason for this is that the that how intelligence becomes useful is not that the more, the
better, but rather, the better the output and the efficiency, the better.
Right? So if you, yeah, have better output, but you have to do a billion trillion calculations that take
you a hundred thousand years, that is not useful intelligence, you know, unless you watched the oh, you
read The Hitchhiker's Guide to the Galaxy, you know, where that computer It's just gonna bring Yeah.
Right? Where the computer thought Forty two.
Exactly.
The number's forty Is it useful? Is it useful though? Like, the answer, maybe not.
So argument here of Nor is basically that human intelligence, if you look at it, it is rather specialized
in certain areas, and we, humans, are really good on these areas of of kind of problem solving.
But but we have an upper bound, you know.
So at one point, like, we're not more intelligent because that doesn't give us any more you know, we
would need to use more brain capacity and and energy to to go there.
And we also, you know, physically restricted.
Like, we can't have a head that is one meter square, for example.
So if you look at human intelligence already, we see that it is rather bounded and it is has an optimal
peak at or like an optimal, yeah, peak at one point after which you it drops off inefficiency.
And the argument here is that AI will probably do something similar where it becomes better until it
reaches a certain peak, and then there's a drop off of, you know, efficiency, but also the output, the
result, because it just starts overthinking the result, although it's correct.
And then it might have an incorrect result.
Right? So yeah.
And and the the last thing here is that the real intelligence as we see it rather comes from connecting
the dots of different areas and different problems than brute force calculations of, for example, math
problems.
Right? I mean, even math problems are not a calculation.
They're rather, hey.
Let's try this approach, and then we can add that mathematical trick to it.
And then we can add this approach, and that kinda is like a zigzag way of getting from point a, your
starting point, to the solution in the end.
So it's not a straight line where more power, more brute force will get you there faster.
It's definitely more like a, hey.
We need to add tricks and and whatnot.
So yeah.
Right.
And I think one of the things that we've seen with AI is that there are problems in some domains that
are unsolved hard problems that in similar domains they have solutions for.
Yeah.
But because those domains are not closely related, it's hard for that information to kinda flow between
Exactly.
Yeah.
And that's why, like, cross functional teams are always a big thing because you you wanna cross pollinate
some of those ideas.
Yeah.
It But AI has been able to do this as well.
Exact yeah.
If you think about very specific AI models that can spot cancer in MRI scans or, you know, radios I don't
know what it's called.
Yeah.
Radios scans in your body.
Right? They they look at the scan, and they can spot cancer cells in it.
Like, even I have a friend who does a PhD in the Netherlands, and the the model they use or there's a
model that can that they trained to spot dementia, like the onset symptoms of dementia in MRI scans.
That is a model that has good accuracy, but only on one single scanner in one single hospital.
Because if they take that model to another scanner in the same country fifty kilometers away, that MRI
scanner will have its own characteristics, its own biases, its own inaccuracies in the image that it
gives that the model loses a lot of of correctness, of validity.
So her job, her entire PhD is focused on finding the parameters that differ between MRI scanners and
then correcting the image in such a way that it's normalized for the model to then do the inference on.
Right? That is just one tiny problem in, you know, one country.
That's cool.
Yeah.
But that shows you how Small one.
It's it's small.
But yeah.
I mean, that's the thing.
If I if I get scanned here in Leiden and I have an MRI scan, I, you know, then go to Groningen, which
is in the north of the country, like, that MRI scan is gonna have a different I can't use the the two
and compare the two, you know, because the scan is different.
Funny is that, like, to us, looking at them Yeah.
They're the same.
Right? But the data that's encoded underneath is just so slightly different that it Exactly.
So so that's why, you know, this this blog post by Nora, it it gives you a think, like a good just a
couple of points to think about in terms of does more intelligence help us, or do we rather need, you
know, more specific and then cross functional, cross area intelligence? Yeah.
So have a read and and tell us what you think about it.
Right.
Alright.
And last up here, we'll keep this short because we're already at an hour.
But so the GCC, the new c compiler.
Is that right? The new g n u c compiler.
I think that's on all the Linux c compiler.
All the Linux distributions.
Recently announced that they will decline any significant con contributions made with AI or LLMs with
the exception of test cases.
And I think I mean, this is it's not really news for us on this elixir focused podcast to discuss GCC,
but this is something that I think we're starting to see industry wide is AI use policies, um, and this
is just a big project that that uses this.
What I thought was interesting specifically about GCC is the language that they used saying that they
would decline legally significant Yeah.
Code.
Legally significant.
And that it was the legally significant code that was written by LLMs in, I guess, the compiler itself
or but it also excluded any legally significant code written by LLMs for test cases.
So the the language there is interesting, and I'm wondering if they have concerns about IP ownership
and because, I mean Yeah.
These these these these big projects have really strict adherence to licenses.
Right? Yeah.
