Keeping the Wheels on the Bus

Morning.

Good morning. how are you doing Ed? What's going on in your life?

well, so the kids are back in school, so that's good.

yeah.

Yeah. Alright. So let's talk about stuff. I feel like you have a bunch of stuff. You have stuff. You have stuff. Should we talk about scalpels first?

Mm, a little. I I feel like I underperformed this week.

Where do you want to start?

Yeah, I mean skillful, I just keep getting sidelined. you know I'm I I set this deadline of October thirty first to launch this thing.

Mm-hmm.

You know how it is because you've launched products. You get to that last ten percent and things just keep growing. And some of this I'm bringing in myself, some of it is just, you know, circumstances being what they are. the first thing that has been tripping me up is that our launch plan had been for my partner Len to do a bunch of videos. Because he does he does the this week in PHP internals video. and he's just really well set up for it. He does really well on camera.

It it is his wheelhouse. It's definitely not in my wheelhouse. And then as as he was getting ready to start filming, we inked a deal with with a client that is pretty intense. And again, we're in his wheelhouse. So he has taken the bulk of that work. He's still doing the PHP videos, he is speaking at a conference this week in Texas. So

There was just nowhere on the calendar for him to sit down and do these videos. So when that became apparent a few weeks ago, I thought, okay, I'm seeing some really interesting stuff out of Claude with these, you know, stop motion videos people are doing. Maybe

Mm-hmm.

I can do something like that. And I tried, I don't know, ten, fifteen times. I I would sit down and just crunch away at it for two hours and come up with like it it was just it was juvenile or it was

One of the things I've noticed with all the stop motion videos that Claude does is it tends to make the motions really, really fast. And so

Okay.

it was hard to pace things with with dialogue. anyway, just a lot of failures. And I would just get frustrated, give up, and go on and move move on to something else. And so then finally a couple of days ago, it was time to give it another try. And I moved from my normal setup on the Mac Mini and running everything through solo, and I

Mm-hmm.

just opened up

Claude Desktop on my you know on on my laptop thinking that maybe if I can just iterate faster on this it won't be quite so frustrating. And I I spent a lot longer actually talking about what I wanted before actually trying to generate the first video this time. And I mean probably two hours of just back and forth, you know, just

creating individual snap photos. it I was just having it create pictures of scenes rather than actually trying to animate anything just to make sure I got

All right.

I was really happy with the keyframes of the video.

Yeah, yeah.

and then it was time for lunch. So I said, okay, go ahead and, you know, I've I signed off on the script, go ahead and render. And I went to lunch and I came back and hit play. And my expectation was that, you know, this is gonna be another kind of failed experiment. And I was like, whoa, no wait, this is good. This is

This one works. And I immediately sent it to you. I don't know if you saw the one that I if you watched the one that I sent you on Slack.

I I've seen I feel like I've seen a few versions and they and they looked very good.

But then. Yeah, and then I've I started, yeah,

I've teased a few on Twitter. And yeah, so what so my discoveries were basically, and I don't know if this is gonna apply to everybody, right? But this is what works for me. I spent a lot of time kind of just creating static images for the keyframes, you know, like after

Yeah.

the transition is finished, this is what I actually want on the screen. and then I took a lot of time on.

just various individual assets and those assets then get saved and so they're reused across every video. So one

W

what is the asset like? 'Cause it's like claymation style, so is the asset

Y yeah, it's it's these clay

clay images. So like most of the videos will have a shelf and then things end up being stuck on the shelf as part of the so that shelf is something.

Yeah. But but what's the format

of the asset? Is it is it like a JSON form w what's in the asset?

it's it's

an SVG and it's

okay.

Yeah. And then it just applies some some filter and the filtering was built. So again, it's it's not like the AI is transforming all of these SVGs on the fly. Like it's there's a script that it runs through. It basically creates a base image and then the clay texture is just filters that it's deterministic filters

Okay.

that it runs everything through that gives it the right you know the the right feel on the edge and things like that. And you know, I'm not

Yeah, yeah.

I'm not a graphics guy. this is not something that I understand at a very deep level. Like I I just kind of knew what I wanted and kept talking to it about what it will like make this a little fuzzier and this this should cast more of a shadow and things like that. So it's just

huh, huh.

you know, really just kind of on the aesthetic, this is my taste and might not overlap with anybody else's, but this is what I want to look like. And so then finally I got it working. So like all of the assets are consistent and now I no longer actually have to plan out keyframes. I can just

Point it at a product, or I can point it at an article, like a guide or an experiment or an idea on the site. and I can say, do a TLDR video of this. And it'll read it, it'll propose the script, I'll make a few suggestions on the script. And once I lock the script, all of the rules around what all the transitions are, when a bot is mentioned, which character is that, when this thing is mentioned, which

which characters that and so all of those assets just get pulled in. So everything looks exactly the same. It knows how I want things laid out across the screen at the, on those keyframes. so I don't have to explicitly describe each one. It knows, you know, kind of what real estate I want to be using. so yeah, I got that.

Yeah, yeah.

Now it's it's a great workflow. I can just point it at any any URL on the site and just say create a video around this content and it'll do it. And then I have

Now as of yesterday I've started experimenting with using it to create slides for presentations.

Yeah. Yeah. And and just one last thing though, the the the voice is pretty good too.

So that was that I actually put very little effort into that. What I actually did was I'm using Claude Desktop. I I saved an 11 Labs key to my Mac keychain and I just said

Okay. Yeah.

okay you have access to the 11 Labs the other thing is I I set a budget that so eleven I'm I'm doing it too sorry, it's not eleven labs actually. I use open router.

