SaaS is dying on its own

Okay, I I think we're live.

think we're live too. Th there was a c I saw a countdown. That was cool.

Yeah. I

assume that means that we're recording now. So how are things?

So how are things? do you wanna intro the podcast first?

sure. Yeah, this is this is Artisan Beer. This is kind of a thing that I've been wanting to get started for I don't know. I think I registered the domain two years ago. it's just casual conversation over a beer or coffee, depending on the time of day and you know, personal preferences. I do have a beer.

Mm. Man, I'm jealous.

gonna l let you know dark dirty secret. It is nine fifty one in the morning and this is an NA beer, but I I wanted to stay on brand for this,

Peter, you and I talk once a week about what we're doing, the kind of things that we're working on, things that we're seeing in the industry.

And then when we were talking earlier this week I said, you know what, we should just record these talks because you know, I think there's some

Yeah.

some good stuff in here that other people would really enjoy. So we decided to try this out as episode one and we'll see where goes.

Yeah, so so we can start. What's going on in your life? I feel like I'm cop

In my life.

I'm copying all these other podcasts, but that's okay. So what's going on in your in your life, Ed?

Yeah, no, that's that's fine.

So in my life, I I'm a single father this week. I mean generally speaking, I'm happily married and I have two young kids. but my wife is on a trip to a conference in Italy. And so

good for her.

yeah, so I'm I'm if I were to turn my camera in any direction, there would be toys and rappers and all kinds of stuff because the children have completely taken over the house.

Are you gonna clean

So

up the day before she gets back?

yeah, definitely gonna She's gonna come back, everything be spotless, the laundry will be folded. I I definitely will look like a competent father, as long as she

Mm-hmm.

doesn't catch an earlier train and tell she's doing that.

Yeah, yeah, yeah. Yeah. Yeah.

How you?

Yeah. So so I'm in Madrid. I don't know, the weather's really nice today and and what's going on here is is the parking is terrible. We have a we have a subterranean parking in the building, but the building is new. And so for some reason it's not finished, so I have to park on the street. And it's it's just terrible. So every morning I go around, I look for an empty spot and all that kind of stuff. That that

But the weather's nice, so it's nice now. It's been super hot all summer. yeah. Yeah.

Yeah,

I have a I have a parking space in the garage and I don't own a car, so I just keep a bike in it. Yeah.

good. Even better. Even better.

And

Yeah, what are we working on? I I kind of know, but tell me w what like 'cause I also feel like it's partially I think there's a lot of interesting stuff there in the sense that I feel one, everything is changing so much in, you know, Wave, right? Th the whole world. And we're all changing our what is our job and what is our career. And also, we're all doing crazy stuff. Like like what are you working on?

Well, we are at the end of October officially launching a SaaS called Scalpels. And what Scalpels does is it just provides sort of unified MCP and a dashboard over a bunch of disparate little tools. So it's kind of hard to explain. And this is this is one of the struggles

Okay. huh.

that I'm having is that it's it's a bunch of little miniature SaaS apps. They're all open source, they're all standalone.

If you wanted to, for instance, replace Laravel Nightwatch, we have a product called Hone. You just install that and you know, throw it up onto Laravel Cloud, throw some environment variables and install a package into whatever app you want to monitor, and it just takes all of that Nightwatch telemetry, throws it to your API, you're not paying per request. there's no dashboard, it's just it's strictly you know agent facing. So

So so your so your mission

you just ask your agent.

so your mission is you're gonna kill SaaS, basically.

No, no, no, no. I'm not trying to kill SaaS. SaaS is kind of dying on its own. I'm just So I'm I'm making a couple of big bets. My my first big bet is on Laravel Cloud. I think that Laravel has done just an amazing job putting this this platform together and it really truly is the easiest way to

deploy and manage a Laravel application. We've moved every single one of our clients as much as possible to Laravel Cloud. And as they add

Same. Yeah.

affordances, you know, we move more things over because it's just so much easier than than running, you know, your own servers, even through Forge. you know, you always have that

keeping the server up to date, handling all the the CVEs, you know, all of that stuff that once you click deploy on a server, you own it, you're responsible for it. You know, Laravel just takes all that stuff away. And, you know, especially in the the age where a CVE turns into a big exploit in the wild, you know, kind of overnight with AI, you know, the the more things that I can hand to experts to manage and not think about it, the happier I am.

I

I was thinking about that. I think that is a big thing, like as in you really want you know, I really want them to handle the security. Like the more they can do and then and then I know they're like connected to Cloudflare, so that's good because I kind of like feel Cloudflare you know they have it going on, like they know what they're doing.

Yeah.

And so and so I feel like okay, between Cloudflare and Laravel Cloud I feel a lot more secure that I want get

act that easily. Right.

Right. Yeah, we just have to

make sure that our robots don't leave any big holes in our applications.

Yeah, yeah, they will. Yeah. You know,

you know. I don't know.

Yeah,

I mean it's you cl you shut one door and no one opens, but it's it's always gonna be a cat and mouse game and it it's especially now that things are moving so fast, it's definitely something that's constantly in everybody's mind. But anytime you can take a piece of your your business and hand it to an expert and pay them good money and know they're gonna they're gonna treat it responsibly and they're gonna take good care of it, it's a win. And this is why it's kind of funny that SaaS is dying.

Because that's kind of been the story forever, right? You you have a problem that you need solved in your business and it's not your core competency. You want somebody else to handle it for you. And so you entrust some other company with with whatever it is.

Yeah,

and you pay thirty bucks a month and it's worth it.