It is interesting because the same argument was made by CodeBerg when they said that they don't wanna
host LLM generated repositories because of the legal implications of it.
And I'm curious to understand and that that's maybe discussion that hasn't really surfaced that much
yet in the community.
But we are what is the copyright question about LLM generated code where Right.
The entire function that it might generate or the entire test case it might generate is literally taken
one to one from another project, another repo that has a license that forbids the copy of that, for example.
You know? And then, yeah, that is something it's kinda like what's it called? When you write your your
PhD thesis and then you take parts of other Plagiarism.
Yeah.
Plagiarism.
Plagiarism.
Yeah.
Where, yeah, out of a sudden, sentences and paragraphs you that you copy paste it from other theses or
other articles into your thesis, you know, they are actually people who scan the thesis and compare them
to others just to spot these little sentences and paragraphs between the two documents.
And I'm curious, maybe this will also start to happen if there is a big law law case against, you know,
one of the big labs that says you generated copyrighted code or you basically not generated.
You just copied copyrighted code into this project that you took one to one from this other project.
You know? And if if there's like, if lawyers can make money with it, I think this will become a new industry.
You know? Maybe so.
And it it and this kinda goes back to what we talked about last time is who is to blame here? I would
say that the operator is the one that holds the legal implication because how do you hold Claude accountable?
How do you hold Yeah.
Codex accountable? You can't.
Right? Just because you although Codex or Claude or whoever whatever LLM of choice put that code in your
project, you're the one that merged the PR probably.
Yeah.
You should have done the due diligence.
Right? But I mean, would I don't know.
Yeah.
But would you do the due diligence on every single line of code that the LLM generates? I mean, you wouldn't.
Right? Right.
So then can't yeah.
I don't think they will go after the clots and the the op the codexes of the world.
They will go after the individual small, you know, software engineers, especially then people who put
their code public again because then they can easily scan it and compare it.
And then, yeah, big projects like the GCC and CodeBrick, they have potentially a legal issue there where
a lawyer could come and say, hey, you host code that was one to one copied from this other project, you
know, take it down or, you know, pay a fine, for example.
So Right.
Interesting.
Let's see where this goes.
Problem.
I'm curious if there's any projects in the Alexa community that have already implemented LLM AI usage
guidelines, LLM, these types of things similar in a similar vein.
I do think that, of course, everyone knows that that, like, drive by, sloppy PRs are are a big problem
right now.
Maintainers don't have bandwidth to review walls of LLM generated text, don't have bandwidth to review
sloppy LLM code from people that aren't invested and committed to the project that they're contributing
to.
No.
So that's a problem, but I wonder if people have started to address that in in our audience.
What do they think? It yep.
It it definitely has some potential to become a problem in the industry, you know.
Because even today, if you copy code from was it like an Apache license software? Mhmm.
You need to put the copyright notice in the code you copy.
And, you know, then you have commercial limitations sometimes.
And you also have you need to contribute back to the upstream repo requirements.
Right? Where, like, if you change the code in any way or you improve it, you need to upstream the the
the fix back to the original repo.
So Right.
Yeah.
Let's see what the lawyers decide.
I mean, this side of the pawns, maybe we have fewer problems, but, you know, like, uh, America Inc.
With all the lawyers, they they might have another problem there or a bigger problem.
Well, we'll wait and see.
They'll still have to settle before we make any final judgment on that.
Exactly.
Alright.
And with that, we are at, well, more than time.
This has been a fun hour, Peter.
It has been.
Yeah.
Thank you very much, Gus, for being with me here.
Thank you, listener, for sticking around until the end.
We will take, well, a break that we always take next week, and then the week after, we will be back with,
yeah, another another episode of Macro Mayhem.
If you have anything you would like us to feature on the show, please tag us on Blue Sky or LinkedIn
or Elixir Forum, and, you know, we will we will most likely put it on.
We always need to plan for time obviously, but, know, we do our best.
So any last words, Gus? Yeah.
Check out our blue sky.
I think we have auto post notifications now.
So when when the recording is live, you can get it right away in your feed.
And Thank you.
We haven't really decided on a release schedule, have we? Last time was on a Monday.
Now it's a Friday.
Yeah.
So that's part of the mayhem.
Yeah.
You just It's mayhem.
Just gotta accept it.
Exactly.
It'll be out when it's out.
Yeah.
Yeah.
We tried for every two weeks, but, you know, like, today, we had to do it a little bit earlier because
Gus is away next week, but around every two weeks.
Yeah.
So subscribe to on your favorite podcast show, uh, podcast app or YouTube and, uh, or our Blue Sky account
or the website, micro mayhem dot com.
No.
Sorry.
Macro mayhem dot f m.
Nope.
Dot f m.
Dot f m.
We're one of those podcast extensions.
Exactly.
And, yeah, you will be notified when the next episode airs.
And until then, thank you very much, and have a good day.
Bye bye.