So that had access to 11 labs and it has access to music models as well because there's music

Okay.

in there. so I saved an open router key, and the open router API lets you set a like a weekly budget for the key. So I set a $50 weekly budget for the key, figuring that you know I might burn through that all in a day when I'm experimenting. But you know, I'd

Yeah, yeah. Yeah, yeah. Mm-hmm.

like on on an ongoing basis, I don't want to spend more than $50 on this thing. And I didn't know actually what it was gonna cost.

but that just felt like a reasonable kind of safe place to put a tripwire so I don't end up with a huge bill. and so I told Claude, you know, okay, now it's time for us to select a voice. I don't have any strong opinions, you know, the language is gonna be English, just grab me a selection from Eleven Labs. And so Claude just built me a little panel and sidebar there, and it I could listen to it it used the the first sentence of one of the the scripts and it

I hit play and I listened to it and then I could click a button to shortlist it. And so I just went through and shortlisted. I came down to three that I really liked. and then I said, okay, do the full script with these three so I can compare like a across a a larger set of sound and landed on the the voice that I liked from Eleven Labs.

Yeah, yeah. Amazing. So so you build an agent unless you wanna dig into this more so you build an agent discussion thing. I build an agent discussion thing after last week, but so why why is yours better, Ed?

I don't know if it is. I mean yours seems kind of more free flow it's

what's funny? It has it has a set of rules that I put some work in and the bots so far they are really following the rules.

Mm-hmm.

as in as in, being good citizens of the discussion board.

Right. Yeah, I I took a different approach because what the the problem that I decided was interesting to solve was you know when I'm doing experiments, often I'll just say, okay, do this thing, put it up on Larabla Cloud, provision everything, blah, blah, blah, blah. So it'll go through and it'll try to provision something. And what

Yeah, yeah.

it'll run into is, I tried to provision this database that appears in the docs, but actually it doesn't appear in the API.

And so I wasn't able to provision this particular database. I had to choose this one instead. So it's finding places where the documentation and the actual API are drifting.

Mm-hmm.

it also will refer frequently to the open API spec, and it'll find places where the open API spec has drifted from the actual API implementation, which is funny. I'm not sure how they're actually doing that. Like I because generally I'll have my spec.

Yeah, you would think

generated from my API

so they always match, but in this case there there have been some times where it's drifted. And then

Yeah.

so what happens then is my agent will leave a note inside of the the meta project that I do all of my work in. And it will the the agents are instructed like before you start interacting with a tool, you know, search for any known gotchas and the workarounds we've identified. And it works fairly well.

but it's it's just for me then, right? So nobody else can take advantage of this information that I've uncovered that, this thing doesn't work exactly as documented. so I just decided that was the use case that was the the most interesting for me. And because this would be used in the middle of a coding session, potentially coding sessions for clients, things like that, I wanted to take a lot of time and really make sure that no proprietary information was gonna make it.

in there. no keys are gonna make it in there, things like that. So I just have a lot of deterministic filtering. I have a lot of trip wires that can just be tripped where you know a report will just get rejected if it suspects that there's anything that is you know proprietary client information. I do, yeah.

Yeah, yeah, yeah. Do do you have chef at all to classify things?

So I have Jev. I mean first there's just deterministic like pattern matching type things.

Yeah, yeah.

if it passes that then it goes to Jev and you know Jev has sort of this

confidence score. the other thing that I won't take is a high severity or critical severity vulnerability report

Fair enough. Yeah, 'cause you're publishing it.

Because I want those to be

responsibly reported to the maintainer. so what happens

Yeah, yeah.

then if if an agent were to try to post about some you know critical severity that it found somewhere, it gets rejected. And the message that comes back to the agent is that this is this looks like it might be a severe security vulnerability. here's how you should report that responsibly to the maintainer.

Mm-hmm.

to try to guide people to do the the right thing. So anyway, I just I I could have done, and part of me is is curious about like what would happen if I created some, you know, just sort of like a social place for bots to chat. You know, people have done that, but think a lot of that stuff is like very manipulated where

Yeah,

people have done that. Yeah.

Yeah.

Yeah, that's less interesting 'cause then it's just, but yeah, I I feel like it's one of those little experiments. Let's let's see how it goes, right? G

Yeah.

g give it a month or two.

Yeah, I just moved all of my well not all, but I I moved the most recent things that I've hit out of my markdown and I've thrown it in here. I'm using it myself and it's you know, it seems useful so far. I did kind of game the system. I you get twenty five reputation points if you get a fix for an issue that's been identified merged. And I made the decision to make that the reward for

For getting a fix merged right after Laravel merged a PR of mine that fixed a bug in the the CLI, the cloud CLI. So I was like, I'm gonna give myself a head start on everybody. But it was a cool experiment to see

Huh. Yeah, yeah, yeah. Yeah, yeah, yeah, yeah, yeah.

can I take a a problem here, identify a PR that's been opened and merged to fix that problem, identify whether that person's a user of the site, and if they are, give them some points. And I can't. So that that just

Yeah, yeah.

works. You don't have to.

It's not like if you were to put up a PR that fixes a problem, you don't have to go in and say, Hey, I I put up this PR that fixes this problem. The system figures out that you put up a PR that fixes the problem, it knows who you are and it will file that it'll add that PR to the the record and when it gets merged you'll get the credit.

Yeah. Yeah.

So

Crazy times.

yeah, so what was the you said you had some rules in in place. Is it is it around like content or just trying to keep them on topic or what's the

it's it's it it's like on topic content style board rules. So it's like these are the things you should talk about, don't be an asshole, etcetera, etc. It's pretty straightforward. But we'll see. I'm just gonna run it for a few months and we'll see what happens. Yeah. And if something happens, I'll fix it, you know?

Yeah. If yeah, if if you notice. I mean that's that's the problem I think sometimes with things like this is that you

I'll notice

because if there's traffic I'll notice that there's traffic. And then I can put

No.

a jet on it and whatever.

No. Yeah. So I I have the JEV on it. And I and so far, I tried a lot of I tried to break it a lot. Well, I didn't try to break it a lot, but I I basically just I spun up a bunch of agents and just said, you know, here's this API, here are the rules, find a way to break them. And every time it did, it would file an issue and then that would that would get corrected.

my god. That's so

funny.