Yeah. Well it's so the the the narrative around why SaaS is dying is interesting because, you know, it on the surface it's like, well, why would I want to pay you thirty dollars a month if I can spend, you know, fifty bucks in tokens one time, build this, and you know, I never have to pay anybody again. Now of course the truth of that is you then have to maintain it, you have to host it, you have to, you know, blah blah blah blah blah. So this is the the scalpel's bet, right? That instead of you vibe coding a replacement for

you know, your Dropbox or your Docu sign or whatever. We already have those things built. And you just click a button, it goes up onto Laravel Cloud. It is updated for you. you know, as long as you don't you know you can

So so you're making it

huh. So so you're doing the vibe coding for me kinda, but properly done.

Yeah, yeah, yeah. I mean obviously the the robots are doing a lot of the work. but

Yeah, yeah. But but well done, obviously. Yeah, yeah.

well done and it's a thoughtful architecture. Part of that is just out of necessity to try to get this SaaS to work the way I want it to work. It's you know, every one of these things has to be shaped exactly the same way. So I've kind of built sort of a a framework on top of Laravel.

Where none of these apps actually have their own authentication in them. It uses Laravel's first party authentication affordances, but they're all wrapped within a package that these all carry. So for instance, issuing tokens or password resets or login or any of that stuff, that's all exactly the same with any one of these apps. And if I ever need to change

Yeah, yeah, yeah, yeah.

that, I just update that package and it it propagates to the whole fleet.

Are you a WorkOS guy,

by the by the way? I'm a total I I'm a convert

I you know, I'm

to WorkOS since since about a year. I'm like it it's great.

I l

I like it, but it feels like it I I yeah. I just don't know if it's a good fit for this. Yeah.

Great. It's just 'cause like it's the same thing, they know what they're doing and it does all the things.

And I'll give you an example, right? So I want an MCP and I just go and WorkOS, I turn on the MCP, and now my my customers they go, they connect to the MCP, they log in the MCP, and it's done. And it's like,

Right.

you know what I mean? Yeah. Anyway.

Yeah.

I just don't know in in this particular context, because basically I'm just handing out SaaS applications to anybody who wants to use them. And I'm trying to keep it opinionated

Yeah, yeah.

in some ways, but then I feel like WorkOS would be a kind of the kind of lock-in that not everybody would necessarily want.

Mm-hmm. Yeah, fair enough.

plus I'd have to do a lot of exploration because there are a lot of different things that that I have to do, of course. Like we have

sort

of a private packagist sort of repla like it's just a hosted status runs on Laravel Cloud. And of course I have to be able to issue tokens and then write those to the your composer alt outjson in order for you to be able to, you know, get these packages. So it's it's sort of ecosystem specific stuff that I'm not sure that something like WorkOS would be a good fit for.

Okay, but

okay, so so scalpels, right? Like so I heard it you can replace Nightwatch.

Yeah. Yeah, you can replace Nightwatch. And

And a bunch of other stuff if you want to.

yeah, that's that's

Mm-hmm.

sort of the point of it. I'm the the thing that I haven't quite figured out is like who the customer is and how I'm pricing it. you know, one of the big challenges is of course all of these things are open source. So in theory nobody ever has to pay me a dime for any of this. the reason that I made this bet, I think I could make more money if I didn't open source it.

But the reason that I

Okay.

made this bet is that what I want to spend

Just

my time doing is solving problems and making great software. I don't want to spend a ton of time supporting the software.

Okay.

And with open source, obviously, you're kind of relying on your users to, you know, at least report bugs in a responsible way. And now it's just so easy to kind of put up a PR. Like if you run into a problem with one of my products, you just put up a PR. If if it makes sense, I'm gonna merge it. And so that it

Yeah, yeah. I should call you Ed. It'll be like

You'll just call me, I know.

Hey, I got a pro I

I installed Hone, it's very good.

I'm I'm glad you like it. well it's just the point you didn't pay me anything, you just you just installed it. So that's the

I did, yeah.

that's kind of the downside to this business model. the the other thing that I'm I'm hoping that I can get out of this is that it just kind of becomes a toolkit for freelancers and small agencies when

They are trying to help their clients to solve problems. Because I know every single one of my clients has come to me and said, I'm so sick of this service, you know, I want to get rid of it. And, you know, I'm gonna vibe code a replacement, or can you vibe code me a replacement? And you know, as a as

I think.

a small agency or as a freelancer, if you say, Yeah, actually I have something in my pocket right now, this is a drop-in replacement for that, so let's just go ahead and do that.

Okay,

so it's scalable, it's a big toolbox of a whole bunch of very focused tools and replacements for lots of SaaS things out there. And if I'm a agency or or something, I can be like, Hey clients, I know you're paying like two hundred bucks a month for this thing. And I have a re replacement that's better, and let me just help you out there. Yeah.

Yeah.

Well and it too, that that turns these freelancers into my kind of premium support tier, right? If if you're a customer and

Yeah, yeah.

you want to use some of these tools, like Hone, for instance, like 'cause we've we've been talking about this, it's this Nightwatch replacement sends your your Nightwatch telemetry to you know, it it sits in Postgres, it gets rolled up, it's really handed off nicely to your agent fee and MCP.

Let let's dig into that. C yeah, can you explain

exactly how it works? 'Cause I I just think I know you've told me, but it's just very interesting and it's not a replacement

Yeah.

for Nightwatch, it's more like a different kind of Nightwatch but then it's also very

Yeah, so

similar.

yeah, so basically it takes the it uses the Nightwatch first party package from Laravel. So it collects exactly the same telemetry that you would send to the Nightwatch SaaS. And instead of sending it to the Nightwatch SaaS, it sends it to your own endpoint that you're because you've deployed this now to Laravel Cloud. So it's sending it to your own endpoint on Laravel Cloud.

And it's being stored in your own database on Laravel Cloud. so nothing is leaving your your infrastructure. All of your data is kind of within your account.

So I just deploy an app on Laravel Cloud, which is one click which is my my Nightwatch backend, my personal Nightwatch backend. Okay.

Mm-hmm. Yeah.

