Ian Landsman (00:00)
Welcome to Token Town. Today is May 28th. I'm your host, Ian Landsman, joined by a special guest, AI developer extraordinaire, Pigford. ⁓ Today also brought to you by Bento.
the email marketing and CRM platform for the AI era. Send your product marketing and transactional email with Bento. Thank you so much to Bento for sponsoring the show. All right. So Josh, I mean, a lot of people in our circles will know you, but, ⁓ you know, I think you've kind of made, I feel like you've had a whole evolution revolution of your identity. You're always a guy who was known for building a lot of products and doing a lot of stuff, but now you're just supercharged by the AI. It's like,
You have these incredible superhuman powers. So tell people a little bit about yourself and maybe a little bit about your AI adventure so far.
Josh Pigford (00:50)
Sure, sure. So ⁓ yeah, I've been building software for twenty years or something ridiculous. ⁓ and you know, yeah, I've always had ⁓ a lot of stuff going or at least have done a lot of things sometimes in the real was one after the other, kind of like stepping stones, but now it's ⁓ and I guess probably what most people know me for historically would be bare metrics, which started that.
Ian Landsman (00:57)
Right? Crazy.
Hmm
Josh Pigford (01:19)
And sold in twenty twenty. ⁓ and then I didn't waste any time and kept building software. So when when AI kind of started hitting ⁓ some sort of you know amount of popularity a a couple of years ago, I jumped in and it just sort of fed my ⁓ desperate need to keep building stuff. And now now it's just like, you know, poured gasoline on the fire. So ⁓
Ian Landsman (01:46)
Yeah.
Do you feel like it reinvigorated you a little bit or you were already going?
Josh Pigford (01:50)
Yeah.
It did in a way of like, you know, software has never been a craft for me. ⁓ it's not I don't I'm indifferent on code. I don't love code in any form or fashion. It's just a means to an end. ⁓ and so this this like helped me take even a further step back and like abstracted it even more, in a good way for for me, at least for my style. and so just the feedback loop.
Ian Landsman (01:57)
Right.
same. Right? Yep.
Josh Pigford (02:19)
So much faster. you know, pros and cons to that, which you know certainly can talk about, but the but for me, you know, especially from like an ADHD perspective, it's like I can just like rapid fire at stuff. ⁓ so it's fun. It's it's a lot more fun for me now, for sure.
Ian Landsman (02:36)
Agree.
Totally agree. I should mention that Aaron's out today because he's on a boy's trip. So brought in Josh to fill in. Um, and yeah, like, so I think you and me are very, very similar. You're more prolific than I am for sure. But in that, like the code was, it's always a means to an end. The products are what get us excited, right? It's like building things, getting users, shipping things and like, yeah, the code. I was never the world's best programmer. You know, I'm an okay programmer.
⁓ And this is I feel like this is our peak time for people like us is you know, it's like, okay Yeah, the code we're cool. Like the AI is already better coder than I am great now that check check on code we're done with that and I can focus on features and ⁓ What the products do and things like that. So really cool time Okay, so, know what we have some things to go through some AI takes and ⁓
Josh Pigford (03:08)
Sure.
Yeah.
Ian Landsman (03:30)
Yeah, let's jump into it. So the first thing I thought was interesting for us to talk about that happened this week was Microsoft had like, you know, one of their events and they released Microsoft Scout, which is basically, you know, open claw, ⁓ kind of dolled up with some enhancements for increased security and things of that nature. they're calling them auto pilots. So I guess they have co-pilots and autopilots. I own autopilot.app. If Microsoft, if you're looking for a domain hit me up. ⁓
But anyway, what's your take on this? I know you were, you were big into the claw for a while there. I feel like I've heard less from you on that. So what's your experience with open claw in general? And then it's sort of corporatization here. What are you thinking about that?
Josh Pigford (04:07)
Yeah.
Right. So I I am I'm no longer an open claw user. I ⁓ I've I've tried a few different other these sort of like personal AI tools and and still dabble some. Open claw was too too unstable for me to ⁓ keep up with. I think like with Scout with Microsoft stuff like
Ian Landsman (04:17)
Okay.
Josh Pigford (04:39)
To me I mean I always think of like either, you know, this enterprise side is what I associate with Microsoft, or like my like my grandparents or something. Like, and so do either of those are they a decent fit for for ⁓ this sort of yes, this like personal AI thing. It it feels early to me only because like, you know, ⁓ it's AI is still super early. ⁓
Ian Landsman (04:50)
Bro.
Yeah. ⁓
Right.
Yep, super early. People don't get that. It's so insanely early.
Josh Pigford (05:10)
for for for any of that. Yeah.
Right. And like and people don't associate like when they think of AI, they like at best think chat GPT. like that's the thing that people have experience with. And so I don't know. It just it feels it feels like weirdly early to try to introduce something like this.
Ian Landsman (05:31)
So you think it's odd too to even do it on top of OpenClaw. Like that doesn't feel Microsofty to me, right? Like Microsofty thing is like, oh, we will rebuild this whole thing from scratch. We'll copy it and we'll build it in the Microsoft way. And it'll be very Microsofty and it'll have all the hooks and permissions and stuff. But like then you're also getting the benefit of like the Microsoft. Yeah, it's got the hooks and permissions and it's all thought through from the ground up as like being enterprisey and corporate. Whereas this is not gonna really have that, right? Cause it's like the foundation's super shaky, so.