Yeah, well it's great. I I

I love that we can work this way. And I'm I'm sure that this doesn't really necessarily qualify as due diligence in the you know, in the old world, but I mean it's I would never have been able to check things with that volume in the battle days.

Yeah, and that's thoroughly.

Uhhuh. Yeah, yeah.

I feel like we have a we had a bunch of topics and and now if we so talking about traffic just

Mm-hmm.

just very quickly so last night so I used plausible for my anal analytics, right? It's very good. but I do pay for it.

Mm-hmm. Right.

And so last night I'm I'm I spent like I'm gonna say an hour with Claude. I'm like, hey, let's make a plausible clone. So I'm calling it possible because it's pa plausible possible.

And and I just sent it off overnight and it installed the tracker on some of my sites and it works. And it's good and it's it it it does the bot filtering and so I'm gonna run that for a month and then I can compare the actual plausible with my possible. Like the output data. And if it looks good, I'm like, you know, back to like I'm sorry, my friends, but

That that

Yeah, well

that's a thousand dollars a year that I'm paying you,

Yeah, and people like us are bad customers. I I don't know that I'll ever replace Fathom, but who knows, I might and actually interestingly, another topic that we're gonna cover, and we can transition to it, you know, now since we're on the topic of analytics is the marketplace that that I'm

Let's do it.

adding to scalpels. so

sorry,

one one one one comment on the possible thing. The actual reason that made me do this was that they don't have an MCP that works.

yeah.

And I was like I was like, man, fuck it, you know?

Yeah.

Well it's funny because that's like we talked about the first time we we did this, it it's not actually usually about price. It's usually about, you know, you're charging me a bunch of money and I can't

You're getting in my way.

use the tool the way I want to use it.

Yeah, yeah, that is such a good point. You're getting in my way and so, I'm sorry.

Yeah, so while we're on top of analytics, have you ever heard of simple stats?

Okay, so it's a it's a larval app. it's traditionally just been a SAS that you pay for, and what's unique about it versus like a fathom or

cool anal analytics or anything like that is that it's you connect it to your source of revenue as well and so it does revenue attribution as part of

Yeah. Yeah.

the analytics. and it's seems like pretty solid seems pretty popular. the owner decided to open source it and announce

Uh-huh.

that you know he made the the repo public yesterday. And so I

did the thing that I do and tried to put it on Larrible Cloud. And and I was successful. I got it on Larable Cloud.

Uh-huh, uhhuh.

and I think he noticed like 'cause he he saw that I had forked it, then saw a bunch of I wasn't opening PRs to the upstream. I was just opening PRs to my own main branch.

Uh-huh.

but he saw the fork and he clicked over and then kinda commented on Twitter like, hey, are you planning to PR this stuff in? Because it looks good. And so we just had a little kind of back and forth on Twitter. And

I said, hey, so I've got this working on Larval Cloud. I'm building something called Scalpels, which basically just gives, you know, turns anything to like a one-click on Larval Cloud, and I'm considering, you know, opening up a marketplace. And because it's right now it's just all my stuff, but I'm considering opening a marketplace for other products. you know, is that something that you would be in favor of? He's Yeah, do it. So I was like, Okay, I have the green light, I have the first marketplace app, I'm gonna do this.

I I

I I love it. It's this kinda fast reaction, right? wait, we wanna do something, we're all good, and then you do it and then it's done. It's like

Yeah. Yeah, so it's

it's really cool. And so I've managed to do it in a way and part of it is, you know, this one was sort of doing it in easy mode because it's a very well maintained Laravel eleven app. If it were on Laravel thirteen,

Okay.

it would even easier, but you know, very professionally done. They I I think he's partner at an agency. so you know, there this is not something that somebody hacked together. This is a really well designed, well architected product.

Yeah. Yeah.

And so I I put up, I think, eight or nine PRs last night. only a couple of them were fixing like actual bugs. They were mostly just things that just optimized to make it easier to deploy to a large cloud. there

Okay.

were some places where, you know, the it was hard coded, the data was gonna be written to the local drive. And of course the local drive is a femoral on

yeah, yeah, yeah. Yeah.

cloud, I turned that into a configuration thing. So it was all sort of

architectural decisions that were made that were very sensible at the time, that now that Larval Cloud exists, that they just don't make sense anymore. And he was completely open to, all of the changes. So so it was great. And I've I have it set up now so that when I actually flip on the marketplace, I'll be serving a fork that I maintain. But it has not drifted at all from upstream.

Because you're you're you're pushing the the the the PRs back.

Yeah. So once once

he merges I I hadn't checked this morning to see if he's actually merged the PRs, but assuming he merges all those PRs, I can actually take his his exact code and I can put it onto Laravel Cloud.

Yeah, d there's a question that that okay, so have a question about that. So right it's in this new model

'cause I know you have security stuff. What if, you know, your security bot says, Well, there's an issue issue there.

And then fixes it. But now this is your own repo, not the upstream repo. And is the promise 'cause the s the promise of scalpels is a little bit that, here's good apps and they're like a little bit maintained.

Mm-hmm.

So are you gonna have to fork all these projects?

Well from a

Yeah, so the way that Scalpels is written right now, and I don't know if it would be possible for Scalpels to fork from somewhere other than the artisan build organization. So right now it's set up so that it the way it works is it will fork a repo from our organization to the customer's organization and then we'll deploy it to their Laravel Cloud account. and I don't know if I could make it so that it would fork directly from the upstream to so like skip scalpels altogether.

Yeah, yeah.

yeah, I I think one of the things that I definitely want to do with the the marketplace is I'm gonna be pretty picky about it. Like I'm not going to

Yeah.

if because I, you know, I had already worked in this code base. I already had had done all of the things I needed to do to deploy the cloud. And I knew that it was high quality, well written, well thought out, and well it obviously very actively and well maintained. so I felt confident saying, Hey, you I'd like to

I'd like to work together on this. I want to, put this, make this one of our offerings. I what I'm not gonna do.

