We Survived the PHPocalypse

Good afternoon.

Hey Ed, how are you today?

Doing well. I finally recovered.

From from the Wednesday madness. Was it Wednesday? Yeah.

Yeah.

Yeah, it must have been Wednesday. are you totally patched and back up and running and everything good?

Yeah, yes, I was totally patched. I just I just manage my own sites now. So I just I literally took them all down. and then when c I had moved them onto clouds a few months ago, I think. And then when Laravel Cloud patched everything, I just put them back up. So I I don't think anything's really at risk. You know, like I would be so you know, I don't think anything was hacked.

previously

Right.

before I took them down pretty early on Wednesday.

I didn't take anything down. and part of that is because I have a lot of stuff and it would have taken a lot of time and I hadn't done no prep work in advance to take anything down. So I didn't have like any static

Yeah.

pages to point things at, or it would have just been like they don't exist versus like having a a coming back soon page. And also Cloudflare had announced that they had some mitigations in place. And so they were blocking anything that was shaped like an attack and

it felt like that was good enough for now. And I just kind of kept an eye on, you know, social media and other feeds to see if if anybody was finding there were any cracks in that armor, I would have, you know, done the work to take everything down. But I just just trusted them to keep things going for me and focused on because

Yeah.

I do actually manage some stuff for clients. So

Yeah.

I th I think you've said in the past that you have

basically referred all of your clients to to like a service company that manages all that stuff for them.

Yeah, yeah, my yeah, they they have they use a different they use a tech provider now and I just do some product work occasionally for them.

Yeah. Yeah.

So no maintenance for me.

Yeah, that's that's nice. So yeah, I needed to focus on prep work for my clients because you know, some of their stuff is pretty straightforward, but some of it is like PHP five point six on some ancient, long abandoned OS that that weren't

Mm-hmm.

gonna be any patches for. So so yeah, I just kinda had to, you know, trust in the

the greater forces to keep my stuff safe while I focused on theirs. And it worked out.

Yeah. I

I I felt the whole thing all in all just we could talk about it like this will come out in a few days, right? I guess by the when you have a moment to edit it and then by then everybody will have forgot forgotten about this, but so we could talk about it in h in in high level terms. I felt a good I felt there was a good vibe on the on the internet about this.

it was funny 'cause people were talking about this and everybody was working on it. And then there was the occasional idiot who was like, Yeah, you see, it's PHP. That's it's not secure. And and the whole world just went like just ignored that person, which I feel like happened a bunch of times. I don't know. If it felt like good yeah.

Yeah. Well, I think we're we're used to that.

So, we can block that out. But yeah, I was impressed. And and it's kind of what I would expect from the community. I mean, obviously I think that Laravel did a fantastic job. Cloudflare really stepped up. the PHP

Mm. yeah.

Foundation did a great job. There were some, you know, kind of careless, reckless people who who made things a bit chaotic by leaking things early.

And then there was a company that decided that they were going to be a a bad citizen and act in what they felt was the best interest of their own customers to the detriment of everybody else, and that threw everything into chaos. but

What was the

company? We might as well name them.

Chain ChainGuard, they had

what do they do?

access. Well, so they had access to the vulnerability early. They're a supply chain management company. And so what they basically will do is they'll build for their clients and ensure that, you know, as soon as something, security related is released, that their clients can get it immediately. so it's just basically a company that they they get paid a lot of money on enterprise contracts to repackage the

work that a lot of open source volunteers do. And generally speaking, you know, I'm fine with that because enterprises do need a little bit of hand holding. But they decided that they were going to go ahead and release their patches to their clients six hours before the scheduled release. And of course, as soon as they released their

All right.

patches, the actual vulnerability was out of the wild. So now anybody who wanted to evaluate it and actually start using it to attack,

because before then it has been speculation and a screenshot or a a

Mm-hmm, mm-hmm.

a little clip video or something showing this thing in the wild. But there was no so like Cloudflare and AWS has said that they were seeing attack shaped requests for, you know, up to a couple of days before the release. But I hadn't seen anything that was confirmed that any actual successful attacks had happened anywhere.

So anyway, this company just decided we're gonna, you know, give the finger to everybody else that because you know, all these companies and organizations, they agree to roll this stuff out all ex exactly at

Yeah, yeah.

the same time so that this doesn't happen. So you don't have all of a sudden before everybody's ready to go, you have the vulnerabilities out the wild. And I've been a little annoyed with people running to their defense and saying that, you know, the PHP Foundation dropped the ball or the security res researchers screwed up or

They've known about this thing since August. It was originally discovered back in, 2006 or whatever. Because there have been there have been variations on this flaw in the language that have surfaced several times over the the history of the language. but it was just in August that somebody figured out how to really actually use it to effectively attack a a PHP application. And I don't know if you looked at the diff, but

I it was just huge.

I did not look at the diff.

It was huge. There was a lot of work that went into this patch. And

Okay. Uh-huh.

I think part of the reason that it took as long as it did is because it wasn't a super elegant solution. There are just so many kind of intertwined things there that they had to make a lot of compromises to make this thing safe. And I'm sure

Right. Right.

that.

There were a lot of places where you look at it. It's like there are three different ways we can go here. Each one of them has pros and cons. And it was a lot of that back and forth trying to figure out how do we get this to be as secure as possible without, obviously you don't want it a great performance. You don't want to screw up, you don't want to break anything. so it's not like it was a a tiny little patch that that was put out. It was a big complex change. And

had nobody leaked anything and had Chainguard not jumped early, well, first of all, if nobody had leaked anything, they wouldn't have given that twenty-four hour notice that they did that, there's this critical vulnerability, we're gonna be patching it tomorrow, getting everybody stirred up. It would have just been, we just released an important security update for PHP. Please, you know, please patch. And so people would have done it sort of

Yeah, yeah.

on like a rolling basis throughout the day instead of everybody hitting refresh right at four o'clock and

Yeah, and a million people wouldn't wouldn't have known what it was and tried to hack it at the same time. Yeah.

Right. Yeah. So this was just

Yeah, but people are trying to throw like the PHP Foundation under the bus and the volunteers that have they've done so much work for the language and yeah,

Yeah, they're just idiots. Hmm.

and that's that's where so like I can deal with the the random idiots saying, PHP is dead or PHP is insecure, but when you start throwing like the PHP Foundation under the bus, I mean their their annual budget is less than a million dollars. They're running the language that powers seventy percent of the internet for less than a million dollars. I mean

Seventy percent of yes,

exactly. It's like the curl guy or or what other people, right? they do this stuff in the weekends or they get paid very little

Mm-hmm.

and then people complain. Yeah, it's crazy. Don't don't be a complainer.