Josh Pigford (05:34)
Right. No.
Right. Yeah.
Yeah.
Ian Landsman (06:00)
I don't know. That's a weird take.
Josh Pigford (06:01)
I it's like ⁓ to me it feels like this sort of weird Microsoft trying to to jump on ⁓ a wave and like stay stay culturally relevant, at least in the technology space, but like they missed that wave a little bit, like they're even late to that. So ⁓ so yeah, to me it's just it's super weird. ⁓
Ian Landsman (06:12)
Right.
Mm-hmm. Yeah.
Yep.
Josh Pigford (06:27)
feels like the kind of thing that just kind of it just dies off and nobody talks about it ⁓ ever again.
Ian Landsman (06:32)
Right. That's the first and last time it's heard of. Yeah. And it's kind of like counter to, feel like to the, like what we're just seeing in the past. I don't think I have it even on the board here, but we could talk about a little bit of like, ⁓ I think it was Uber is like, we're putting caps on all of our spending for AI usage per employee. And I will know a lot of other companies have jumped on board with that. So it's like, it doesn't even now, like they've already missed not only the boat and open cloth, but now it's like all the way to like,
Josh Pigford (06:36)
Right. Yeah.
Ian Landsman (07:00)
No, you can't even run, how are going to run a claw with like a $1,500 a month cap on your tokens or whatever? I mean, like, it's just not going to work. ⁓ You kind of need a bunch of tokens as sort of its whole thing. feels weird that way too, but yeah. What do you think about that? Company's limiting, limiting the spend.
Josh Pigford (07:21)
⁓ I I it makes sense. I don't like I feel like in the past sixty to ninety days that's become such a giant core conversation in the AI space that, you know, it felt like everybody everybody was talking about ⁓ everything being so heavily subsidized and like, it's gonna you know, the eventually the the goodwill and all this money's gonna run out and it's gonna get really expensive. And then it just sort of happened really quick.
Ian Landsman (07:41)
Mm-hmm.
Right.
Josh Pigford (07:49)
And ⁓
and so I don't I don't know how like I don't know what's realistic on the cost side. Like, you know, most tech costs costs go down, right? Like as as technology improves. And here either either it's going up because companies realize there's a lot of money to be made and so they're just raising prices 'cause they can, or it's legitimately like they were so heavily subsidizing it like they're just they're just bleeding money. I who knows what's true.
Ian Landsman (08:00)
Right.
Mm-hmm.
Right, very hard to know.
Josh Pigford (08:20)
Right. So, you know, I don't know. Like, ⁓ there's I think everybody in my head it's ⁓ everybody's subsidizing this stuff and hoping that it gets cheaper. ⁓ but whether they reduce the cost or just keep raising it because they can, who knows?
Ian Landsman (08:31)
Right.
That's the thing that's tricky, I feel like, too, with like...
feel like for people like us, the cost benefit is super clear. I am very clearly way more productive. there, you could never convince me that I have not three, four, five times more productive. It's certainly encoding than I was before. but when you're in like a corporate structure where it's not so clean as like one person makes all the decisions and they're just doing whatever they want to do and they can maximize that productivity where it's like, no, there's like a whole department to people. might have like
10 repos involved in one feature, right? Like you have all this stuff, all this interconnection, all this approvals. Like, so you don't even know if what you're spending, it's gonna be much harder to suss out if what you're spending is actually increasing productivity, because even if one or two devs are more productive, but there's 50 devs involved, it's hard to know if you've got actually the feature delivered faster or not, you know, and all those kinds of things. So yeah, and then the other part of that being like,
Josh Pigford (09:35)
Sure.
Ian Landsman (09:39)
Companies are used to paying for software for developers and other employees, but it's like 50 bucks a user or a hundred bucks a user or whatever it is, right? It's like more of a baseline. We know per user XYZ. So this is like just sort of per user's uncapped, unlimited spending is a whole new thing that probably not going to make a lot of sense in the corporate world.
Josh Pigford (09:45)
Mm.
I
I think the transition is what's sort of interesting to me. ⁓ you know, how do you take ⁓ an engineering ⁓ team or organization or whatever and take them, you know, how do you how do you help them become sort of AI native in the development sense instead of you know, I feel like there's maybe the majority of engineers, ⁓ career engineers.
Ian Landsman (10:01)
Yeah.
Josh Pigford (10:26)
⁓ are at best copy pasting into chat GPT, you know, to ask questions like they're like using it as a replacement for Stack Overflow. And yeah, and which is like that's not terrible. I mean, whatever. That's that's that's it's an improvement for sure. But it's it is not the same as like you and I like having agents sit there and run for hours churning out code. And ⁓ and I think that's
Ian Landsman (10:30)
Right, I kind of agree.
Right, they're just doing that.
Right. It's an improvement, right? Yeah.
Right.
Josh Pigford (10:53)
the the cost difference in that, like giving an engineer a chat GPT license, is very different than having to pay API costs for for a clawed account for every engineer. And ⁓ to me I I think it's like you you end up with this sort of like great reckoning of having to that's I I think in my head, this is what happens where everybody starts getting laid off because they realize there's like this some
imbalance of like productivity. ⁓ and there's a probably a weird mismatch in like it being very hard to have a team that's a mix of AI forward engineers and then engineers who are like anti AI completely, right? Like 'cause they feel threatened by it. And so, ⁓ I don't know how you I don't know how you reconcile that on a team without just like starting from scratch or having everybody be on board with it. ⁓
Ian Landsman (11:25)
Mm.