Yeah, yeah, yeah. that's the whole point, 'cause L if

it's not curated, then why would I use scalpels?

Right. Well, and it's like and that brings me to this the kind of question that's been bouncing around my head since I made this decision and took the action to create the marketplace. Now, one of the things I do want to have on scallopels is some version of Campfire.

Uh-huh.

and it might be two different versions of Campfire because I like the Rust version. and also the PHP version that that I built. So it was kind of like a fresh start on what DHH had done.

nope, throw that all away. Start over. in both cases, it will deviate extremely far from upstream, and there's no way that we could keep the fork connected. So it's going to be disconnected, and it's going to be my

Yeah.

own hard fork that I'm maintaining. And I'll

Yeah.

I'll have my agents keep an eye on the upstream repos to see if they patch if they merge in any patches or anything that that makes sense, and and then I'll have my agents evaluate whether we have the same problem and

And fix that, but there won't be any sync repos or anything like that. Now, does that belong in the marketplace because it is a fork of somebody else's thing, or is that a scalpel's product?

Okay, it's a tricky one, right? So

Okay. This is very interesting. So so I think I think you would shoot yourself in the so so I think you just have to think of this as the iPhone. So on the iPhone I have apps. Some of those are made by Apple. Some of those are made not made made by Apple, right? Whatever. But as a user, if I'm looking for something, do I go to the Scalpel app or the Marketplace? It should just be one thing. There's just a lot of apps. Yeah.

Well it it is just gonna be one it is just gonna be

one thing, but the so with with our internal projects our internal products, the the product page is generated. It's a it's a template, there's no I

It should be the

same thing.

I think 'cause

Yes.

as a user it's gonna be confusing wait there's these things and then there's these other things. It should just be there's a lot of apps, Ed makes sure they're they're they're good. Some of them are made by Ed, some of them are not made made by Ed. And I can one click install them.

So I should just I should just have the

template account for marketplace. So in in the case of a marketplace app, there would definitely be more information. There's gonna be so in this case,

Yeah.

it's not it's not like he shut down his SaaS, right? He's still running his SaaS. So I wanna have a block, I

Right. So you're doing backlinks.

wanna have a block on there that says, don't wanna self-host, you know, go here, they have a generous free tier. you know, things like that.

Yep, yep, absolutely, absolutely.

And that's stuff that I obviously don't have in the standard template, but you maybe you write.

Yeah, but then it's just

versions of the same thing, right?

Yeah, maybe

I just I just add some more conditional logic to the template and if then I can I can add that sort of thing. And I might find use for that sort of block. I could probably put that in hone today. that you know if you want, you know, robust dashboard and you know and and just point point them at night watch and recommend

Yeah, exactly. And and some of

them are gonna have screenshots, some of them don't have screenshots. That's all variations on the same

Yeah.

one page with one button. But as a user, I I don't have to know I don't have to look in two different places. I just know there's a lot of stuff here, I can pick the ones I want.

And Ed makes sure they're good and and then it I'll be able to tell this one was made by somebody else and this one was made by Ed and it's fine.

Yeah, I just wanna I I

think I think you're right. they would all land in the same directory. They'd all be, you know, if you do a search, they'll all they'll come

Yeah. Yeah.

up in that mixed thing. I just definitely wanna have it so that like the marketplace ones will have a marketplace tag on them. part of it is because

Yeah, for sure.

I think the the reason that somebody would wanna put their app in the marketplace, because I don't wanna always have to be hunting people down and saying, Hey, would you put this in my marketplace? I want people to I want people coming to me. But then

I have some apps I would like to put in the marketplace and then and then and then you

are Apple,

Yeah.

So yeah, that's that's the point. And I'm not sure, how ultimately I monetize this. It might be like in in this particular case, I'm not even gonna pursue that. This is just

Yeah.

I'm I'm doing it because I wanted the experiment. He was, you know, happy to play along and he wrote good code and this is great. So I'm gonna kind make this the the first test case. But I'm thinking that, you know, particularly when you have if you if you're running a SAS and you have a self-host option and I can make that self-host option for you one click, then I can also

give a lot of, you know, marketing space for you to drive people t back to your SaaS. also it

Absolutely.

keeps people in in your support funnel versus in mine, which if you're trying to sell the people, you want them contacting you. So

Yeah, yeah, no, I think I think it's great. You are Apple and then if one day you figure out how you take thirty percent, fine, but you know But but y you the reason I go to scalpers is I know these are like selected items. There's a lot of them and I just wanna I'm like, hey, you know, can I get an analytics? there's three of them.

Yeah. Yeah.

I think that's great. And and and then you have some criteria on who you accept and who you don't accept and and it and you have some bots that are checking stuff,

Yep. Yeah. Well and the nice thing is because we have all of these tools at our disposal now, I can have a giant catalog and and I just have things that are keeping an eye on it for me and picking up any C V E's and picking up any, bug reports or any exceptions that happen, it just all just gets taken care of.

And the value is exactly it's kept secu it's curated in the sense that it's good stuff, it's kept secure, it's all that stuff, right? That so that's value. This is this is like

Mm-hmm.

a curation, a curated marketplace. it's one click install, so that's value. Makes my life

Mm-hmm.

a little easier.

I get do I still get the the one what would be valuable if everything had an MCP and it would all be through the same MCP? Maybe.

Yeah, well

so for our apps, they do. you know, just about everything, there might be a couple that don't currently have an MCP because I didn't really see like how an MCP would necessarily fit the particular thing, like image background removal. Although I guess for that, you

Yeah.

know, letting your agent just grab a a random image from your hard drive, throw it up to the endpoint, have the background removed to get it back. But I think that agents can probably just use on device tools anyway, so that's probably yeah.

Yeah, maybe MCP is like it'll it'll say which ones have MCP and I can filter in the search by MCP.

Right.

But so like for the for these marketplace apps, then if I'm going to do that, either I have to make

I like it.

big additions to the code or I have to, you know, collaborate with the the maintainer. I mean in this case I I would assume that if and I haven't looked, I didn't look to see if there's an MCP in this.