Yeah. So anyway,

Don't be a hater. It's not

I guess

that hard,

Yeah, my big takeaway is that the PHP community acted perfectly. I mean, we did we did what we always do. We stepped up, we took care of each other, we took care of our clients. It was it was all well handled. some of the big corporations that just sort of leech off the PHP community, maybe not necessarily they didn't act in such good faith. And you know, I my reluctance

Yeah, yeah.

now I

So what what I worry about for chain guard's users, right, is now that they've proven they can't be trusted with embargoed information, are they gonna be the last to know the next time something big happens? Because if I'm a security researcher somewhere and I know the chain

guard's gonna leak stuff, I'm not I'm not gonna give it to them in advance.

Yeah, yeah. And and yeah.

And it's the like in security you can be an asshole. Right? you can be that company that that 'cause then I don't know.

Yeah. Stupid people. But

It the whole thing

sub supply chain security in open source only works when everybody involved acts like a good citizen. As soon as somebody steps out of line, it just throws chaos into the mix. And and the natural thing to do and the right thing to do is identify those chaos agents and cut them out and just don't let

Yeah.

them be involved because we can't have chaos. We need we need planned, well-executed rollouts.

And everybody who was working toward this ex this well executed rollout did everything in their power to make sure that it was well executed.

Yeah, yeah. It it I found like I just felt that like 'cause people always talk about like, it's the community and the community, right? And it kind of feel you know, and you're like, Yeah, sure, whatever, but like we're all just trying to, you know, do our thing.

But it did feel like it did feel like a strengthening event for Larel, for PHP, for the for the whole ecosystem. And it feels like we're all much stronger now. It also feels like like all our apps, like everybody's more aware of security now. Every and and you know the whole the whole ecosystem is is much stronger now. And and something will happen again. There'll be another thing in in

yeah. Yeah.

six weeks or in three months or whatever, right?

And it'll be for every language and for every ecosystem.

Yeah, the the thing that scares me

And if they react sittily, then

that's gonna be a silly experience for everybody for the next few years. And if they react well, then we can feel good about it.

Yeah. Well,

I I hope we've learned some lessons from this and that people do recognize that, coordination really actually does matter. I don't think there were any huge detrimental effects. I don't know of anybody who was down for a significant amount of time because they couldn't get the patches because, you know, because of the chaos. It was it was a few hours of stress and headaches, but things resolved pretty quickly. And e everybody who was aware and wanted to get patched, I think, is patched and

so the the scary parts of this now are, because PHP is such a widely used language, you know, how many people have WordPress sites that they paid fifty bucks to some guy to throw together for them three years ago and they're sitting on some shared server now, and they're hearing all this stuff about like a PHP exploit. They they have no idea that actually applies to them. so there's I'm sure there's a ton of stuff out there that's not patched. And then you get like this API key leak from Railway, and that's

so that some of these companies that specifically market themselves to the vibe coders, you know, you don't know have to know how to write code. Yeah.

As will keep your stuff secure. Mm-hmm.

And so then when they're getting leaked, of course, these people who are just they're not technical at all, they just, you know, prompt and prompt and prompt till they get the app they want. And all of a sudden their API keys are leaked and they don't even know like where do you find those things? How do you rotate them? That they don't live in this world.

No no no no. And y you know what another lesson was for me? So like this whole go with comp go with providers and companies that that you feel a good a good vibe of. So you know people say like sure Cloudflare is like a single point of failure or Laravel Cloud is a single point of failure. But but those are two companies I feel they have a good vibe.

And they do the right thing

Yeah.

repeatedly, consistently, and then so when something happens, you're like, great, they're on my side. And they and they were on everybody's side. And then you and then you're like,

Yeah. Well yeah.

thank God, I'm not running some I don't have to do everything myself.

Yeah. Well it's not like it's not like Laravel has a perfect track record of never screwing up. There was a a time a couple of years ago where they pushed some wrong button and it messed up something with IP addresses or something at AWS. So anyway, I don't know what they did, but they screwed something up and everybody who was on vapor

It happens.

suddenly went down and was and everybody was down for a couple hours and you know immediately there was a big mea callpa that you know a real

a really good post mortem. I think we all got some credits on our accounts. everybody's gonna screw up. But when they did screw up, there was no trying to blame anybody else. It was just a very clear we mess this up, we take responsibility. Here's how we fixed it. Here's how we're gonna make sure it's not gonna happen again. And, you know, because nobody's ever gonna be perfect in this industry, the what I look at primarily when I'm trying to choose who I want to partner with as far as, you know, taking out as a as a provider that I really lean on or, you know, do any work for,

That's what I look for. And I've I've been really impressed with Laravel throughout the entire history of it as a company, from when it was just tailored running the show up to now being a a growing corporation. They've kept that, the buck stops with me ethos. And yeah, they just grow up sometimes and they they resolve it. And I think you you experienced the pain during during this whole thing because you had shut down your your sites that were on cloud.

I shut down all my sites.

Yeah, yeah, and then they were doing

And then it took hours to be able to turn them back on.

Yeah, then they were doing these like rolling rollouts of of the you know, I imagine hundreds of thousands of of boxes that they are they are running, right? And

Yeah. Yep.

and deployments were shut down and I couldn't restart my s my site without a deployment. And therefore everything was down for a few more hours than I would have hoped. But but

But it's all it it's fine, it's all personal stuff. If this was high revenue client stuff, I would have, made a different plan. But the reason

Right. Well what what I was gonna say though

I shut it down was that it was low low maybe l lose a few new clients for a few hundred bucks, but nothing serious.

Right. but my what I was gonna get at is they're they're aware of that feedback because you weren't the only person that experienced that. a lot of people experience the same thing. And of course, this was not something that they had designed and planned around. This was like a what's the best we can do on twenty four hours notice? but they it's not like they took what happened on Wednesday and then got through it and put it to bed. They kept copious notes of all of the feedback and everything that went right and everything that went wrong.

And they're really working to harden that system. you know, they they have people who are dedicated still to doing just this.

Yeah, so next time it happens. Mm-hmm.

So the next time it happens, it won't take hours and hours and hours. The minute that your environments are safe and ready to go, you'll be able to come back up. so yeah, I've been I've been really pleased. And and they're so responsive too in the

Yeah. Good providers.

in the wake of this, you know, yesterday.