And then you know, Nightwatch has this amazing dashboard where you can like click into queries or click into requests so you can dig down and see, you know, everything you could possibly want to know about that request. And I didn't try to replace that stuff. but honestly, none of my clients use that stuff. What they want is like what queries do I have that are slow? what

What are my routes that are slow? Those are really

Yeah.

actually the two big questions. And then of course, what exceptions am I hitting? And so all of this data just gets rolled up into into kind of a well, we call it a roll-up, but it just optimizes the data that comes in. So I'm not just trying to query raw data every time you every time your MCP hits it. It's not going through and getting all the requests, it's is getting aggregated data.

And so the way that I use it myself, and I don't know how how deeply you've integrated it, but the way that I use it myself is that I have basically a s an MCP aware script. It's not a it's not an agent, right? But it's just it's a a bash script essentially that speaks MCP and it reaches out and it grabs the MCP the data via MCP on a crowd job.

And it hands that data to a cheap agent, the ch a cheap LM, that just basically s looks, is there anything here that looks like it needs to be handled?

If there is, it kicks that off to you know like a Claude or you know a something like that to a higher end model to actually like for instance in the case of an exception, right? It'll take the exception, say, this is an exception, we should handle this. It'll send all that data off to Claude. Claude will do a full investigation, write tests, you know, everything required to handle that exception.

And then hand me a PR. So before I even see that an exception happened, I have a PR op that fixes it. And this just kind of runs in the background all the time and it just sort of helps to heal my applications. If I put up a

Yeah, yeah.

if I put code in production and after this most recent commit, one of my endpoints is suddenly way slower than it was before.

the same script notices that. Like whatever you did in this in this commit here has slowed down this endpoint. we should take a look at that. And it'll just kick off that investigation. And if it's something that can be resolved, you know, again, I'll just get a PR. And I so it's just allowed me to kind of make our fleet of applications self-healing and self-improving. And what's nice about that is that I can I can feel confident about having

as much stuff as we have in production. We have like six different SaaS products that are currently in production with users now. We're gonna have, you know, several more. My my partner Len is working on a bunch of he's focusing heavily on multimedia. and so he has several more that are going to be launching here in the next couple of months. And

We're

a tiny team. So if you know, having this much stuff, getting this much traffic, having this many users touching it, you know, we would need DevOps people. We would need developers constantly monitoring customer feedback and server logs and accepting reports and responding to everything that comes in. And now kind of all that stuff is at least triaged, if not fully managed. And Hone is a big piece of that. I also have

I I don't know if you've played with a session recording tool at all as the gardener's walking around behind me.

I have not, I have

not just just just maybe for the audience. So so this is what happens when we talk. Every time I talk to Ed Ed, he's like, and I've launched two new apps. And

Yeah.

I have three more ideas. And and anyway, so so that's I don't know. But no, I've only used Hone. I j I just use it. I'm I'm not as fancy as you. I just I just ask Claude every now and then, like, hey, have a look at Hone.

fix whatever need you think need needs fixing and it kinda just does it. And it's like, and the cues, you know, there's something weird happening with the cues and this and that. But yeah, because Nightwatch is just lovely, beautiful I could spend all day in it, but then I can't spend all day in it, you know?

No, you don't you don't look at a time. And the there are certainly companies that have people whose job it should be to dig into dig into that sort of data. Yeah.

yeah, if I had a big business. Yeah, yeah, yeah, yeah.

But you know, if a team like ours with three developers or, you know, most of my clients who have, you know, between three and ten developers, they're they're all trying to push up products. And so when they log into something like Nightwatch, they just want the headline, like, what are my exceptions? What are my slow routes? And s you don't need a dashboard for that. and

Yeah, yeah. I I

also f yeah, sorry, go ahead.

Well th and then it's just the price thing, you with with Nightwatch, and this is just out of necessity. You cannot create a SaaS like Nightwatch without having people pay per event because it it just it's so resource intensive. There's just no other pricing model that works.

The problem is, you know, especially in this time when bots are providing so much traffic to to websites, whether you want them or not, small businesses have websites up where there's not necessarily a strong correlation between requests or events that would be triggered in Nightwatch and revenue. So you just very quickly get priced out. And, you know, it does allow

Yeah, yeah.

you to throttle things down. So you only get a small percentage or a fraction of a percent of your requests. But

to some extent

It's

the the clients feel like then they're they have this big blind spots because you know I'm paying all this money for this monitoring tool, but it's actually only getting one fiftieth of a s of a percent of all of my all of my requests. And by the way, I had to turn off queries altogether because you can't it's it doesn't matter if you throttle your requests way down, all of your queries, if you're monitoring queries, get sent as events. And so it's so now you have that big blind spot too where you're

Yeah, yeah, yeah. It's

not able to track your events.

it it's funny because it's so beautiful and it's so you really want to dig in and be like, click and what happened and

Okay.

my god and you can see which user was affected and all that kind of stuff, right? And it's it's just such an incredible product. And it and again, if I was an enterprise, then yes, but

For anybody smaller, it's like and I it's such it's so incredible and I can't and it and I can't maybe quite afford it and it's particularly I don't have time for it. And also now we have agents and let can I just have Claude handle that stuff, you know, like just just

Yeah.

just Claude, just do it, you know. yeah. Yeah, it's it's it's

Yeah.

it w it's so crazy. It just blows my mind that, you know, anyway, this is the world we live in, right?

Yeah,

well and that's kind of the the point of the the thing that I kicked off yesterday that I'm actually most excited about isn't even it's not a new product, although it's going to spawn

it's not even that.

a lot of new products. I don't know if I even told you about this, because I think something that you and I talked about a couple of days ago spawned this idea for me. And it might have just been the the concept that you and I should just record our conversations because there's good stuff in there.

I've started making the roadmap for scalpels public, but I did it in a way where it's structured as a lab. And so what I do is I write up an idea, and then I kick off that idea and I write up the results of the experiment. because all of these, everything for scalpel

And give me give me an example of an idea.