Well, w you c you could just push it back, right, and add one. But you know why I like it all like 'cause this is going in the direction of you know how

It's a little bit like those I don't know if this is this is the right way, but like these marketplaces for like plugins for WordPress, which, you know, had is had its fifteen years of of of success. Now it's a marketplace for like full apps on on on Laravel Cloud. Self hosted. The value is like y your bots keep an eye on the

To make sure the stuff is good and secure.

Right. But so if

if we're thinking about this in terms of a WordPress marketplace type situation, do I do I maybe move away from everything is free? Because I think if one of the draws of a marketplace might be if you can put up an app and then just sell the app. and I take a cut.

I I would wait I would I would say leave that decision for in three or six months when it's

Yeah.

like super popular.

'Cause you only g people are only gonna make money on it once it's popular, right? So

Right.

so whatever. If it works, if this becomes a thing, then then you

Yeah.

can figure that out.

Yeah, there's a certain model simple stats. Obviously they're still running a they're still running a SAS. And so what they what they want

Yeah, yeah.

out of their self hosters is to self host it for a while inside I've outgrown it, I need some support, I'm gonna start paying for it. You know, that's why you you Right.

Exactly. My my business is big enough now, I can't

be bothered to self host this this stuff.

Yeah,

so they're not looking to sell something through me and make a cut because we're really what they're just trying to do is get people exposed to the product. And that's

Yeah, yeah.

that that's where the free tier makes sense. And maybe I limit myself to that type of of partner for now.

I think

and and so this is a good segue into a question I have is like 'cause as you figure out the model, you also figure out what is your what is your metric, right? Like what is the thing that you want to grow? And I imagine it's something like amount of people installing amounts of apps on Larrel Cloud. Maybe.

Yeah, that's that's

That's probably a better

one, but something like that.

Yeah, it's it's weird because I'm I'm making up this whole business model on the fly. I don't really have so

Yes. Week by week. huh.

I'm not really sure what I should be measuring. but in what I what I keep trying to focus on is I'm trying to put out good software and good, you know, educational enlightening content. and kind of empower the people who want to own more of their stack to to do that confidently. you know, and

what shape that takes as far as a business, I guess, can evolve. because my

Yeah, yeah.

my primary goal right now is, you know, it w we're in a time where I have this unique opportunity to create a lot of really good software and get that out for the world to use. And you know, it I guess in some ways I feel like the the problem solves itself at some point. You know, you put enough good into the world and and the opportunity will emerge

I think so. I think so. I think that's fair.

for me to then start

Harvesting that. And if not, the thing, so the I built two two new scalpels products this week. And and

Okay, what what are they?

they're they're boring. the so

That's good.

the reason I built them was I went back to my existing clients. I keep harvesting, I keep like dipping into the same well. Went back to my existing clients and basically had a conversation like, what is it that you wish existed? What would you vibe code tomorrow if you could? and two things.

Mm-hmm. Mm-hmm. that's such a good question.

Two things emerged. One was one of my clients said, you know, we have a bunch of sites, they're spread across lower level cloud and lower level forge. And they're already using the whole MCP. So they have all of their analytics that they're pulling in into their agents, which is great. but when they need to get something out of the logs, they have to remember where is this hosted? I have to go like open it up and get the logs, or you know, tell my agent where to go to find the logs. And it would just be really nice if all of my logs wound up at the same time.

So

I have had this problem with hone. I'm like I'm like, Claude, what's going on with Hone? there's this stuff. So so can you can you fix it? And it's like, I can't find the logs.

Yeah. Yeah. So well and this is this is too a really like you can't have hone monitor hone because if you do then you're gonna wind up in this this recursive loop and it's gonna break everything. but you could have this other product that's just keeping track of hone's logs.

Is it then a just a j a general log product? Logging.

Yeah, it's just a general log product. It's

it is paper trail. If paper trail didn't bother to build the website, it's it you just

Yeah, exactly.

It's pa paper t like any SAS without the website. So and paper

Yeah. So

trail is just because that makes me think of QuickWrit 'cause that's literally built for that use case. It's append only

Right. Well and

I I looked at I looked at that, but it's because that's it's old and it hasn't been updated since twenty twenty four and it's kind of in maintenance mode.

Quickwits?

Yeah.

No.

is it is it maintained? I thought it was in maintenance mode.

I just upgraded it.

Mm.

Let me triple check.

You could be right. I may have missed it. I ended

Uh-huh.

up going with DuckDB.

Yeah.

Which, you know, is it's it's kind of the same shape of things. It and it works. I I mean I got it working well enough that like it will definitely meet my client's needs. you know, there are all kinds of different ways that you could like I I could do click house if if I wanted to you know take more

Yeah, yeah, yeah.

time and

So anyway, yeah, it's it is that shape of product though. And I'm not going to revisit right now. I I think the next time that I rebuild this will be the time that I rebuild hone and that will be when the Rust runtime becomes generally available because just that AJS endpoint.

Yeah, which feels like in

a few weeks, right?

It should be soon, yeah. and I think the

Okay.

the actual, you know, cre getting the next version that has the Rust run runtime shouldn't be hard. it's just gonna be a matter of like for existing users, you know, how you upgrade, because now upgrading is going to involve creating a second environment. Because the way it'll work, right, is I'm not gonna take the whole thing and rewrite it in Rust. I'm just gonna take that ingest endpoint and and do that in Rust. everything else can stay.