Or I'm sure they were still busy cleaning up some pieces. In fact, I know for sure they were because they just posted they got Forge updates available all the way back to five point six. So I know they were still working hard on this stuff all through yesterday. I was working on something kind of related to this, you know, my own future planning. And I reached out to one of their engineers and within thirty seconds I had a response and we engaged in a conversation. So, like these guys are willing to

set down their their tools and and have a chat with you and you questions in a way that most companies just don't. So yeah, I've been I've been really pleased with them. And and like I said, overall I think everybody did an excellent job.

Yeah. Yeah. So that was that was the PHP day. You know, shall we talk about some other stuff?

Yeah, we can.

st

this

is this is taking over my week, so everything that I have to talk

Yeah.

about is related to this in some way.

Well, I'll give you an example. So so Nuno and and some people launched this social network in the mid I think on Tuesday or something, right? which was quite funny and then they had to take it down because whatever. The whole security thing.

yeah, not only

not only did they launch it and then have to take it down, in between that they launch on Product Hunt and they took number one on Product Hunt and they

It was it was number one. I know.

got some big article written about them in the in the product hunt newsletter. So it was a big high profile, look at this shiny new thing. By the somebody would click on it from the article to see it is down. So just horrible timing.

Yeah, yeah, t yeah, but

but you know, like all good and and and they built it in two days, right? Like in just like a like a forty eight hour kinda like, you know, throw the throw cloth at Laravel. which makes sense. I think it's good. I've I feel I feel there's like a a I mean there's a bunch of social networks and they're all terrible.

Mm.

And it's

not that I want one new social network, but it's nice to have some options.

Yeah, social's a weird one, right? Because it's you have such a network effect with the big ones. you know, like a Twitter, even Blue Sky to

Mm. Yeah. Yeah. Terrible.

some extent, because there is some there is some sort of mainstream buy-in for Blue Sky. But you look at like Mastodon and it's never really taken off. And I'm sure that DevNull will be sort of the same thing. He's gonna have like a a niche set of followers, but it's never gonna become huge.

Mm-hmm.

and that's that's fine. the

I think it's a better experience, right? To be able to just be in a bunch of individual communities with like-minded people. I think that's a better model as a user for social media. It doesn't necessarily have to be that you just have one app where you communicate with everybody. but the problem is, especially for us, because we're we're builders, right? We want to not only interact with people who are like-minded, but we want to get the word out about the stuff that we're doing. And you kind of need that network effect, the big, you know, the Twitter

Yeah.

and the, you know, things like that. So

There are all the economic forces that drive against what, socially would be a healthier experience, which is just small niche communities where you get to hang out with like minded people and, share cool things.

Yeah, yeah. So d so did so how did the whole thing impact your your building? 'Cause y you know, like I know you you're building a ton of stuff. You have a deadline in a week. Is it a week? It's I think it's next week Friday.

I do. I do have

a deadline in a week and I got so screwed last night.

Big launch.

my god, I got screwed last night.

What happened?

solo crashed. I had a bunch of work happening and just after I went to bed, solo crashed. And I have this outside agent that monitors things and is supposed to just bring things back up for me. And for some reason Claude decided

no, restoring these sessions is your call, not mine. So I woke up this morning to you're the one who has to make the call to actually restore the stuff. I'm like, what the hell do you mean I'm the one that has make that call? So it's still this fight with with the AI to recognize that like what you bring to me is stuff that there is more than one right option. You know, when there's only one path forward, do the take that one path forward. And in this case, the one path forward is resume the work.

It's not like you can either resume the work or change course or, you know, it's it's anything that's binary and there's a right answer and wrong answer,

Mm-hmm.

do the right answer. And it's it's just so hard to get you know, a stupid, you know, math problem to do that consistently.

So y so your workflow is your workflow every day I kick it off and in the morning something's done?

yeah. So I don't

Is that like satisfactory?

Like do you feel is that fun?

so what's fun about it is that I spend the the time that I actually spend working with AI. I'm doing planning. I'm I'm reviewing. So like I don't know if we've we've talked about this before. I build everything at least twice. The first thing I do

Yeah, yeah.

is I have like a quick 10 minute session where I say, you know, this is a thing I want to try. So I want you to build it, just push directly to main, no tests, no, no nothing.

Mm-hmm, mm-hmm.

Get it up on the Laravel Cloud, give me a link. I wanna click around and play with it. And I'll play with it and I'll give it some feedback and you know.

Ask questions. So I use that as the jumping off point for an idea. We talk through it and then we talk through any of the architectural stuff, any of the hard decisions that have to be made. And once I'm really, really happy that we have a locked in requirements document, I then just kick it off and it just builds. And it the whole thing is is really automated so that all the tests, the code quality stuff, you know, there's architectural validation, user stories are validated.

all of it just kind happens behind the scenes and I don't look at it. So yeah.

Mm-hmm. Does it have a lot

of your like this is how I like my actions or whatever in in it?

Some, but I've gotten to the point where I don't even fight that anymore. You know, to some extent I I think that if you try to impose your will too much with these agents, then you end up degrading the quality. There's just a way that they wanna do it and you know it's

Yeah, but but you know what

I also feel? I also feel like it if you set up the initial conditions, so

A simple example, right? It's making a controller and it's making it kinda weird. And if that's like in your first hour of like setting up the basics, then like three months later it's still gonna be making all these really weird controllers. And maybe it doesn't matter, but they're

Yeah, well I think part of it

all it's all weird, you know?

Yeah, I'm it's not so much that I get weird controllers or anything like that. What what I will get sometimes is directory structure decisions that are a little bit strange.

Yeah.

and you know, those are the thing is when I catch those, they're really easy refactors. Like, you know, why why is this stuff in this folder? Why is this folder called this? Let's change the name to that, whatever. Like that's that stuff's all really easy to fix. I don't have problems with the actual

Yeah.

code quality. I don't they don't build like super complex controllers or like that. They know that I like the action pattern, so.

they'll they'll resort to that. They

Yeah, yeah. It

know that anytime there's something that's, could benefit from event sourcing that we use verbs, and these are my standards for how I like to use verbs. And you know, I I use UUID versus versus a snowflake and this is why, and like all of that stuff is is recorded in its memory and it does comply with that. And then I have a separate thing that goes through and it checks all those things. So like the I so I guess my answer is yes, I do have my preferences. Yeah,

So you do have a list of coding s standards or preferences.

but it's not

It's not super comprehensive. It's the things I actually care

Yeah.

about. And like 90% of stuff that it could do, I don't really care that much about. As long as it's safe, secure, you know, well written, not super complex code, you know, whatever decisions

Yeah, yeah, yeah.

it makes are probably fine. There's a small handful of things I really actually do care about because are things that I know that if they're not done this way could bite us later. And those are actually verified. So I try to I try to keep as loose a grip on it as I possibly can.