That's
a really interesting thing because I actually just had that come up this week with a customer emailed in about help spot or a help desk product. And they have like, they're, they're a team that's like in the EU and the U S and they're like the EU people. This is kind of counter everybody's intuition. The EU team is like all about the AI. They're using the AI stuff. They want more AI than even we have, or they're going to have, they're doing their own side things, whatever. And then
They're like, we can't even get the U S people to use it. Like the U S people will not use the AI features. Right. And so like, we're going to turn it off for the U S and whatever. And so right now they're just going to deal with these two teams using it in different ways. ⁓ but yeah, probably that's not the end state. Right. So how that all works out, which side wins is quite fascinating. Yeah. How that goes.
Josh Pigford (12:37)
Sure. I I I I
guess I wonder if some teams will you know, it feels like it's inevitable. You can't avoid it ⁓ from it as far as like building engineering teams. But my guess is like you'll have ⁓ people s saying, well, we're we're gonna stay, you know, n
like handcrafted code, whatever you want to call it, right? ⁓ and like that's that's like the hill we're gonna die on. And then they'll like in five, five plus years be like, God, we can't keep doing this. ⁓ or you'll just see them like say, like everybody has to use it. Here's the systems we'll put in place. I d I think like I was working with a team ⁓ like helping with some development stuff ⁓ a couple months ago and I think that their engineering team overall is maybe a couple dozen people.
Ian Landsman (13:02)
Right, pure. Yeah.
Yeah.
Mm-hmm.
Mm-hmm.
Josh Pigford (13:30)
And they
had they had made the switch and were like super into clawed code specifically and had built these ⁓ pretty incredible systems of skills and sub agents, all like specific to clawed code. And like the entire dev process was basically just like you'd have a linear ⁓ ticket written by ⁓ a skill and then like to get it built, they would like
Ian Landsman (13:44)
Mm.
Mm.
Josh Pigford (13:59)
say skill slash whatever the linear ID and then it would go and like pull all the information from wherever, pull from the app, pull from different repos, whatever, and and then just build the whole thing. And then it would have a couple things for like QA and then like how to merge it and all this stuff. And all I was doing was just like running skills over and over again. ⁓ and just like making sure that it wasn't doing anything super stupid. And ⁓ and it felt it was weird, but it like also felt a little
Ian Landsman (14:04)
Mmm.
Right.
Yeah.
Josh Pigford (14:28)
bit like, this is the future. Because like, no way we were pushing out stuff every day that would normally take months to to push out. And ⁓ so that that felt like that's where it's gonna go, but like they had to just bite the bullet and say like, this is and I say that it was for a new product. Like this is established company, but they were launching an entirely fresh product. And so like that they were they almost had to start from scratch to be able to like really do this and not fear they were going to break everything. ⁓
Ian Landsman (14:30)
Right.
Right? Yeah.
Mmm.
Hmm.
That does give them that opportunity though, right? Cause it's like, this is a new product and we can kind of, you know, green field it and like just like start every new, new structures for everything. Whereas like when you're working in an existing product, although I sometimes feel like the AI is the best in existing product in a certain sense. Cause it's like refactors and things are so much easier, like stuff you would just never even take on before. Cause you're like, well, that's going to be months of wasted time, like trying to figure that out, whatever. And it's like, well, we can send the bots off on it and
Josh Pigford (15:14)
Sure. Yeah.
Ian Landsman (15:26)
that come out with something reasonable, know, it's all right.
Josh Pigford (15:26)
I do
I kind of felt like ⁓ in an existing code base, especially one that's like maybe a couple years older or plus, you know, the w to me it feels like AI is very good at replicating patterns, right? I mean, it is. That's the sort of like the the whole root of it. But like the problem I found in like old code bases was that it would sometimes try to replicate the pat like what were pre AI patterns that maybe weren't great because I'm not a
Ian Landsman (15:41)
Right? Yeah.
Mmm, yeah.
Josh Pigford (15:55)
not some sort of amazing engineer. And so like I don't want it to it required a little bit more hand holding, I found, but you know, still good. Right. Yeah.
Ian Landsman (15:57)
Right.
If you wanted to improve on what's there. Yeah. Gotcha. Yeah.
what are you doing? Move on something a little different here. ⁓ what, what do you do for your setup in terms of, know you've gone through a few different things with like how you manage your token consumption, right? So like, you on plans? Are you on per token? Are you, you I moved recently to like, for just corporate reasons to try to get us all on the team plan.
But it definitely sucks compared to the personal plans because you just have way less default tokens. And there's like now all this micro billing happening and stuff, which is, I feel like does like the suede me a little bit, even though like cost wise, it's not even like I'm spending that much more. It's just like annoying to ever have to think about the tokens verse when I was just on the personal max, it's like, fine. I have two personal max plans. I never think about the tokens. So, yeah. So I don't know what you're set up there. think people would like to hear that.
Josh Pigford (16:51)
Yeah.
I ⁓
I am currently I do everything on a single five X Claude plan.
Ian Landsman (17:03)
What, single 5X? Wow, okay. All right, what do you got? Yeah.
Josh Pigford (17:05)
Now here's the deal, Ian. Here's the deal.