Is an idea an app?

It could be. it could be an it could be an app. It could be so the first one that I did was and sort of copycatted you a little bit on this. I got access to the Ruby runtime on Laravel Cloud. And

yeah, yeah. We should we should talk

about the worst stuff after this for a moment. But yeah.

Yeah, we'll we'll get to that

in a moment. So I got access to the Ruby runtime on Laravel Cloud, early access, and I was told right

Uh-huh. Mm-hmm.

away, like this is very early and not everything's gonna work, and so don't put a production

Mm-hmm.

workload on here. I'm like, that's great, that's that's right at my wheelhouse. I like to break new things. and I took Campfire from thirty seven signals, which is designed to run on a single server and really opinionated way. I mean it's there's

There's just no like with Laravel you can choose your storage bucket. You can choose your database. In this case, your storage is on your local drive. Your database is SQLite. There's there's no path to

I'm sorry, yeah. Storage on your local

drive, who does that? It's it this is this is like this is like you're living in the Middle Ages, you know?

well, I mean dh

does and it saved him a ton of money. So, you know but you

Even so.

know, again, this is this is where, you know, he kind of lives in this bubble where it's like, Yeah, I can save all this money by having racks of my own servers. Like, yes, and you can afford to pay people who can manage racks of your own servers. For you know, for me, there's just no way. First of all,

I would have nowhere to put a server rack. I would have nowhere to put in

Yeah, yeah.

a fat enough pipeline to run that server rack. Like I need

Mm, yeah.

the cloud, you know. I think and most of the internet needs the cloud. Now, the way that Campfire is designed, you could spin up a digital ocean droplet or whatever and you could throw it on there, you know, inside a Docker and it's just gonna work and that's fine. but of course I'm going all in and

Betting big on Laravel Cloud. So I wanted to see: can I make this work on Laravel Cloud? And I knew in advance, when I wrote up the idea, I knew in advance that you know storage was gonna be an issue, I knew the database would be an issue, there'd have to be some refactors there. And what I told my agent to do was to get this deployed on Laravel Cloud using the Ruby runtime.

Mm-hmm.

Mm-hmm.

With as

few changes as possible. So we're going to fork it and keep the diff as small as possible. I knew there was just no way that I would be able to, you know, basically fork it and put it up without having to change something that's going to conflict with upstream. So long term, if this is going to become something that if if campfire for level cloud becomes a thing, it has to be a hard fork that I just maintain.

I don't know if I'm gonna do that, but you know, but I knew that like going in, there was just no way that this was going to be something that would just work out of the box.

Isn't this isn't

this a wild time, by the way? Like continue, but I

Yeah.

just wanna say like the words you you said the words campfire for Laravel Cloud. You know, like I just

Yeah, yeah.

I know I know, you know. Anyway, continue.

So so I get it working. and there were more I was bitten by more things than I expected that I would be, but this nothing actually turned out. I I fully honestly expected that I would hit a hard blocker. And I would get to the point where I would send an email to the Laravel team and say, here's what I tried to do. I couldn't get it done. This is where I was blocked. But

In the end, I actually got it working. It's up. You can chat in it. You can upload pictures. Like it it works. there was some some serious funkiness around you know, it it uses WebSockets for all of its interactivity.

I see.

so it's it's not one of those like when when I write applications that use WebSockets, the WebSockets are a nicety. They they make things feel a little snappier and more real-time, but the apples still work if they're not working.

Yeah, yeah, yeah, yeah. But this

is full on reg

In this case.

web sockets or you or you just have nothing. Yeah.

Yeah. Yeah.

In this case, without WebSockets nothing works. And there was a a problem with it. There was a header not sent. Or there was a header being sent, but it was being stripped out by by either Cloud or Cloudflare somewhere. And so the the WebSockets were working and it does not speak pusher. So I couldn't use reverb. And I thought, okay, well I'll I

I created a branch and just told the agent, all right, make this thing speak reverb and see if we can or speak pusher so we can throw it on reverb. And and it did it. So I was like, this is

Yeah, yeah. S big pusher like everybody else.

yeah, I know. That's another one of those like I don't yeah, I'm not I'm not I

Like weirdo people. But yes. but yeah, it is such a funny time.

I'm certainly no DHH DH fanboy. He's

He's very opinioned. In my opinion, he's often wrong, but he's extremely successful. So

He he seems he seems brilli both Yeah, he seems brilliant,

very successful and an asshole. But but who knows? I d I don't know. I've never spoken to the person, so

There's

yeah, there's there's a lot of that in the industry. So anyway, s basically all of these like it only works this way is kind of fighting against this very strongly held opinion. And this is why I know, for instance, like the i during this exploration I saw that there was a a PR open on Campfire that would allow you to use an S3 compatible bucket.

I ended up sitting there on merged. I that'll never get merged because it's just it it's in violation of his core principle. So that's just that'll be a non-starter.

Yeah, yeah, yeah.

Which is fine. Anyway, I proved it. It it can work. So it was written up as an idea. And then I did the experiment. I wrote up the results of the experiment. And I have this section on the scalpel's website now that's called the lab. And so what I'm doing now is every

day I'm getting

I like that idea

a lot, actually.

Yes.

So now every day I'm getting on and recording and trying to stream if I get streaming to actually work. some problems with the X API. And I don't currently when we set up the Artisan Build YouTube channel,

somehow we messed it up and it's not a brand channel, so we have to do this conversion process to get it turned into a brand channel, which will allow me to stream to it. so at some point I'm gonna be doing these as live sessions, but for right now I'm just kind of recording myself and I'll throw them up on YouTube later. And

Yeah.

basically what I do is I talk through here's this idea I had. and this is what I'm thinking about trying to solve. And and then I kick it off to an agent. And when I do these experiments, and this is how everything in Scalpels initially starts, because I

I don't necessarily know if every problem I'm trying to solve can be solved. I just tell the agent this is throwaway, we're not gonna actually use any of this code, this is just discovery, push straight to main, get it deployed to the target, and success is it meets these criteria. Now for scalpels, the criteria are basically it has to only use things that are provided by Laravel Cloud. I don't wanna have I don't want to force anybody to use any other services, and it has to be able to do meaningful work.

in an environment where all resources that can sleep or scale to zero are do that. So yeah.