Because I think it does better work if you don't grip it too tightly. And the the great thing about Laravel is that it's such a well-documented, well-established, how to do things that it generally will just do the Laravel standard thing by default, which is great. And then to the extent that

Mm-hmm.

I want to deviate from that a little bit for my own use, it does that. And really the only variations from the Laravel standard that I have are that,

The the architecture of these these apps that I'm creating is a little weird in that all of the authentication and all those views are kept in a package. and so you basically have like a blank Laravel app with no concept that the app itself has no concept of a user. It the user doesn't exist. That's all in the package. And the actual application is just solving whatever the core problem is. and it took some iterations to get that right, but now that I have a right and the system knows exactly what it is, like.

when I tell

Yeah.

it to create a new product, it's gonna use this this framework on top of Laravel and it's gonna do a great job. So

Yeah, it's

funny, I was I was spinning up a thing yesterday. And that and the things that I was I was trying to get right is like I did get a little annoyed at how it started to do the action, so I'm like, no, no, no, fix that. And then and then I wanted to name things clearly at the beginning, because else it'll start it it just like makes makes a model and it just names things very generically.

And then and then that is like it's like I like things named very specifically and clearly 'cause then it's also easier for the model to understand what that is. don't say URL, say what is that URL f you know what I mean? stuff stuff like that.

Yeah. I

I think the reason that I get around that and don't run into that very much is because that domain language is established during my planning process. So there's no I don't have I

Right.

don't ever have to explicitly tell it you're gonna call this model this thing, but as we've gone back and forth on the plan, that's what I always call it. So that's what it's gonna call it in the model.

Yeah, yeah, yeah. So you do pay attention to to the naming. 'Cause I feel like naming is such a key thing, initially, what when you set it up. what do we call this? yeah, yeah.

Yeah. Yeah. And I've run I've run

into that too. And what often when I run into naming problems is because and this is part of the reason I've started just building everything twice. you know, I get through that first iteration and I discover, you know, actually if we did this slightly differently, we could, you know, open up this whole new thing that this could do. Or, you know, we do this much more efficiently if we did it s you know this other way. So I kind of discover those.

Holes my plan by building it once and I don't care what anything is called or how anything is working. but it's during that that second planning iteration when I'm actually getting serious about we're going to turn this into a product, that I actually set all those naming conventions, but I don't set them by saying, and call this model this and call this model this. I just do it by, you know, in the conversation I'm having with the AI, I'm

constantly and consistently referring to a specific thing within the domain by a name and then when the code gets written the second time it ends up using that name.

Yeah, yeah, yeah. Cool. So okay. So is the launch on track?

I it is. you know, theoretically I could just launch tomorrow. Like there's nothing

Yeah.

tech there's nothing technically holding it back. I do want to do some some UI cleanup. it's just a little bit it's it's like one too many clicks to get some places and there are some places where the primary call to action is a tiny button and the the secondary call to action, which was like the thing that I cared about early in the development phase, is still the big button. So

All right. Yeah, yeah. Yeah. Yeah.

Just that thing,

you know, getting getting those sorts of things smooth out and and you there's a little bit of documentation stuff that I want to do. And obviously I'm I'm redoing a lot of the front page and because it's grown way beyond just the initial vision was I'm gonna create, you know, fifteen little open source SASs and throw it up here and you

Yeah.

can install them on one click. Now there's the the lab, the educational piece, the guides, the

I'm bringing in third parties. you know, I have simple status we talked about last week. That's that's up now on the site, so you can install it. and and now I have this this new thing that I started creating last night that is a scalpel set, but just like anything else from scalpels, but it will need to work if scalpels is down, which is a consideration that I hadn't had before. So back to the PHP problem, you know.

I said the reason I didn't take down any of my apps is because I didn't have that static, coming back soon page. Well, now I do. So I have this this thing that you can

Okay.

you can deploy to Laravel Cloud and it gives you a Rust runtime and a go runtime environment. And both are just serving basically a static page. so that if my reason for doing two environments in the same app is because if it's Rust that's vulnerable.

They but they support

yeah, but they support so it's not just a a static page, right? 'Cause 'cause they do support static pages too. It like it does something.

No, it's it's well it's they do support static pages,

but the reason that the reason that I went this way was because what I want you to be able to do is tell your agent, PHP is vulnerable, it's gonna be patched in a few hours. I need to take down all my PHP sites, please do that with this message, and then you give it the message, hit enter, and then it's gonna use the MCP and it's gonna turn off all of your PHP sites, it's gonna point the domain at this either Rust or Go, whichever.

And the reason I did both is because maybe it's gonna be Rust that's under attack and you have to shut down all your Rust sites, or maybe the Go that's under

Mm-hmm. Mm-hmm.

attack, you have shut down your Go sites.

Okay.

I feel it's very unlikely that you'd have to shut down both your Rust and your Go on the same day. And if you do, then just power down Laral Cloud until they figure it out. Because like at some point, there are just some problems that are too big. There's actually a third option. Yeah,

Yeah, yeah. At at some point the internet goes down. Yeah, yeah.

there there is a third option that this supports, and that is a Cloudflare worker that is serving the page.

But anyway, it's basically just an MCP powered kill switch so that if I were to run into this problem again, I could just say shut off all of my PHP stuff. And you know, the and via the MCP, it would do this work. And then when everything's patched, I would say, okay, turn the PHP stuff back on. And it would do the the DNS changes. And it's not actually reaching the cloud filler for DNS, it's just changing the network settings inside of cloud. So the

Okay.

If you're if you're using either the if you're using the Rust and Go solution, then everything happens inside cloud. We don't need a a Cloudflare API key. But as I was planning this out, it occurred to me that, there might come a day when I wanna isolate scalpels. I just want to take down scalpels because whatever's out in the wild is so bad that I just don't want

Yeah.

it up until it's patched. In that instance, then obviously like you wouldn't have a dashboard, you wouldn't have an MCP in order to do any of this stuff. So I

Have built my own version of so that the this one Rust go thing that I told you about, that's gonna be in Scalpels generally available. Anybody can set up their shit pages and have them ready to go at a moment's notice. And they're fully branded. Like you you just tell your agent, make this look like a a coming back soon page for this site. And if you do all that work in advance, then you just have this thing ready to go. so that you can just shut down and turn back on with your agent very easily.