Pre ⁓ before Claude, you know, they mean man, they love to like freaking burn some bridges. and be I was I was doing everything in conductor, which is, you know, sort of dev platform. I don't know. Yeah, harness is probably the right word for it. ⁓ when I was doing everything through conductor, ⁓ making use of my subs when those were allowed, I needed, I needed two.
Ian Landsman (17:15)
Yeah.
Mm-hmm Harness,
Josh Pigford (17:35)
20x Claude plans. And then I also had a whatever 5x codex plan. So I was spending that's 500 bucks a month. So two $200 a month plus a $100 codex plan. ⁓ and I was using the I was using those up a ton. Like I I I I track all that stuff and I was using something like 18 to 20 billion tokens.
Ian Landsman (17:37)
Mm-hmm.
Josh Pigford (18:04)
It's just stupid. And ⁓ and then when you know when they said we're cutting all this off and basically can't use conductor anymore, ⁓ without just like yeah, without all these workarounds, I went straight into the clawed desktop app, ⁓ clawed code in the desktop app. And I don't know what they're doing. I don't know what they're doing, you know, subsidizing it even more, whatever.
Ian Landsman (18:19)
Mm hmm. Okay, yeah.
optimization or whatever.
Josh Pigford (18:29)
I am still pumping out a massive amount of code and I never hit my limits and I'm on I'm spending a hundred bucks a month with Claude and that's it.
Ian Landsman (18:36)
Interesting.
So you're a desktop guy. You're not, you didn't go to the cloud code CLI.
Josh Pigford (18:39)
I've
so I said what I started with was just in the terminal. And then when I started like really getting ⁓ prolific and like how fast I was moving, ⁓ I needed work trees. and I because I'm not a developer in the traditional sense, like I don't love seeing code or like I I'm a visual I'm a visual coder in this like I need my dev tools to to make sense visually to me.
Ian Landsman (18:42)
Mm-hmm.
Mm. Mm-hmm.
Right. Yeah.
Right, Yep.
Josh Pigford (19:07)
And that means rarely looking at code. And so I don't
need that interface. Like I don't I don't ⁓ you know need VS code or so I don't want any of that. And so Claude Codes desktop app is like perfect for me in that sense. And it's got a bunch of built in stuff that helps like, you know, browser tools and it does all it automates like going computer vision stuff. Like it's great. And ⁓ and so and yes, and so and apparently it's much cheaper.
Ian Landsman (19:16)
Right.
Right.
Yeah. Interesting. And so maybe, optimal.
Yeah. Apparently more optimal. So that's an interesting, you heard it here first folks give that a try. If you're hitting your token limits, check out the cloud code desktop app. All right. I'm going to, I haven't actually tried that in a while since, ⁓ yeah. Right. Yeah. So I'm going to, now I'm excited. Now my afternoon. I know what I'm doing here. I'm going to go check that out. So that's, that's what I'm going to do. that's, that's fascinating. So yeah. So.
Josh Pigford (19:34)
Yeah.
I know. It's crazy. Well, I hadn't either until they they made me because c they killed off conductor. So
Ian Landsman (19:56)
That's what I'm on to the corporate plans max out at like the five. You can't buy a 20 X. You can only buy a five X. ⁓ now do you do any of the, the other thing I like to do is I'm also a Claude guy, but then I will run it through codex usually for like a second opinion. Are you a second opinion guy or you just do it all Claude or what do you do?
Josh Pigford (20:13)
Man, I I
that was that's been the biggest pain point for me is cause conductor because it supported both, I would be in the same work tree and I would do like an adversarial review thing with codex and like loved that. And I have not because of like work trees, I get there's gotta there is 'cause it's all files, right? It's all sitting your desktop. ⁓ but I haven't I haven't gone through the the hassle. I have like
Ian Landsman (20:20)
Mm-hmm.
Right.
Yep.
Yeah, it's probably hard to point it out. There's a way to do it for sure, but it's not easy. ⁓ Right. Yeah, it's there. Yeah.
Josh Pigford (20:43)
built a couple of additional skills for doing these like adversarial reviews ⁓ as part of my process. Yeah, but ⁓ I have not I I don't ever jump over to codex for
Ian Landsman (20:46)
Mm-hmm. Just in line, yeah.
Yeah. All right. That makes sense. Yeah. ⁓ I do think that is like an interesting and important step. Cause when you do the adversarial reviews, you do get a lot of interesting stuff out of that usually. ⁓ yeah. So, but I do also agree that like, I don't think it's so important that it's, well, I don't know if I agree with that. I do think this different model is kind of interesting in that you get like a little different take on it, but, ⁓ but even within the same model family, I do think just doing it at all, like
Josh Pigford (21:04)
yeah.
Ian Landsman (21:19)
make forcing it to take the separate perspective usually do get a lot out of that. Okay, so I don't know if you saw this about codex sites. This seems like something that's up your alley. So I wanted to get your take on codex sites. For those who don't know, this is a thing open AI launched where right within open AI, you'll be able to just say, build me a site and it'll be able to make a dashboard. It has access to a database, has access to file system and has access to all that kind of stuff. It's all powered by
a friend of the show CloudFlare. So it's like D1 and R2 and all that stuff underneath and workers. So I think it's kind of interesting them integrating this. I feel like I can see you building a thousand things that you just have like doing little things just like via sites. So I don't know what you take on sites.