Scale to zero is such a

my god. Like they don't talk about that enough. Like the home page of

So no, no, it is

Laravel Cloud just says scale to zero. I don't know, it's just unbelievable.

Yeah, no, I it

that is that is huge. Because especially like a lot of the problems that I'm trying to solve with scalpels, like image background removal is one of those that like every one of my clients is needed at one point or another. And so they're paying these monthly subscriptions, they're paying for credits or whatever, and they're often not using it very much, right? So I didn't want to create something where you provision this thing and it's just there crunching away, costing you money all the time for you to use it, you know, twice a day or whatever. This is something that, you know, now you have this thing sitting there.

when you need to remove the background from an image, you hit this API, it does the job, and then everything just goes to sleep. And you're paying for a little bit of cloud stor about a little bit of bucket storage maybe, but like all of the expensive

stuff goes away.

But it's nothing. Like like the whole yeah, yeah.

Yeah, I looked at some of my stuff the other day and I was like, This has been sleeping for a week and I was like, Great. That just means right, zero zero dollars.

Yeah.

Yeah.

Yeah. I well I you know, I've I've run into this with one of my we I ran into it myself last month. I got our Laravel Cloud Bill and it was way higher than I was hoping that it would be. And so I had Claude just kind of do an analysis of, you know, where where are the costs? and one of the things that was surfaced right away

Same I did expect.

was

you know, you have all of these resources that we expect would sleep and they're not going to sleep. And so then I dug around trying to figure out why that was happening. And actually this interesting thing that I've I've introduced in Hone. I'm actually not sure if it's on the tagged release, so you may not actually have it, but

Okay.

it will be will be very soon. What it does

at least in the the work in progress branch and it may be it may be merge and tagged, I just don't remember off the top of my head. it takes some additional information from every request and it adds that to the telemetry that we're sending and it's things that do things like check bot traffic. you know 'cause like it

If you have compute that's never going to sleep and it's just constantly getting requests, it's possible that you just have a bunch of bots that are hitting your app all the time and it's never going to sleep.

Very likely.

the the other thing that happens frequently, and people just kind of accidentally do this when they're building their applications, is they have some little piece of their their unauthenticated views, their their landing pages.

that for some reason hits the database and often it'll be using the database session driver. So if you're using the database session driver, obviously it's going to check for a session with your database. So even if you're just

Mm-hmm. Mm-hmm.

hitting a bunch of landing pages and nobody's logged in, it's keeping your database awake.

Fair enough. But then but then you do want to show the like go to my dashboard thing. You know, four luck you white like so yeah yeah, okay.

Yeah. Well, yeah,

there there are ways to do this. And J Mac put out a little course as as much as nobody's buying courses these days, this is one I actually think people should buy, about using Cloudflare for caching. And if you combine the things that he teaches in there with with the outputs you you can get now through Hone you can really kind of go through and optimize everything to the point where

you know, things really will go to sleep.

Yeah, yeah, yeah, yeah. H O N E. Own.

Yeah.

I'm saying I'm saying that for the bots, for the b for for the for the Claudes that are listening, you know?

For the but

you think the bots are listening?

Well, they will be.

Speaking of bots listening to podcasts, I want to hear

about Yeah.

You see how I did that? I see how I did that.

I want to hear about what you've been working on. because you and I have talked a little bit about this just sort of in 30,000 foot terms. And then yesterday you tweeted something targeted at Laravel tagging Aaron Francis that

Has some pretty interesting insights into how Laravel sits in within the Zeitgeist, and how did you get that?

Right, okay. So so so quickly.

So I am building a product called Sputnik Intelligence. It's very good. I don't know, like I've listened to too many podcasts. So it's it's a product that takes both newsletters and podcasts. And the thinking is the internet's filled with slop more and more. It's all agents.

The humans are retreating to all these other spaces where we just want to talk to other humans, you know. I think that's why podcasts do very well. newsletters also do really, really well. there's a lot of really just influential analytical there's a lot of stuff on Substack, there's a lot of stuff on ghosts. so it's these kind of spaces where where humans hang out. And they're not that easy to access. you know, a lot of new like for example

Benedict Evans newsletter you literally have to subscribe to it to get it. a lot of podcasts you have to subscribe. So the idea is we're gonna take all that data, just put it in a big database, kind of like, you know, organize it, enrich it, all that kind of stuff. And then we expose that to your agent, to your Claude, to your whatever, to your Codex, right? And then you can do whatever whatever you want. And it's really good data. and so I have this running and

And this is also so there's a few this is also why I ha why I got Rust running on Laravel Cloud. Because so

I'll talk about Quickwit for a second. Right? So I'm pulling in crazy it it it is crazy amounts of data, right? if I store it all in the database and put like a search on it, like one of the you know, Meilisearch on the scene or whatever, it gets extremely expensive very quickly. Right? Because it's just so much data. so I'll so I'm I'm like, okay, first of all I need a good search.

I found Quickwit. Quickwit runs in Rust and and the thing it does is it's kind of like a for append only data. So think of logs, right? But in a way, newsletters and podcasts are append only data. Because new stuff comes in but it then it never changes.