Building my own though for Scalpels that will actually provide documentation for how to deal with all of this stuff. basically how to have your agent do it, and a really scaled back MCP that will still return all of the tool requests from the main scalpel MCP, but almost all the time it will say, you know, sorry, Scalpels is currently down because of a PHP vulnerability. We're working to patch. But this

one particular thing, the one that would allow you to shut off and turn back on your sites, is going to be actually available through an alternate MCP that is it's Go powered. So basically I just have a separate like a backup runtime. So it

Yeah yeah, yeah, yeah.

can be like a there'd be a Go version of scalpels, which is just info only and then that one basic MCP that will allow that that critical feature to continue working.

even if I have to pull the plug on scalpels for something like this coming up. And you know, part of this is just, you know, intellectual exploration. Like how far can I push this? You know, because this is like how I enjoy working. I like to try to solve really gnarly hard problems in a really clever way. And if I can only lean on Larva Cloud, that's a great way to do it because you know, I like to keep my footprint as small as possible. And it's probably overkill. I will probably never need this, but I'm gonna have it. And

And and it was fun and and

And I'm learning a lot along the way.

and not that hard to do, right? I'm I'm doing yeah, I get it. I I feel it's it's the age of doing stuff like that. Okay, so so you're launching you're launching next week, right? With with with the set of apps that is there now, right? With

Yes. there are a couple others

that will be in there as well. But yeah. For the most part,

Which one?

that's there will be a couple others. There's something that will let you take sc

Okay. Is the analytics app done?

simple stats is up. the third party analytics tool is up. Yeah. I

simple sets, okay. Yeah, yeah.

I haven't done my own if you wanna put yours up, the one the one that you created, we could talk about that. I'd be happy to

It i

I would have to I would have to review it a little bit, you know, I don't think I'd feel comfortable just just throwing it out there. Although it it works.

I think

It it's very good. You know, it's it's g

Yeah. I was kind of holding

off on doing analytics until because my lessons from Hone and you know, so the the really heavy ingest ones that I currently have leaning on PHP.

Yeah.

I think if I'm gonna do an analytics product, I wanna just do it in Rust. And now that Rust is generally available, I might kick that off. and then I think I'm probably gonna redo Hone

You sh you should do it in Rust and

in Rust.

And and there's another thing, right? there's this whole because the problem is also like, okay, sure, I can run my own analytics, I can run my own home, but now I'm running a few more boxes on Larvel Cloud. Which which is

Mm-hmm.

fine, you know, and makes sense. But but it's it's interesting because like you're gonna basically have to reinvent analytics and then you have to be like, okay.

When we ingest stuff, do we just kinda like where does it go? Does it just get dumped into cash for now and then it gets saved to the thing? And what if it's high volume and all that stuff, right? But and you know

Yeah. Well, and this

is this is again why I build things twice. So what I'll do is I'm gonna basically

Yeah.

say, you know, I might choose Fathom for instance, say, like let's let's create as close to Fathom Analytics as we can and you know, just vibe it. Like don't ask me any questions to throw it up and then then once it's up, then I'll try, you know, I'll throw some real volume at it and just see, you know, how's this gonna hold up for

Mm-hmm.

my busiest clients. And the thing is, I'm not trying to replace a Fathom Analytics for a big company. I'm not trying to replace Nightwatch for a big company. I'm trying to put in something that, you know, when you're a brand new startup and you want to have this information, but you don't have the budget to pay for it, you have these free tools that you can just kind of plug in, just paying for a little bit of compute and a little bit storage and

you're not paying for events, you're not paying for seats, you're not paying any of that stuff that you would have to pay for somewhere else. You don't have to pay for it here. All you ever have to pay for is just your resources. And

And you don't have the limits.

and I fully expect that you will someday outgrow it. And it'll just get to the point where, you know, my analytics tool is just not robust enough for you. And then, you know, go to Fathom, become one of their big customers because they're

Yeah.

they think about this stuff all day long and they're they're gonna have a better tool.

Every one of these is gonna have a better alternative. but you know, at this certain level of business, especially like just kind of getting started with a with a new project, how many people start new projects and they don't have any analytics or they don't have any telemetry

You know what I also yeah.

Yeah, yeah. What I what I'm starting to like also is just this idea of of you have your stack, so so you know, like at some it's like the stuff you build, right? Like at some point if you're a builder, either you're you have a good business and you have something that that is working really well and you just focus on that.

Or you're gonna be building a bunch of stuff. And if you build a bunch of stuff, it's nice to have a stack and just now if I start a project, I'm like connected to hone, connected to Probable, connected to to all my things in my stack, right? And I know exactly how it all works and it's all and it's like it I c I I haven't actually turned that into a skill yet, but I should. Like you know, like it's it's just just a markdown file. And your whole thing is set up and that doesn't

really doesn't cost you anything extra. You don't have to sign up for a bunch of things. Don't have to figure out like which email should I sign up with and where do I put my credit card and it's just easy and then

Yep. Yeah, and

sorry, I'll say one more thing. And then and then what I'm trying to figure is out is as I figure out my stack that I like, and some of those are paid products, right? A lot of them are just stuff that that is now vibe coded for ourselves. how can I make

Also the workflows in that stack work well together. So for example, my security workflow or even my marketing workflow. How can I make it so that my marketing thing knows it can create things and post things and figure out which one is working and y you see what I'm saying, right?

Yeah. Yeah. I I think

That's what I'm starting

to try to figure out.

Right. And I think where you'll probably land, I where most people are landing, is that you don't want to have to jump into a bunch of different dashboards. You don't want to have to try to connect a whole bunch of different things. and the the AI tools are getting good enough now that you don't really have to. If everything has either an API or an C P or a C L I or some way that you can get your data in and hand it to an agent, you can have agents kind of remix things however however they want.

Yeah.

So I have this I don't know if you've been into

If you see it on the Scalpel's website, I have this product called Capstan. And it's very experimental. I'm the only person using it right now. I haven't even rolled this phone out to clients.

Capstan

I have not I have not seen Capstan. Is that is that the pri

Yeah, so I've even

the yeah, what is that?