Josh Pigford (22:07)
⁓ I I think it's great. ⁓ I do have a bunch of stuff where I use Cloudflare's. ⁓ they've named it multiple things now, but it's like their workers, pages. Yes. So I have I use that a lot for just like one off landing pages for things, ⁓ or you know, a couple of pages. Like I have a couple of Mac apps that need to need a like a little marketing site with some stuff. And so it's super easy to just spin up a Cloudflare page, whatever, and throw it there. and this feels like the
Ian Landsman (22:15)
Pages or something or workers. Yeah
Mm-hmm.
Right.
Josh Pigford (22:36)
A way to like remove some sysadmin kind of stuff. So I'm a big fan of that. ⁓ I so to me this feels certainly kind of ⁓ like lovable esque. and I'm yeah, and I'm and I'm a fan of that stuff, and I and I'm a fan of that ⁓ specifically like as a dad watching like like my youngest, ⁓ she's 16.
Ian Landsman (22:51)
Right? Step one towards that.
Mm-hmm.
Josh Pigford (23:06)
And she's like, she's not she's not super interested in development or coding, but she's like loves hobbies. She's like super into DD. And ⁓ and so like she's been she started building this whole like library for like her DD characters and her friends to all like store this stuff and like so she just needed to build this like database site and it's been doing that on Lovable, but it's it's a ha it's it's a slog in to like build anything like that. And
Ian Landsman (23:11)
Right.
Mm-hmm.
Mm, that's cool.
Right? Mm. Right. Mm-hmm.
Josh Pigford (23:33)
something that's probably arguably more accessible with this sort of all in one platform that sites claims to potentially be. I think it's a great yeah, who knows? But
Ian Landsman (23:41)
Right. If it actually is that right. Yeah.
I have the same exact thing like a month ago, me and my 13 year old, she like rides horses. And so we built this like horse name generator on her own site and whatever. she was soup. I mean, she's not no interest at all in programming, no interest in any of this. And then this is like totally opened her up to like AI and program. Like she's all super into it. And, but it's like, yeah, you build it, but then you have to like wire up all the BS parts of like, what was, we got to put it in get hubs. We can push it to this other place. And it's like,
Josh Pigford (23:52)
Mm-hmm.
Sure.
Right.
Ian Landsman (24:10)
That's all super
complicated and makes no sense with the like, I'm just building a website. So yeah, I'm pretty jazzed to, I haven't tried it yet, but I think maybe if we can move her thing over there, it'd be pretty cool to just, then she could just easily go in and do stuff on her own too. And not always like.
Josh Pigford (24:26)
I'm I'm curious, Chat GPT is so weird about there's there's just this like lack of feature parity between their web version and then their their Mac app and their iOS app. It to me, ⁓ especially for you know kids these days, like they they're so mobile first on everything. the ability to for something like sites to be accessible via just the Chat GPT mobile app would be
Ian Landsman (24:35)
Yeah.
Right.
Mm-hmm.
Yeah, just in the mobile app, yeah.
Josh Pigford (24:56)
You know,
they're sitting in like class and just like building a site. Like that would be great. I I think. But
Ian Landsman (25:00)
That's super. mean, this is,
yeah, this is one of the things I've been most excited about. And that, I mean, if I always say AI stuff, which if you're like us, the new stuff comes out you're like, that's great. I can use that. I get it, whatever. But it's like, this is one of the first things where I'm like, oh, this is like reintroducing people to websites and things like that. Like you can have your own website and it's not a pain in the ass, right?
Josh Pigford (25:19)
I'm
curious if this like does, you know, you mentioned that though because you think of like geo cities or like ⁓ then like blogs and all that kind of stuff, right? That which we which we certainly like were deep into all that stuff. And like curious if it brings any of that back, the because it's so easy to just spin up something, your own place on the internet, which I I feel like our you know, our kids and did not grow up with that. Like they grew up with social media and like that replaced all of the the blog.
Ian Landsman (25:26)
Right. Yeah. Right. MySpace.
Right.
Yep. Yep.
No, right.
Josh Pigford (25:49)
personal site stuff. So well, I'm curious.
Ian Landsman (25:50)
sure. Yeah, that
I hope they market it and push it that way because I think that would be interesting. And they're the most like, you know, chat GPT is much more consumer oriented. So I think it makes a lot of sense for them versus Claude. Although I hope Claude gets this too. ⁓ but yeah, I think there's some opportunity there. I hope they lean into it and like let you, I don't know if you can map domains, but I hope they let you map domains and like, just add more features. You can just do more stuff and the AI is going to be obviously quite good at these sort of simple websites and
Josh Pigford (26:12)
Yep.
Ian Landsman (26:20)
adding dynamic features to them and all that's going to be really easy. so yeah, really cool. All right, let's do some Q &A. Producer Dave, hop on in here. There he is.
Josh Pigford (26:25)
Yeah.
Producer Dave (26:31)
All right, so ⁓ this has been a very interesting episode. I think Josh must have attracted a whole new audience. Are you all familiar with Banker? Have you heard of Banker, B-A-N-K-R?
Ian Landsman (26:42)
Okay, all right.
Josh Pigford (26:49)
man.
Ian Landsman (26:49)
No.
Josh Pigford (26:49)
Run run far away.
Ian Landsman (26:51)
Okay. Sounds like you are, but okay. boy.