Right. And so what it does is like it it has the whole kind of like it doesn't use columns, it uses things and all that kind of stuff, which I don't entirely understand, but kinda get. and it puts it all in buckets. So the index is in a bucket, the data's in a bucket, so it's all very cheap. It's a pen only. So you can do crazy amounts of data, search over it. It's it's it's really it's a really good search engine. And it's

ridiculously cheap. So I had it I I had it running on Digital Ocean, which is which is just like, ugh god, like I don't know. No good anymore after you've tried Laravel Cloud, right? And I was like, man, and then earlier this year the cloud people started talking about, we can do we can do other things. So I started like tweeting like hey can we get Rust? Can we get Rust? At some point they were like

Yeah, you can get in the Rust ba beta. So I was like, Great. And I j I was just like Claude, let's deploy this to cloud, this Quickwit thing. it all just worked.

kinda one-shotted it and now I so now I have Quickwit Search Indexer, which is built in Rust, running on Laravel Cloud, in just like a separate little app. It's all kind of like close to each other, so everything's fast. And it just kinda works. Anyway, so that was the Rust experience. I I learned a little bit about Rust. It's it's I feel like a lot more people are using Rust.

Yeah. Well two things

Now because it's really fast.

it is. Two things come to mind. One is that I wonder if that might end up being a better ingest engine for things like Hone once it's kind of production ready. because I think that's I

Do you need search over the thing? But it's keyword search, that's what it's built for.

Yeah. I don't know. I just well just the the fact that it would be a columnar database versus or the that other like the

I assume it's kind of like what yeah.

Yeah, the the the thing, the thing, right? Yeah, yeah, yeah.

Yeah. You can tell. We're just developers. We don't know any of this like database stuff. more it's probably structured more like a click house, which is what's what Nightwatch uses behind the scenes.

Yeah, the reason it's so

it's so cheap and fast is that it doesn't need to keep the index in memory and the reason it can do that is that is that b it's append appends only data. That's how it works.

Right. Yeah. Well and something

something like Hone of course is just sending data in it's it's never updating anything. It is it is again just append only data. So the data shape is probably appropriate for it. I don't know that I would necessarily need the the search affordances, but as far as the

Yeah, yeah. And then do you

the data shape, it makes sense. I just you know, I went with what I know, you know, Postgres and you know do roll ups, but would I be able to avoid doing the roll ups if I were to do something more like this? Maybe. I don't know. R Rust isn't

Maybe, maybe because

in

GA yet, so I probably won't experiment with something that's working now, but

But but yeah, but you know, like maybe because one thing it does very well is to say, find me like a trend, as in you know, how often is the are people talking about the straight off hormones, right? So imagine doing that in like in like a database or even a search index. Like your your typical search index is not good at that, but but Quickwit is though, because

And then you just get immediately back and you can do a trend line and stuff. It's really cool. Anyway, yeah.

Yeah. I'm gonna I'll I'll dig into

that a little bit more at some point. But the other thing about Rust that it kinda makes me laugh, and you you know, your your statement about how everybody's getting really interested in Rust now because it's super fast. the day after I finished my experiment

And tiny memory, tiny memory.

Yeah. But the day after I finished getting my experiment done to get Campfire moved over to Laravel Cloud, DHH tweeted that he's now ported it to Rust. So that's the

But it's it's it's but think

about it, right? Like it's very very small memory, like it is the smallest memory load of of any of all these things, right? So you know, anyway, but but I wanna I wanna there's so much to talk about, like 'cause 'cause one side thing and and we can talk about this for many weeks, but one side thing is like I could never have imagined I could do all this stuff.

Right, it's it's just wild. And then the other side side thing is yeah, Laravel is no longer just PHP. But but anyway, I wanna say one more thing about so Sputnik intelligence, Sputnik, right? So sp so Sputnik pulls in all this data, it's great. and then you throw your agent at it. So so I just asked my agent, like, hey, check what Aaron Francis has been

talking about in terms of his goals for like Laravel marketing and then have a go at like all our data in Sputnik and kind of come up with like some thoughts and recommendations and stuff. And so it just it just goes off for 20 minutes, you know. And that and then it came back with a whole bunch of stuff like, hey you know so so if you want to position Laravel as you know the best

way to develop with AI.

Wales is doing this exactly the same thing and it's doing it better than you. Here's what it came up with all these things that I just pulled out of all these podcasts and newsletters. And and I was like, I don't know, Howan, you guys should pay me and and and use this ever every week 'cause it's just really good data and you cannot get that. I'll

can I say

Yes.

one more time?

Yep.

Just very shortly. Aside from the crazy Rust stuff, because it's a Laravel app, right? But then we got the crazy Rust stuff. We also have a bunch of crazy GPU stuff. Because we're like doing all these transcriptions, it's pretty expensive. it's a lot of GPUs, it's a lot of crazy stuff. So I'm just saying I'm saying that because like don't try to do it yourself. Not because it's hard, but because it gets expensive. just just get the data.

Anyway, that's my pitch.

Yeah.

So yeah.

Yeah. No, I mean what I was gonna say is that so it's it sounds like while I'm busy trying to replace SaaS, you're trying to replace comms agencies. so my my wife works in communications.

I am selling to commerce agencies.

Yeah. So my wife works in communications and the the sort of story that you just told is the sort of thing that

she would often sign big fat contracts with agencies to provide and they would provide, you know, maybe a a weekly digest email of some like here's here's how your business is appearing in in the world and

Yeah, yeah, but to be very clear, like I am selling to these kind of things and and it and the thing I'm selling is is not the report or the this or the that. It's the data. Because you can get like podcasts are just RSS, they're very easy to get. But but you don't get the transcripts. And the real thing that the humans are talking about is in the transcripts. It's like so and so s is saying this thing and it's spe right?

Same for newsletters. Newsletters are not that super easy to get because you have to subscribe to them. And so I'm I'm just selling the data. I'm selling the data so that they can throw their agent at it and make all the reports they want and get all the insights they want and whatever. Like and also like my hypothesis is data may still be a good business. Cause like code maybe isn't, right? I mean

yeah,

I mean as

I don't know.

in as in as in like I'm selling this and it's so hard to write the code, right? But I'm selling data. Yeah, yeah. I don't know. I don't know, man.