So this is my AI

it says experimental.

is very experimental. So it the the first thing I did with it was you remember when there was that thing that that came out a couple of weeks maybe a couple of months ago, maybe now, where Google had been indexing shared clawed artifacts.

Yeah. Last year, yeah.

So when you when you share a private

artifact within within your organization, all of a sudden Google is picking that up and indexing it. So now your private information is on Google. so I was just like, you know what? I don't think that I want to have client reports, things like that being put up. Because the thing is with with the way that the cloud artifacts are set up, if I want to share it with somebody, I have to basically it kind of just becomes public and a hard to guess URL,

Make it public, yeah, yeah, yeah.

which means theoretically I could get indexed. So I

just created something that uses

a more sensible share system where yes, you can create a share URL where anybody who has that link can see it, but you can also explicitly lock it down to certain people. And so I wanted that ability so that I could have really granular control over who can and cannot see these things. So

Mm-hmm.

it was basically just a replacement for the whole artifacts concept that Claude and OpenAI offer. That you just basically tell

Yeah.

it to use Capstan instead of

its own native artifacts. And it does the same thing. It just builds an HTML artifact and it puts it up and we serve it securely and you know, it works great. I have now it it will ship a Go binary to so there's a client, it's a Go binary that I I just decided I'm gonna make it depend on solo. I was trying to make it kind of agnostic so it doesn't really matter what you were doing on your side.

But solo just offers so many things that I don't have to build that now you have to have solo running on each one of these drain machines. But basically now I can create a a PRD, throw it up into Capstan, read it, you know, interact with my agent to get it refined the way that I want it. And then I can just say, you know, hand this off to Len. And then Len will get a notification saying that this PRD has been handed to you. And Len can then kick it off in his solo agents and

So yeah, it's that's kind of where I'm moving my entire workflow is into Capstan. I don't even remember how it got on that. but that's so that's the big thing I was trying to get built out last night that that crapped out is

Mm-hmm.

kind of the the last few pieces of that. yeah, workflow.

So it's it's basically

yeah, yeah, yeah, workflow and

It's yeah you're gonna run out of names, man, like

No, no, I know. I'm gonna run it. Well, it's a good thing I have AI, you can come up with names and I can switch languages too.

Yeah, yeah, yeah, yeah.

I'll just I'll just switch this well, I can't name things German because then, the the boxes would have to get a lot bigger. but I can switch to Spanish names. That would be easy enough.

Uh-huh.

So but yeah, well the reason we got on this is

There's a whole bunch of new new

apps that I hadn't seen.

Yeah, then there are there are more to come. But the what brought me out of this was you were talking about like trying to get your whole workflow put together. And that's kind of what I'm going

Yeah.

for with Capstan, is that you would then be able to you know have all of the data from these various sources get pushed up into this one centralized place that you can act on. Like I if I wanted a dashboard that showed me you know my kind of a side-by-side graph GitHub.

commits and hone data and like just take data from various sources. you know, I can just easily have Claude create that dashboard for me and I can set up a schedule so that it updates every 30 minutes. Like I don't really need real time for a lot of this stuff. If you don't need real time updates and if you don't if you are okay with kind of read only type stuff, you can create any sort of dashboard review or anything like that that you want in this. And that's I think that's gonna be pretty powerful.

You see, Cloud has Claude docs now. So the so they were they went into the into the office suite, which which fair enough, they should, right? Because everybody's like, where do I store my Cloud stuff in Notion? That's what it seems

Yeah. I just I haven't played with it, but

I but I wonder if it still has the same limitations that artifacts have. Is it like it's either private to me or it's public to the world? and I'm sure like if I'm gonna pay the extra money to have a Teams or whatever subscription, then you know, I probably would be able to limit it to just within my team. but

then you're locked into. the the the other thing that I'm trying to avoid is vendor lock in. And we've talked about this before where I really am enjoying cloud right now. Opus five point five has been really good, but it is the best today. It wasn't the best three months ago. It might not be the best

Mm-hmm.

in three months. And I want to always be able to reach for whatever works best for me. And I want to be able to continue working when a provider is down. And so I'm trying to avoid that

Yeah, yeah. No, I I

vendor lock in.

yeah, yeah, I agree. let me I have a question for you. So zooming out, right? so I think we are both getting very like you know, and everybody like there's a certain set of people who have who are lucky to have the time to do stuff, you know, and the inclination and the skills. And we're getting very good at building the most incredible stuff.

and at the same time the models keep getting better and better and cheaper and faster and better and cheaper, right? Like all the time.

And for me, if I look back like if I say a year ago, my product ideas were not ready for for October twenty twenty six, you know? And I find it very hard to say today my you know, product ideas are gonna be ready for October twenty twenty seven. I don't even know. And but so my question is I think we agree that the models will continue to get much, much better, faster, etcetera.

smarter. So sure, we have now developed the ability to build incredible stuff, I'll I'll build a search engine in Rust, why not, right? Like whatever. but but what does that look like a year from now and how does that turn into money?

Yeah, and that's the the tricky thing. And I think that part of my decision to make

Or happiness. Let let's ask me differently.

How does that turn into happiness? Money being part of that, right?

Well, I don't know, I don't know that it'll ever turn into happiness.

well, happiness I'm already achieving because I get to experiment and find, you know, just kind of push and find where the edges are of things. And I like that. I like to be

Okay.

able to kind of, you know, I like to be able to push things until I fail and see, okay, I can go this far today. I'm gonna try this again in six months and see if I can push it even further. I'll bet I can

Yeah.

because I'll I'll have learned stuff, the models and tooling will have gotten better. Like, but always just trying to push and find that boundary, and I get to do a lot of that.

and this kind of what's unsatisfying about client work is that for the most part, clients just want to be able to, have a button click happen a little faster and stuff like that, which is fine. It pays the bills, but it's not that's not what excites me and gets me up in the morning.

Yeah, yeah. Download an Excel. Uh-huh.

What excites me and gets me up in the morning is that I have this crazy idea that I'm 90% sure it's gonna fail, but what if what if it falls into 10%? What if that what if I do actually get it working? How cool would it be? And there may not be any real financial value in it, but this is kind of where I

This is a big driver of why I decided to make scalpels free, because I feel like it would be really hard for me to create a SaaS today that's going to be able to run ahead of what the models can do. So instead of

Yeah.

trying to get you to pay me to use my software, what I'm hoping I can do is get you to pay me to be part of the the movement, part of the organization, you know, be be one of the the freelancers that's in the network that is getting

recommended to people who are looking for opportunities to confidently own software that they used to rent. and so kind of trying to build it on that community support. Cause I mean, I never have launched anything and this this included and maybe especially where I'm thinking this is going to be some sort of a unicorn I'm able to sell this for, you know, hundreds of millions of dollars.