Producer Dave (26:51)
banker bot banker dot bot. ⁓ Just during this show,
six tokens have been token named token town have been launched and and I'm getting there. I'm getting DMS people are in the Twitter chat. They're like accept the fees. It's like this whole fascinating thing. ⁓ This is all everybody is talking about right now.
Josh Pigford (27:03)
They're they're the worst.
Ian Landsman (27:04)
⁓ it's one and crypto thing. Okay.
⁓ great. We're part of the culture
Josh Pigford (27:17)
They're
Ian Landsman (27:18)
now. All right, great. Yeah, that sounds terrible. All right. Don't don't if anybody out there gets approached by these guys, it's not official. Yeah. All scams with no crypto on this show. Hmm. Beyond that.
Josh Pigford (27:18)
they're a they're a they're a leech on society. I've n I've nothing nice to say.
Producer Dave (27:21)
⁓
Josh Pigford (27:26)
It's it's all scams.
Producer Dave (27:33)
So beyond that, honestly,
beyond that, the biggest thing is, ⁓ obviously, Aaron's not here, but somebody had to put a plug in for Solo. Solo routines with work trees are coming soon. So if you're trying to do stuff with work trees, ⁓ seems like that might be good way to go. But the other thing I was going to say is, because again, I'm not a ton of other questions. the only other thing I'll say is I have been a Claude desktop user for a long time. I'm also ⁓ not.
Ian Landsman (27:42)
Mmm.
Okay. Yep.
Okay.
Mmm.
Producer Dave (28:02)
primarily a dev, so at some point they're gonna figure out we don't need chat and co-work and code, like they will figure out how to merge that into one, but yeah, what's what, but yeah.
Ian Landsman (28:12)
What's what, yeah.
I thought ChatGBT is gonna combine all their apps into one. think I just saw something about that the other day. Like Codex and the ChatGBT app are gonna be one thing somehow.
Josh Pigford (28:15)
I think
Somebody did mention something that yeah.
To me, you you'd ask Ian at the beginning, like the whole I mean, it ties into the open claw sort of personal AI stuff. Claude's co work has has replaced a lot of that sort of thing for me. And and really it would almost entirely replace it for me if there was a mobile version. I mean, like there's some sort of mobile accessibility to it. ⁓ because right now I have I have like personal AI stuff set up through Discord and so I can ask access that anywhere. And ⁓
Producer Dave (28:36)
Hmm.
Ian Landsman (28:37)
Mmm, yup.
Producer Dave (28:46)
Yeah.
Ian Landsman (28:51)
Mmm.
Josh Pigford (28:54)
If I could if I could do cowork from anywhere on like via the Claude iOS app, that would that would cover most of it for me.
Ian Landsman (29:00)
Yeah.
Producer Dave (29:02)
All
right, we do have a question that came in. I think this is a good one. What aspects of people's AI setup do you think they overthink the most often?
Ian Landsman (29:04)
Okay.
Mmm, that's a good one. Take that one, Josh.
Josh Pigford (29:14)
I so I think people actually don't, it's not that they overthink it, that they they probably underthink it. They're probably not hey, ha. ⁓ I think I think most people are stuck, you know, you've got I mentioned this earlier, they're like copy paste into chat GPT as a just as a simple QA kind of thing, is like step one to just sticking your toe in the water. Then ⁓
Producer Dave (29:22)
⁓ Contrarian take.
Ian Landsman (29:22)
Hmm. ⁓ hot take. Yeah.
Right.
Josh Pigford (29:43)
You know, maybe that next step is, well, let me like open up my repo in codecs or claud code or in via terminal or whatever. And then you're just like working in it in this like one stream at a time. ⁓ and then like the next step past that is the is the work tree stuff. It's the ⁓ like I I've seen a couple people talk about like the death of ⁓
GitHub issues or like internal ⁓ like Kanban boards or something where it's like, well, you don't need to write anything into these to-do lists, just spin up a work tree and start building it. Right. And like I j like generally that is what I do. And so a given project, I'll have 20 work trees in progress. and ⁓ and I'm just sort of like churning along on those. ⁓ and because of like I've built these sort of internal skills that I use.
Ian Landsman (30:21)
Just do it, right? Just do it, yeah.
Right.
Wow.
Josh Pigford (30:41)
Things tend to run on their own without much intervention from me. for you know, it can be 20, 30, 45 minutes. So like I I need to do something else. And I can be doing something else. and I think like people don't make use of that enough. Like they stay in this sort of single lane of I'm working I'm working in the main branch and I'm just pushing stuff out one at a time, sitting waiting, go watch YouTube while it's churning away on this thing. Like, no, no, no, you could you can be doing a lot more.
Ian Landsman (30:44)
Mm.
Producer Dave (30:59)
Yeah.
Ian Landsman (31:03)
Right?
Right.
Josh Pigford (31:11)
you know, to the extent that you want like this existential ⁓ anxiety like the but it's fine.
Ian Landsman (31:15)
What a, a, good.
Producer Dave (31:18)
Ian, good. I was
gonna ask if you had gotten on the WorkTrees bandwagon yet. I don't think so, right?
Ian Landsman (31:23)
Only a little bit.