Right. Yeah.

No, it it does seem it seems like you're on to something here. just because the the way that you know, I just watched that video quickly, but as as I was watching just the the statements flash by, I thought, wow, you know, people used to spend, you know

hundreds of thousands of dollars per year to get this data from an agency and it would come in when the agency, you know, on the whatever cadence they set versus like, I just I want to talk to Claude about how my, you know, how this product is being received by the public. And

Yeah,

yeah, and and that's I think that's how it is. You sh they should use Claude and then they need access to the data.

Yeah.

over.

the the other things I wanted to just go through sort of just like a

Yeah, yeah.

a quick punch list are the the sort of things that are happening in the the news, not so much the news, but the things that people are talking about. I want to get your take real

We should, we should.

quick on there's

Chef.

people are yelling at Figma right now because they're so restrictive around who can use an MCP.

Or who who can use their MCP? That they're not allowing certain harnesses to connect and you know, so what's what's your take

I

on on that s problem?

I think it okay, I I actually have a take. I hadn hadn't heard this but so I have a take. so it's really tricky. I can totally see why they do that because if I can just throw my agent at Figma and and I am actually doing that, in a in a consulting project with a customer.

If can just then first of all just the amount of traffic they get, because I can just say, Claude, built me a whole bunch of stuff in Figma, and it's just gonna hammer it. And imagine that times a million, right? It's just gonna hammer it. So so it causes a technical cost problem, fine. And then and then strategically, so so so last year I'm talking to like one of one of my product managers.

and we use we we used to use linear, right? So linear is also famous for beautifully designed, it's just the UI is incredible. It the UI, the user interface is incredible. And he's like, Yeah, I never go to Linear, I just use Claude. And I'm like, and that was the kind of like before the whole Claude got

Mm-hmm.

super good. And I was like, What? You don't use yeah, he's like, Yeah, I just never go there. I just say like create a ticket and blah blah blah. And I was

Really surprised. And so for my own products, what I do now is I'm like, and this is similar to Hone also, right? Forget the user interface. Because I think people just want to chat with their agent and have it do stuff. And so sure, you still sometimes need something in terms of UI. Fair enough, all that kind of stuff. But in the big picture

for my products I just build the MCP. But then if you're a business like Figma, they're not gonna survive if if it's just an MCP, right? It just doesn't feel

Well,

like that that they th they will just like disappear eventually.

Right, but I I think that's inevitable anyway, because the so my take on on this is this is this is actually the real story about why SaaS is dying. it's the only way that they can protect their business model and you know keep their cost reasonable is to kind of do what they're doing. But the problem is that that introduces a bunch of friction into the process of using their tool.

And there are other tools, and I'm not going to say that Claude Design is a one-for-one replacement for Figma, but for a whole lot of companies and a whole lot of individuals, it is. you know, there still are designers who really know Figma and do really great work in Figma. but if but if I have to jump through hoops, right, yeah, yeah, yeah. That when that's what they're that's what

Yeah, but then they become then they're like we're only for the best stuff and now we're like a niche business or a smaller business and that's no good.

they're becoming, it's a niche business. And

The problem really fundamentally for me as a business owner is when I have a design that I've signed off on, I want that converted into code without having to jump through hoops to do a handoff. And so when you have things like in sort of the traditional SaaS way of doing things where you're paying for seats or you're getting features behind certain

levels like you can you can only get this feature if you have this level of of subscription, like any one of those little things that causes friction and you've built it that way in order to entice people to upgrade. Now people are just they get frustrated like I should be able to do this instantly. And they're not going to look at, you know, should I pay the money to upgrade? Should I no, they're gonna say I'm just gonna vibe code my own

And

so it's all of the dyn dynamics

It's yeah.

that are residual from the way that SaaS has always been operated. with you know, things like personal billing, gating features and things like that, that just it stops you when you say to your agent, Hey, Claude, I need you to do this. The moment that Claude comes back and says, I can't do that because whatever, I'm immediately looking for something else that's gonna solve my problem.

Because it's so easy

It it's

to find something else. So it's not so much a matter of saving money. It's a matter of I don't ha I don't want to have my tools interrupt my workflow.

Yeah, there's something there and and it's really funny because that like when I build products that's where I've been moving. I'm like so so Sputnik for example, right? Even I'm I'm kind of thinking like like the base level subscription is MCP only because because you don't need anything else. A and you don't wanna be using any UI. Why do you need a UI? Like it's not it's just data, right?

I mean in the case of Sputnik you probably don't need a UI. And so and so and then maybe the upgrade path is okay and you get like I don't know, API access if you want to automate more stuff or something like that or or whatever, but or custom stuff, but but yeah, it's just like sign up, get your MCP, get a few skills, you're done.

Yeah. Well that's why, you know, part of the reason I I try to stay with small, really focused things that aren't super expensive to run is that I can keep things really simple. Like it's just one tier. You if you want access thing, you buy this this one thing, you get access to everything. I'm not gaining anything, like you just use it however you want it. a little harder to do in a bigger business where you have, you know, real cost concerns.

But yeah, I think that's

Yeah.

that's gonna ultimately end up being the thing that kills off SaaS is not so much I don't want to pay the subscription anymore. It's I'm paying you all this money and your system is stopping me from doing what I wanna do. Either because I can't use my MCP

Yeah. Yeah. Too many limits.

because you're blocking my particular harness or

Because you just have some silly thing hidden behind a gate, or you just didn't think through your MCP well, and so there are some things that are still dashboard only or whatever. If I'm going to pay you money, I need you to get out of my way and just give me what I'm paying for. And so this is why I think the

Yeah, yeah, the yeah the one

Figma's really dropping the ball.