No, no.

Like that's never it. It's just can I get this to the point where it can pay enough that I can give myself a nice

healthy salary and just wake up every morning and do exactly what I want to be doing and be able to say to the the wider world, I'm not gonna do client work because I don't want to work on making your button faster. I wanna I'm just gonna do this stuff. It's fun, it's fulfilling and it provides

Okay.

real value to the community.

So you are build okay, so a year from now, October twenty twenty twenty seven, in the year one anniversary of Horizon.beer you are building stuff you you're happily building lots of crazy stuff and and just experimenting or building things

And there there is enough activity around that

to somehow to make sure you get paid whatever you need to get paid, right? Like d that's good.

And how do the models getting much smarter fit in that? 'Cause I I always I always

feel like today we have five point five, right? It is great.

But a year ago we had

Sonnet four point six or something. I don't know, it wa the world was quite different a year ago. What I'm saying

Yeah.

is like in a year the world may be quite different and then so

Right. Well that's again, this is my point. The the offering around scalpels is it's useful tools that are, you know, potentially gonna quickly evolve. Like hone may end up being a Rust tool, like exclusively Rust in the next three months when it started off as as

Mm. Yeah. Do it.

you know PHP and Larable, because the models are good enough, Larable Cloud supports the runtime now, like there's nothing actually stopping me from doing it. So I can just kind of constantly improve the way that I'm solving these little problems and

I'm not asking you to come and pay me for software. I'm asking you to come and, pay me for you know, the join

Join my movement. Yeah.

the learning journey, right? You know, like get in and read the code and, you know, get on calls with me. I I'm you know, I'm thinking about how I can kind of turn this into sort of a community type thing where, you know, low barrier to entry,

Okay.

everybody just kind of come in and get your hands dirty and get involved. And you know, I'll just keep it all I'll keep all the balls in the air and just

Take a paycheck for doing that.

Yeah yeah, I just I just wonder myself. I feel like the stuff I'm building now versus the stuff I was building six months ago versus the stuff I was building a year ago is very different. And that's just because the models have gotten better. Not not technologically more complicated, but more like just the world needs different things,

A lot of it comes down, it's

I think you know the models are getting better, but I think the the biggest driver of the improvements that we're seeing are in the the tooling, the techniques, the harnesses. we're all just getting better at harnessing what the models can do. so I think that

Mm-hmm. Yeah, also.

even if even if anthropic were to stop research tomorrow and opus 5.5 is the last thing that they put out, I think that opus 5.5 in a year is gonna be

way better than Opus 5.5 today because there is so much opportunity for just wrapping it in the right harness and the right tooling and improving on that stuff. And I th but and I I really do think that's where we're getting most of value. I think you're you're in a better position than I am though, as far as like your products go, because I, you know, the other hesitation I have around building anything that looks like a SaaS right now is that,

Basically the harn the the the labs are trying to do w everything that we're doing. if you're a Figma, they're gonna bring out a cloud design and try to kill you. If you're a my Google or Microsoft, they're gonna bring out an office suite and try to kill you. and those work their way down the food

Yeah.

chain and try to swallow up as much as they can. And so I don't want to be

Yeah the

in any sort of a business where an anthropic or an open AI or a meta could look at it and be like, we can create a tool that does this. But

Data is a much harder thing, much harder problem to solve. I don't think they're going to go after that anytime soon. So making the the curated well f well shaped data your moat, I think is a great way to go.

Data's interesting.

Another interesting one is being the system of records. if software w

Mm-hmm.

anyway, there's all this stuff about motion, we can talk about that next time.

But yeah, in interesting times.

Yeah. The the downside to the models and the two link any better is that I think that what happened Wednesday is gonna happen a lot. Yeah, so I've been thinking

gonna happen a lot. But I feel it was good.

the whole the whole ecosystem is stronger now. And I don't mind if it happens again in two weeks, you know, something similar because it'll be good. It's these it's like you know, you're getting your webs down, you're getting stronger. It's good.

Yeah. It's funny because I've been, you know, I hadn't heard much from any of my users from our CVEs, you know, in in a long time. And all of a sudden I heard from all of this week. And so it kind of put that back on everybody's radar. They're like, I I pulled this up and I saw, you know, what what a couple of people said was, I I patch PHP, but I see there are, you know, like twelve other high criticality C V E's on my server. Like, what do I do about that? I'm like, Well, that's easy. Just

have your MSP write a script that will update those things, but or have your have your agent w write it and you know, throw it up on your server and run the script. It's it's not that hard. but yeah, just the fact that suddenly this thing that I put out and people were you know a little bit excited about for a couple of weeks and then just stop paying attention to all of a sudden they're paying attention to it again. You know, I wonder if we get enough sustained hits like this, if that product will eventually really have legs or

Alternatively, do I just take it and turn it into a scalpel and make RCVEs the the site just be the product value service that just like you know the s the tool is now free, but if you want somebody to actually

It feels it feels to me that's

yeah, I think that's where you should go because the security stuff will like the shocks will keep coming and and probably for at least the years, you know, it's not gonna be like three months and we're done, there's no more security issues, right? But it'll be this this occasional like little spike and and it doesn't feel like you can put the committed attention

Do a separate product if your product is scalpel, so I would yeah, makes sense to me.

Yeah. Yeah. Well, I think too it it then becomes like a good opportunity for Legion for security experts. I don't want to do a ton of security work, but I would love to get a finder's fee from, you know, agencies

Yeah.

or freelancers who are really, you know, tightly bound into that. Or better yet, just have them come in and pay me monthly in scalpels and you could just, you know, eat what you kill. I'll just throw as many referrals at you as I can and you know, so it's all it's all just sort of like

Up in the air and I don't know. And I I don't know. I it's it's weird. Again, I I think I say this every week, discovering the business model as I build. but what I'm excited about is just just expanding the horizons and getting people excited about, you know, what I'm working on and building things that are actually useful.

Yeah, yeah. And we should wrap up soon, but I w I'll ask you one last question. So if a if a builder person or Lauerville engineering person goes to scalpels, right, like what are the three things they should probably pick out just to start?

Yeah, so that's what I'm working on right now. I'm trying to redo the front page. I think obviously

Yeah, 'cause a little circle's

not gonna scale my very well.