Sometimes the AI does it on its own and that's fine. But I have lately been hitting up. So normally I've been running with like multiple checkouts when I'm doing multiple things at once. But I think I've been thinking the last week or two, I'd need to move to the work trees and it's just going to be faster. Like going across the checkouts has its own annoyances of like keeping stuff in sync and all that. like, ⁓ so that was, that's going to be on my list pretty soon. Are you doing Josh, the whole like,
Are you really having each work tree spin up its environment so it connects to the database separately with its own database or whatever?
Josh Pigford (31:55)
Yes.
Yeah. And like and so I want to preface by saying I have no clue what work trees do. I have technically no idea. And ⁓ and I had never even heard of them until like a couple months ago. But then right. So you know, maybe it's been around for literally twenty years. I don't know. But okay. So ⁓ but so conductor is what introduced me to the at least the concept of this, and it became
Ian Landsman (32:00)
Mm.
Nobody does, that's fine. Yeah. Right.
Nobody had, right? It has, I think. It's been around a long time, yeah.
Mm-hmm. Right.
Josh Pigford (32:24)
one of the big things with conductor that was so great was its ability to spin up environments unique to every work tree. So and I've and I've transitioned this. It's a little bit more work to set up and clawed code ⁓ desktop, but I do have these sort of like pre run hooks where every work tree ⁓ is on its own unique port. Each one gets ⁓ in some projects, they each get their own
Ian Landsman (32:40)
Mm-hmm.
Mmm.
Josh Pigford (32:53)
unique database on red Redis database ⁓ Postgres ⁓ all it's unique to that thing and and I can then open any given work tree into the browser and like it's just working on that thing. And so yes I've I've that's core to my work tree setup is separate ports and all that stuff. Yeah.
Ian Landsman (32:54)
Mm-hmm.
Right, all the stuff, yeah.
Right? Yeah.
Yeah, I need to do that. think Dave brought up, think Aaron's tool solo, which is what I primarily code in ⁓ on Cloud Code command line. Aaron's working on some black magic that's going to make spinning up a lot of that stuff straightforward from solo. So I think that'll be pretty cool.
Josh Pigford (33:31)
It
it that's one of those that feels like you you're it makes sense when you like talk about it out loud, but like until you sort of experience just instantly spinning up this dev environment and you're not even thinking about it and it just all sort of works, ⁓ it feels insane. But
Ian Landsman (33:43)
Right? It's all magic. Yeah.
that's really cool. David something else or no?
Producer Dave (33:53)
The only thing I was going to say is, for you, know you like being in the code. You're a solo man. For someone who's not in the code, Polyscope does work trees in the Laravel world. Polyscope does it incredibly well. And because it's all tied in with Herd, you get the same thing where you immediately can launch it in the browser. And in Laravel world, it'll even sync with night watch issues and linear issues and stuff. So it's like this whole, I had this happen where I broke something. I was setting up a queue.
Ian Landsman (33:58)
Mm, sure.
⁓ right. Mm-hmm.
Yeah.
Producer Dave (34:19)
I broke something, it triggered it in Nightwatch. I went in the polyscope, I hit the fix this Nightwatch issue button basically and it did it. And I was just like, I'm living in the future. This is insane, but yeah.
Ian Landsman (34:26)
Right? It all, it all worked.
Yeah. w let's be clear. I don't like looking at the code. I'm a blinders man. put on, I have a Luke Skywalker when he's on the Millennium Falcon, he's got the blinders on, but, yes, but, ⁓ I do like Claude code on the command line for other reasons that are not really the code. just think.
Producer Dave (34:35)
Ha ha ha ha!
Ian Landsman (34:50)
I just think it works the best as a harness. The desktop apps, don't like how they all have this sidebar of crap that I don't want that sidebar crap. The stuff I'm working on, yes, but it's like all this stuff that I don't want to look back three months ago and see what I did. I don't want to look back yesterday and see what I did. So I have to archive it and do all this stuff. So I don't really like that all of them have that and I hate that on all of them, but I am going to try cloud desktop again here because maybe it's the benefits outweigh the negatives.
Josh Pigford (35:15)
It's the secret, Ian.
It's the secret sauce. Yeah. Yeah.
Ian Landsman (35:17)
That's secret sauce that does. And
apparently token usage is one quarter of the other harnesses for reasons unknown.
Josh Pigford (35:25)
Look, maybe maybe
I've just outed myself and there's some bug with my account and like I'm about
Ian Landsman (35:30)
They were like, we love this guy. Somebody internal just like did a little if Pinkford, like just let this guy go. Yeah.
Josh Pigford (35:32)
She's right, it's fine. I
mean, I like the thing that tracks all my token usage also includes how much it would cost and it's like twelve thousand dollars worth of tokens. And like, yeah, like I mean, I'm not I can't I can't afford to do that every month. ⁓
Ian Landsman (35:43)
Right, when you were doing the...
Something's going on.
Right.
I love the idea that it's actually still using that many tokens, but you just have a credit card you forgot about and it's just like there's 150 grand of flood bills on there, you know.
Josh Pigford (35:55)
That's right. ⁓ I'm gonna
collections are gonna show up at my front door and be like, here's your hundred thousand dollar bill for Claude. god.
Ian Landsman (36:03)
Exactly. Yeah. Two months of Claude. Yeah, exactly. ⁓ man.