As it is, I don't want to rebuild the MCP server in Rust. I'm like I'm perfectly happy with the the larval app. The only thing that I feel like would benefit from Rust is that that ingest endpoint. and then at that point I'm gonna go with some a a different database behind Hone as well. you know, I did Postgres just as

Yeah, makes sense.

sort of a

It was a proof of concept and it worked well enough. And it's, it's handling a lot of traffic for one of my clients has a lot of stuff going into it. And they they have it up on a big beefy, you know, server or you know, cloud instances, but you know, it's holding up pretty well. I've been I've been really impressed. So Postgres isn't as big a bottleneck as I thought it would be. I I kind of did fear that like once this client went ahead and turned it on on all of their stuff,

because I I piloted it on three of their applications and they have forty of them. So

Mm-hmm. Uh-huh.

last week they just went merge on all of the the package PRs, which then just activated it across the whole board. And I'm sitting there staring at Laravel Cloud, going, Okay, when is this gonna crash? And it didn't. I mean,

Uh-huh.

you can see the CPU went up some, but yeah, it's what I what I originally provisioned for them, which is not what what

It's it's so

scalpel's provisions by default. I did make it bigger, but

Yeah, yeah. It's s it just seems so useful. I something I like is the more atomic you can make each of those, and if they if they are combinable, right, and if it all kinda makes

Mm-hmm.

sense, then that's also nice.

Yeah, well that's one of the things that I have on my marketing whiteboard is I I've thought through various use cases for combining them. And so I'm going to have kind of like package landing pages where you'll have like your your self-healing software landing page where it will have things like hone and the session

Right.

replay and the logs and all of that stuff. And you basically install these things and then

Here's a a script that you can download, or maybe maybe create a a a Mac binary that you can download that will go through and use the MCP to gather up all the data. And then you define how you want to fire off the agent, whether it's Cloud P or the solo MCP or whatever. But basically just, give away a free you want to make all of your software self-healing. You know, here's the step-by-step guide. Point your agent to this, have it install all these things,

Yeah, yeah, yeah, yeah, yeah.

and now.

your entire fleet becomes self healing. Every time you throw an exception, it'll raise it an issue and it will kick that off to an agent, it'll fix that issue. every time an N plus one is detected, that'll just be resolved.

I like it a lot.

Mm-hmm.

I like it a lot 'cause 'cause 'cause my my there's a part of my brain that is like I want it, I want it. So you know what I mean? so so my my feeling is I want that. and and I like that it's atomic, they're maintained, they're self hosted, it it just all

Yeah, well the the nice thing is I know I can build this. Yeah. Yep.

makes sense.

I know I can build this because I've already done it for, this client that I that I tell you about that has 40 things. They they have all of this stuff set up and I've set up some rudimentary automations around they use solo, so it was really easy. I could just have everything drain in and use the solo MCP to kick off an analysis with a cheap

model so it'll go through and it'll read the the most recent stuff from home and it'll read the logs and decide is there anything here it is what's happened in the last fifteen minutes that's actionable. And if it is, then I'm gonna kick it up to an Opus and Opus will do a deep dive and then run it through run it through the whole the the life cycle. you know and

Yeah, yeah.

this

This tooling is part of and really I I need to start I need to do some video series on solo and the way I use it because w has had a breakthrough in the last few weeks. And I haven't even mentioned this to the client yet. I'll talk to them about it the next time we meet, but they do a lot of you know, what what we would call vibe coding. You know, people who are not developers in the organization are actually just adding stuff to production in various applications. And that's been a

Okay.

a real source of headaches, as you can imagine, for they pay me to keep the wheels on the bus and they keep, you know, throwing away the lug nuts.

Yeah, but but before

before we dig into that, you gotta keep the wheels on the bus. But also but is the bus moving faster? is it is it good is it good for their business that they are doing this stuff?

so initially

the answer was no to both of those. but I have onboarded them into the way that I work in solo, which is that that meta project that just keeps all the notes and and things like that. but the the key thing that it does is any code that gets written, what they had done initially was they just had like the the r standard like claude app in GitHub that will do like the standard Claude PR review and then

So they would have that done and they tell our agent, handle the review comments. It was like that that first thing that we all did when that became an option. I've turned off, I don't have any code reviews happening in GitHub now. I use so

Yeah.

many GitHub action minutes that anything I can do to save on that, I do. So the first thing I turned off was those claw things. But what happens now is any work that gets done inside solo, there's an implementer, there is a code quality reviewer, there's a security reviewer, there's like

several review rounds that it goes through. but it all just happens out of sight. And so it used

Locally.

to be that they would kind of feel like blocked up and in a hurt. They didn't want to to keep iterating with the agent. Like once something was working, they wanted to push it up and just get it out the door.

Mm-hmm.

Because it's the thing that's it's it's the thing that's right in front of them, right? It's it's taking their attention. They're they're focusing on this feature right now. And until it actually gets up and merged that

it it's the thing that's on their mind. But now with the way that I have them working in solo, they're working the way that I work, which is basically I I get through defining all the requirements and I I don't have to give any architectural guidance or anything like that anymore with anything that I do because all of that is locked into my documentation. So in inside this this meta project, all of my guides are

all my guidance for how I want software written and what the standard of success is from an architectural standpoint, that's all just

Uh-huh.

that's all encoded in there. And then there's deterministic stuff, PHP stand and rector and all the other tools that are keeping things on the brails there. So anyway, long story short, they're able to just define requirements. They're be able to just talk about, you know, here's the problem I want to solve. Let's talk through solutions. And once they're comfortable with the the solution,

And the the trick that I taught them I got from somebody

Let me ask you a question.

on Twitter is is have the agent before you kick anything off, have the agent describe back to you in his own words the problem is it's solving and the solution you've settled on and make sure that you're in agreement. And

Yes, huh.

so now so they're doing that and they say, Okay, go. Now they move on, right? So they're they're doing something else with their day. And so they're not

The human, right? Yeah, yeah.

the human is doing something else for their day and this stuff is crunching away in the background. It might take eight hours or ten hours or whatever.

But the stuff is getting reviewed, is going through all this deterministic tooling. There's the whole code quality, the deterministic code quality suite that some of these things take a long time.