No, no, no, no. well, because the offering is is just different. So I think what I'm trying to lean into is empowering people, whether the developers or not, to own confidently some of the software that they rent. So, you know, part of that

Mm-hmm.

is just it's gonna be sort of the knowledge tools and some of it's gonna be the actual tools, some of it's gonna be the

the marketplace, a third-party open source software, being able to install something like Simple Stats where, it you you don't have to be a developer now to just install this. You just click and it's done and it just works. so yeah, I think it's it's really about empowering confident ownership of your software stats.

Yeah, yeah, good. But but what are like for developer people, what are the the three things you think they should install? if you're gonna highlight three. Yeah, yeah.

the three actual tools they should install.

I'll say hone as the first one, it just seems obvious.

I think hone

definitely wake for the logs. So those two are a a no brainer. you know, people have different opinions on like the R R web session recording. I'm not a big fan of it. I don't use it in any of my

Maybe no.

own stuff, but I have clients who have been paying, you know, a hundred dollars a month for hotjar or more forever and being able to just replace that with something that does exactly the same thing. was really

Yeah, yeah.

nice.

So I think that one's definitely optional. But if if you're one of those people who really thinks that, that's a valuable tool, then you should obviously install that. And then I I'm again not super excited myself, but I know I guess I have some clients who will be real really into this. I have this the next tool that I'm building is and it's spread to go. I haven't published it yet. Is when you have a deployment happen.

It will screenshot various pages on your app and sots. And then what you can do is you can have like a a multimodal model basically assess, you know, what changed in your on all these pages and do those changes line up with what's expected to be changed from the code? Like if there's a copy change, does the copy change appear in the code? and then it can flag if there are any unexpected changes. So if you did something

That actually broke layout, you'd be able to, or you know, anything like that, you'd be able to then get notified that, hey, this this last commit broke this layout on mobile.

Yeah. It feels like Claude can just do

that. No? Like ha

It could be

have a look at the site and it just has a look at the site.

Right, yeah. No, it it certainly could do that. The thing is now it has screenshots already there and the screenshots happen

Yeah.

automatically on every deployment. And then you could just set a cron job up that will just do it for you. So that's the other thing with almost I mean, you can say that about basically everything that we're doing here, except for maybe hone. You couldn't really just have Cloud pull telemetry out. You have to actually save that stuff somewhere. And the same with, you know, the logs, you have to have Cloud save that stuff

Yeah. Yeah.

somewhere. But a lot of stuff that people are building is stuff that Claude could do, but you have to sit in front of a keyboard and ask Claude to do it.

And so what I'm doing is what are some important high value things that you can do that I can do the first 80%? And then all you have to do for that last 20% is sit down and say, Claude, create a job that will run every 10 minutes and pull in this data and analyze it.

'cause again, if you make some things like this is all deterministic up to the point where Claude is asked to analyze the screenshots.

Yeah, Claude has a look at it. huh. Mm-hmm.

Yeah. So you're you're saving tokens by not having Claude go and do that work of getting the screenshots and you know that other stuff. So yeah,

Sure. Lots of good stuff.

I mean that's yeah, I don't know. I mean I still don't necessarily know if the developer is my front page user. I'm not sure what my front page user actually is.

my marketing plan is all around the product pages. And that's what I'll actually be like when I s when I turn on advertising next week, I'll be advertising specific products in specific places, targeting specific types of users. and I'm not you know, so I'm not actually gonna try driving traffic to the front page. So it's that sort of like I serve several masters here and I'm not really sure who should get featured on the front page and how can I tell the story in a way that kind of makes sense no matter who you are.

Yeah, I wouldn't I wouldn't

worry too much about the front page. Like it's more like if you can find the channel where people are interested in this stuff and you can find a way to like show them this stuff, right?

Yeah. Yeah. and that's that's just what like especially as I'm have some of this back office stuff out now, the DocuSign alternative. Like there are all kinds of places where people are constantly talking about wishing they could replace DocuSign with something. Reddit is one of those places. So I'm gonna be throwing a Reddit ad up in some channels where, you know, I notice that people are often saying they wish they could replace DocuSign. And

Huh. Yeah, let me know how that goes. I'm curious to see how Reddit ads work or if they work. I guess it all depends.

Yeah. I

I have yeah, we're doing a lot of experiments. So for our B2C app, we I think this week have started, or if we haven't started, we'll be starting next week, a bunch of ads across the meta platform. So Facebook

Yeah.

and Instagram and threads and whatever, for a B2C product. And so that'll be interesting to see what kind of conversions we get from that. but my my long term plan really is just try to get content out.

They're in the places where AI is picking it up so that when, you know, cause in six months, what I want to be happening is when somebody says to Claude or OpenAI, Hey, I I really hate DocuSign, you know, how what else can I do? I wanted to say, well, there's this free open source one that you can get as scalpels, you know, click here. so that's

Yeah, yeah.

so my contest strategy is all around basically just putting information out where the search engines are or where the the

The agents are getting their data that they're using for their n without web search response or even with web search response.

Being Reddit, what else?

well Reddit YouTube. for some reason

Yeah, you too.

they they really like YouTube content.

Yeah, especially

Google yeah, likes YouTube content.

Yeah. And I think I've seen some evidence that podcast transcripts find their way into the training data. So

I I imagine so it's just good content, right?

Yeah. Well, I think it's especially because one of the things that I think the labs are aware of is that there's just so much AI generated data now out on the web that if

Yeah, yeah.

you're just generally scraping the web for content, you're just kind of you start cannibalizing. So you're no

Yeah, yeah, yeah.

longer taking human communication and using that to train your AI. You're taking AI information. So you're going to it's kind of this doom loop that you don't want to get into.

And I think they recognize that AI podcasts are pretty easy to identify. And that the vast majority of podcasts out there, at least any podcasts that have any significant lead listenership, are not AI generated. So if you want to be able to model real human conversation, the thing that you want to grab is like our transcript, because this is how actual people actually talk to each other. And so it

Yeah. Yeah. Yeah.

is a high value target for training AI models.

Yeah, talking about that, Sputnikintelligence dot com It's really good. You get podcast transcripts, also newsletter transcripts, but only the good stuff. connect your MCP. Anyway

Yeah, I'm

gonna go help my kid do his math now.

Yeah,

there's a thing. I I think we can wrap up and we'll we'll talk next week.

All right, sounds good.

Set zero.

Right.