I think so. I'm sure they will be I mean yeah. The one thing I've found over the past let's call it let's say this year, right so far, is that I I also feel you just want the MCP. The one thing is like and and people will catch up, but very large amounts of people d do not know how to use even

Claude Cowork or Codex. And that's just getting started, right? So if my customers don't even use Cowork, then then they cannot be my I'm I think that's actually okay, like because the rest of the world will catch up. And every I I kinda feel like Cowork and Codex are just killer killer apps. Killer apps. Killa killer, they can make decks,

Yeah. Well, but to that point, what do you what do you th

they can make Excel sheets, they can do all the stuff people do at work.

Right. No, but but do you think that what's your take on AI for normies? Do you think things like Grokbot and dots and and all these things is I found I found that like Cowork is great. I can take I can automate business things that I need to get done. it

Yeah, yeah.

But I've I still just cannot get AI to be useful to me in my personal life, other than like the obvious I take a picture of a part on my appliance and say, This is broken, how do I find it? You know, because it's it's much better at searching the web than I am. So like that augmented

Yeah. Yeah. Yeah. Yeah.

s search. And of course living in a country where I don't speak the language fluently, every once in a while I was happy to be able to take a picture of a sign and say, Can I park here? But but

Yes. I did that yesterday.

Can I can I park here? It was very helpful.

Yeah. But I just don't think that like Grokbot and things like that, that try to just automate things more geared toward like personal life.

I don't know, I fe yeah.

Yeah, I feel like okay. So I have a t two takes on that. I think I think on the whole kinda like you know, bot buddy. I think I'm not sure that I don't think that's there today. It may get there. But then but it's also like the AI is just gonna go off and do stuff. But then I think the the really nice thing about

Codex and Cowork, like the desktop app experience, like that kind of thing, is that I still have a job. Like I'm still doing my job. Hey, can we make a deck for this customer? And I'm making the deck. I'm doing it in in I'm saying, Claude, make the deck, right? Then I have a look at the deck and it's great. And I send it and I'm still I'm still getting paid a salary. You know what I mean? And that's just a really good model. Because like Claude's doing all the work but I'm telling it what to do. I don't know.

Like that sh feels feels like I really wanna adopt it, you know.

I don't know, I just think that they haven't cracked a a user experience that I think is gonna work for like my mother.

Yeah. Y you know what's happening?

Yeah, d I just don't know. I

don't know it's gonna I don't know how long it's gonna take for it to become something that's really useful outside of the context of of business. I'm surprised I'm saying that because, you know, six months ago I would have said I don't know if it'll ever become useful outside the context of software development. But it has reached into several of the parts of my business and made things a lot easier. And so I'm able to do some automations that aren't software related. and

And take advantage of it for like like you were saying, you know, doing slide decks and presentations for clients, things like that, that you know, it it used to be bad, now it's good. but

Yeah, yeah.

I d I don't know, I just it and might just be because I didn't see that coming. Maybe I have a blind spot and maybe you know there will come a day soon when just everybody has their, you know, grok bot or you know, bot buddy or dot buddy or whatever.

that just kinda follows them around and makes her life easier. but I just yeah.

It c it it could be, but but

this year feels like the year of like, you know, the the Codex and the Coworks 'cause it's just really good. And also the year of like the Claudes and the MCPs and you know it's just I just I just and and yeah, and

Yeah. That's feels like enough. It's been a big year.

it it it is enough. It's October first. And also like so one last thing. so since I think about two weeks

I am talking to my agents, like with voice, and I'm and this is just maybe a tip. I'm using Muesli U E S L I. It's open source on GitHub, so you can just download it on on your Mac. And then it downloads a parakeet model and it just transcribes everything you say. Just press the function key. And it's it's very, very good. Like and and also like I I was listening to a podcast yesterday.

where the anthropic guy says, like was talking about this this kind of like working model of talking to your agents and it's and so it's not a it's not cleanly written because when you speak it's not like clean sentences and stuff. But it's it's much better because you're giving a lot more context. You can be like, wait, what I really mean is this and and kind of dig in and and and so what I say, you get a lot more text and there's a lot more

Context in there. And the model can figure out all the ums and ers and and kind of messiness because it all just goes it's all just like words that go into the into the you know the latent space and now you're giving which your input, you know what I mean?

And I'm a bit do you talk to your is that how you work?

I I don't, but I I can see where it would work, because the transcription is gonna be at least as good at capturing what's actually intended than my build than my typing is. I mean if you if you see me on Twitter you know that about every eighth or ninth word there's some sort of a catastrophic typo. And I don't correct that stuff for agents. I just you know, I type what I wanna type and I I hit enter.

Me neither. And the and but I think the typos

the typos also don't matter that much. It's more about if I can just like s and maybe it's personality, if I can just speak out loud, I will say more things. 'Cause then I'll be like, yeah, and the context of this is like from the business perspective and blah blah blah blah blah blah blah. And it just all goes in there and it's not a very cleanly written thing, but it's like as if I was talking to like, you know, my head of engineering kind of thing.

Yeah. No, I I might give it a try at some point. I've I've been reluctant because I do kind of like typing. And the other thing is that

I do a lot of my work in the evenings when, you know, my kids are asleep and my wife is asleep. And so just being able to sit in silence with my headphones on, listen to music and down in the basement and clicky clack away, I I'm not disturbing anybody. but I could see, you know, 'cause I do tend to when I get excited about something, I tend to start talking louder and louder and louder. And if you know, if I'm having

Uh-huh.

a disagreement with Claude instead of going caps lock, I might end up yelling at it, waking up the family.

Yeah, but that's that's cool. Yeah.

We should wrap it up. It's been an hour.

All right. Sounds good. Yep. Yeah, I think so

Good first podcast. I think it's good. All right.

too. We had some some good stuff in here. And so we will do this again soon.

Yeah, like and subscribe, people.

Yeah, yes, definitely like and subscribe. All right. we'll get the next one on the calendar. All right.

All right. See you later.

Thanks. Bye-bye.