Because there are, you know, tons of thousands of tests and you know, cause some of these are really old applications. But all of that slow stuff now is happening completely out of sight. So they're not having to take any shortcuts. They're able to really just take the time to think through: here's the problem that I'm encountering. Let's talk about solutions. Yes, I like this solution. Let's implement that. You know, answer a few more questions, get the agent to describe exactly what the plan is, click go. And since they've really gotten it.

into the the swing of doing things this way, the wheels are staying on the bus. They're moving faster. I'm finding that my time, you know, because they they only pay me for a a small chunk of time every week. And it had been, you know, historically that that time was all spent like, okay, you have this that's blown up and this site's completely down. And like I go through and I'm just kind of mushing things back together and getting it back on track. And then like, okay, now I've run out of hours for the week.

But now I'm able to go in and think through, you know, actual real, you know, real stuff. Like I did a deep dive analysis on their larval cloud and their larval forge spend and usage. And was able to give them a really good detailed report of here's where you're over provisioned, here's where you keep running into some memory issues. So maybe you should think about, you know, refactoring this so you can use managed cues or whatever. Like I'm able to do the sort of thing that, you know, will just make all of their stuff better, which is what

they had intended to be paying me for. They didn't really intend to be paying me for, okay, we're gonna spend six days breaking shit and you're gonna come in and spend

Yeah. Yeah.

one day putting it back together.

So yeah, that's the I think that's the value in what I've what I've built for scalpels and from the standpoint of an agency or a freelancer, if you can start using these things to wrap your whole fleet in in this kind of set of cameras that's keeping an eye on everything for you in real time at all times and can respond to it. it can just it keeps everything in a healthy

Yeah.

state so that you can focus on the bigger picture, the strategic. How can I save a bunch of money? How can I

how can I im improve my page speed to increase conversions? the things that we we used to think about before it was, shit, everything's broken.

Yeah, yeah, no, that you see that all makes a lot of sense. And I think just the fact that you have you this is coming out of real work, I feel it's coming together. And and this makes me think of a similar thing I've been trying to find time for to like sketch this out on paper, which is aside from like building stuff and keeping keeping the the bus running and stuff

on the marketing side, so I'm I'm starting to like and I think you're similar, right? You're starting to build these various systems and automate these various bits and pieces and and figure out what works, etcetera. But how does that become a thing that just accelerates and and what are all the different metrics you need and what are like the atomic little things you need and how

Mm-hmm.

do

they'll fit together. You know what I mean? I feel there's a similar system on the marketing side.

There

there probably is, and I'll tell you where I land on that. I think that the reason that that I'm as successful as I am at writing software using AI is because I really, really know how to write software. So I know how to automate the things because I know I already know how to do the thing well. With marketing, I don't know how to do the thing well. So for me to try to automate doing the thing well.

It's like yeah, that's so that's yeah,

Yeah. Maybe not your wheelhouse, huh.

it's not my wheelhouse. So like, yeah, I think that there definitely are opportunities to automate a lot of things that have to be done from a marketing perspective. I don't really you know, that's not my forte. But fortunately it doesn't have to be. we have somebody who is very successful in marketing and she's she's worked with

local businesses and influencers and and whatnot to blow up brands and make them big. and so she's we we have her on the payroll and she's given some some guidance and a lot and like anything that I'm doing now with regard to marketing, she's signed off on. And as we get, you know, in the next week or two, as we get closer to launch, she's gonna move her focus from what she's been focusing on, which is you know, another

product entirely. It's a B2C thing that we're doing a big Christmas launch on that has nothing to do with scalpels. And I've I've basically not touched any of it other than moving it from Forge to Cloud. That that landed on me for some reason. But that's been our focus getting that Christmas campaign running. But once once they actually click go on that ad campaign, you know, her her job then becomes sort of more reactive.

It's like, okay, this channel's working well. Maybe this one's not. you need to rebalance and spend, things like that. But she's not going to she she's not gonna have a bunch of creatives to sign off on and things like that. So she'll shift then over to scalpels and will help me to you know, really get the, you know, stomp on the gas pedal. Which is why I'm kind

Mm-hmm. Mm-hmm.

of you know, feeling the crunch to actually make that October 31st deadline to launch because

That's when, I'm gonna be paying somebody to do marketing, whether I'm ready to market or not.

Yeah. Yeah. I get that. It's interesting because

just the idea of a launch, right?

I just throw stuff out there and I I I you know, do figure out a few ways to promote it and then I kind of try to find out what what's clicking.

but I f I feel

Yeah, I mean

like I've never done an actual launch.

Yeah, and it it's a perfectly valid way to do this sort of thing. up to up to the point where you're like

I'm not saying it's good, It's

you're validating your idea, right? You you know that people are actually interested in it. And I know I know already that people are interested in scalpels because I mean you saw yesterday on Twitter somebody reached out and said, you know, stop making videos and and send me a contract to sign. You know, and I've had other outreach like that, people are like

Yeah, I saw that.

People are like, I wanna pay you. And I'm like, okay, great. I I will I will make that possible for you very soon. so I know that this is gonna this is gonna work, it's gonna make money. you know, I I'm a hundred percent sure that I'm leaving a ton of money on the table by choosing the business model that I am, but I you know, I don't know, I have my reasons. You know, I the the my mission on the in this is again, you know, put out great software and help people ignore their stack and you know.

I I I don't wanna I'm just really trying hard to avoid the subscription thing 'cause that's what I'm trying to help people replace.

Yeah, yeah. Fair enough.

Yeah.

But you

you've been doing some some cold outreach. Yeah.

I'll tell you quickly about let let's wrap up with that because then we're I think we're

out of time. But but so Okay, so I'm running Sputnik Sputnikintelligence.com. It's it's very good. It's

So so so, you know, we pull in podcasts, we pull in newsletters, we give them to your agents. So I got a surprise signed up the other day, like somebody signed up, started paying, and I'm just like I'm asking my cloud, so what's this guy what's this person doing, right? Like 'cause I'm I'm trying to figure out like all the different use cases and why are people using it? 'Cause there's so much stuff you can do with it. And turns out they are doing opposition research, in the US. So it's like election season.

It's the season, yeah.

Apparently also,

I I did then then I dove in a little bit. Apparently also it's always election season, to some degree. So they're doing opposition research, I'm like, okay, I can see a very specific use case for that, which is, you know, elections are coming up, what's my opponent saying on podcasts? Can we use that against them? Right? Essentially. perfect tool for a perfect use case. So I jumped in, I was like, great, I'll make a landing page for that.

and i'll i'll i'll just send out a whole bunch of called emails so i so i i started using lamb lists which seems very good to be quite quite honest so so one of the things that they do and this is interesting for like SAS dying they have something that you cannot replace so they have like all these people using

LAMList, right? And all these domains need to be warmed up for email so it's not spam. Right? So what they do is they basically send emails to each other to each other and reply. And it's this automated system. And I I don't know how secret that is, but but they warm each other up. So you can only do that if you have thousands and thousands of people doing it.

Mm-hmm.

Hence hence they cannot die because they have a real key advantage.

Well, no, you don't need thousands and thousands of people doing it. You need a lot of domains and you need a lot of agents sending email back and forth to each other.

Yeah, but then owned by different people and at different stages. You know what I mean? there's a variety

Yeah.

in there that is just really help looks really healthy. anyway. Anyway, so I got this very interesting use case. I was like, great, let me do the next day I did the landing page. I I set up the campaigns, one for Democrats, one for Republicans.

I wish I could charge the Republicans triple. I kinda should, but whatever. And and and it's very slowly ramping up because of this, we don't wanna be spam. so I'm sending out like ten emails a day, but I'm getting a few replies, no signups yet, but I just I just I liked this s little s I like circles that reinforce themselves. So whenever I see a use case

that I think is very specific and valuable to people that they would sign up for, I can now do the same thing.

It it's funny that you said it was a a surprising sign up and surprising use case. And it's it's an experience I think that all of us who've been doing this for long enough have have had that, you know, you spend six months building something, you know, with a a particular use in mind. And then the first person who signs up uses for something completely different that you never never envisioned at all. And it's like, that's that's really interesting. I should have thought of that.

Yeah yeah yeah. It I it's not the first person that signed up, but yeah. yeah, it's funny. It's it's just

I'm finding I like it like with AI, I really like having these figuring out it's like what you're doing with building with AI, right? You're figuring out these circles and what's all the data you need there. 'Cause then they can go faster and faster, right? And the and the agents can do it. I don't know, baby steps.

Yeah, that's cool. I'm I definitely wanna stay in touch on your cold email experience because you know, that's something that I so I was I was gonna try Reddit. See if we can make this the last thing we talk about. but I I think I told you the last time we talked that you're like I've not been on Reddit in years. So like I have zero reputation on my Reddit account and so I kinda have to start from scratch and just sort of join communities and and be useful.

And so I decided to do that yesterday. And I spent 20 minutes crafting a thoughtful, empathetic reply to somebody who was feeling kind of like they're losing their identity in the age of AI on Reddit.

Wait on t on Reddit? wow.

And so I spent 20 minutes, you know, really making sure that I addressed everything that this person was concerned with and kind of trying to encourage them and and give the kind of my perspective on on the shift in the the the identity

Yeah.

of the developer.

Yeah.

And

immediately the moderator flagged my post as AI and removed it. And I'm just like, this I I was on Reddit years and years and years ago and got off it because it was like, you know, so there's some really good discussions that go on there and really sincere people trying to to share information. And but then the people who actually control the way the conversation happens are just the worst people on earth.

And so that's why I just I got off and I was like, no, I'm not gonna do this anymore. And now I'm in a position where because of the way this is how the the AI models are get getting. I mean, basically, AI is trained on Reddit. So, like, if you want your stuff to be in the world of of AI, you have to have a presence on Reddit. So I'm like, okay, I have to subject myself to this. I'm sure it's better than it was before. And my first experience is 20 minutes of trying to be really sincere and upbeat and positive.

Immediately get flagged as AI. I'm like, I'm done. I'm never going back to never going to run it again. So I'm gonna lean in on

Yeah, you you can't let it get to you, right?

it's just

I know, but I just I try to protect myself from things that really bother me from a an emotional standpoint. And so just not having read it on my

Yeah, yeah. Also you have some other

channels. I feel I don't think you have to do every channel.

No, I don't. And and I do like we talked initially about that street team idea that I have, and I still want to lean in on that. I want to find people who you know, is it you know there are some benefits at scalpels if you're a supporter. So if you're paying me, you get like early access to things, you get a direct line to me for a feature requests and bug reports and things like that. So like I'm I'm incentivizing j it's sort of like that sponsorware model that like any open source maintainer has done. I'm just kind of trying to wrap it up and make it sort of business sized.

but the other thing that I'm I would definitely be open to and I'm hoping that I can find is people who want to be on in that supporter tier and want to have their, you know, their product badge or whatever on the site as as a supporter of scalpels. You don't give me a dime. All you do is you be my people in Reddit. you know, and when you find somebody who is like, hey, is there a

Can somebody, point me toward a self hosted open source alternative docu sign? Just drop a link to our our product in in Scalpels. You know, you don't have to go out and aggressively market, but every anytime you see somebody talking about a problem that they have where we have a solution, just, you know, just make a mention. And I will, you know, give a nice featured spot in the in the supporters page. 'cause I just don't want to be doing that stuff. I'd much rather other people be

Out there.

Yeah.

Yeah. And it's probably better too. that that makes sense. all right. Shall we wrap it up?

Yeah, that's that's about all I have. So

Good chat.

yeah. It's been fun. I will try to get this one up a little faster. Let's see how how well Claude handles the getting everything ready to publish thing for me this week.

Yeah, Claude can do it.

We think Claude can do it. But I I'm I'm hoping that this is gonna be like the the video problem that I had where, I I try and struggle a handful of times and then all of a sudden it clicks and it just works. And

I think so. I think so. Yeah. Yeah, have a good day,

Yeah. That's the plan. All right. Thanks. You two.

see you so