All right. Sounds good. All right. Thanks, Dave. All right. We got a couple more topics here. Then we will wrap it up, but this is just our, what we call our quickies, which is a couple of things that I thought were interesting. ⁓ alphabet raising a bunch of money, which you don't really think of Google, you know, alphabets, parent of Google. Do they need to raise money? They're Google. They literally print money from
Producer Dave (36:11)
All right. See you later.
Josh Pigford (36:29)
For white from who? Who are they raising
money?
Ian Landsman (36:33)
10 billions from Berkshire Hathaway. Um, and then they're doing like a 30 or 40 billion more from just open market, like bond sale type things and whatever. like raising money and selling, or I guess they're going to issue stock, I think too. Yeah, they're issuing stock, um, actually. between stock and bonds and everything, like 80 billion, uh, AI infrastructure is expensive. I think this kind of gets back to the beginning of the conversation where it just really is expensive to make a bunch of servers.
Josh Pigford (37:00)
In my head they've got
they've got like a trillion dollars. Like eighty billion doesn't sound like much for Google.
Ian Landsman (37:03)
Infinite money.
Yeah, I don't know. That's the thing. It's like, I think there is something, there's some idea that maybe it's also like,
It's just a way to like kind of announce how expensive this all is and like prep the stockholders and investors that like we really need literally as much money as we can get our hands on. But yeah, and the weird thing about like a lot of capital expenditures in business previously are like bigger things that last longer. So what's going to be really interesting is like five, six years from now. I mean, the chips are only good for like five or six years. Like they literally die. They just get outdated.
So you're going to, who knows like you're going to have to, you got to do this every six or seven years. Right. I don't know. It's going to be interesting. ⁓ and kind of related to that and thropic, ⁓ filed for S one with the sec so they couldn't potentially go public. They haven't announced going public yet, but, I guess with space X juice in the market, maybe it's time for, I'd rather be in entropic than space X. So I don't know. I'd rather buy.
Josh Pigford (37:46)
Every yeah.
Sure. I don't know. They've they've
the whole X their like the plays that they've been making with their like selling hardware, server, rack stuff, compute is kind of an interesting take. Like I don't you know if
Ian Landsman (38:16)
Mm.
That is, yeah. Well,
they built it for themselves, but they didn't need it. So then they were like, yeah, let's lease it Anthropic for a billion dollars a month. man.
Josh Pigford (38:24)
All right.
Well in their
whole move with ⁓ with Cursor and like, you know, giving themselves the option to to buy them for what for whatever? Was it sixty billion? Was that what it was?
Ian Landsman (38:33)
Yeah.
So what do you think?
Let's say a little side adventure quick on that cursor. I don't know, man. I'm not a cursor believer. I've tried it like two or three times. I don't get it. I don't want it. I don't want to see the code. I don't don't I don't know why I'd use this over Claude or even Codex. I don't know. You got a hot take there.
Josh Pigford (38:51)
Yeah.
I I there was a time before Claude Code, ⁓ where I I was a big ⁓ cursor user and we so like maybe maybe finance, I like ⁓ previous startup. ⁓ we were there were two engineers, two of us working on it, and we were spending two thousand, three thousand dollars a month in on cursor.
Ian Landsman (38:58)
Mm-hmm.
Josh Pigford (39:19)
And it was it was great. It was fantastic for what it was at the time. ⁓ but yeah, but then once like Claude Code came out and we sort of transitioned away from it and I haven't I haven't really touched it since.
Ian Landsman (39:21)
Yeah, okay, right, because it was, yeah, there was no other options kind of.
Yeah. I think that's the thing. That's why that deal is really crazy to me. Cause it's like, well, open AI and Anthropic are never going to let that market, like they are going to always be on the cutting edge of that market. They could even reserve models at some point and things like, like they have a lot of control there that like cursor. I don't know how they're going to really match that, but I don't know. Right. SpaceX just a dumping ground for like everything. Right. So I guess it all, it all ended up in there and maybe something will pop out. ⁓ okay. So final bit,
Josh Pigford (39:47)
Yeah.
Yeah.
Ian Landsman (40:02)
What do think about Opus 4.8? Cause I feel like, um, there was a lot of big claims made. I feel like it's probably not as down as 4.7 was for people, but still not like still pretty muted reaction. So what do think?
Josh Pigford (40:16)
I think muted is good ⁓ because going from four point six to four point seven was was bad. ⁓ it just yeah just felt like it changed when that when that swapped out ⁓ as far as like skill usage and like just workflow, it broke a ton of stuff, at least for me. And I that felt like the reaction. With four point eight it was almost no change. ⁓ whether it's better or not, I
Ian Landsman (40:18)
Okay.
Right.
Mm-hmm. Yeah.
Josh Pigford (40:45)
I don't know. I got like I I'm spitting out code so much all day, every day that like as long as it works, I'm kind of indifferent. ⁓ it's it's not like it it's more efficient. Now I don't know what I don't even know what that would mean. So ⁓ I think the fact that it didn't just like burn everything to the ground is probably a a positive
Ian Landsman (40:46)
Right.
Yeah, I've had pretty good luck. I do. actually feel like it's a little bit slower, but then they've, but they've added these other modes, which now I'm just running in the higher modes. So it's just thinking more. So I'm trying, I don't know. What do you do with that? What's your effort level? What do you like as an effort level?
Josh Pigford (41:11)
Yep.
I leave it on ⁓ high and claw. Mm hold up. How do you say that? I feel like I've well I c I've been ha I have been having some some ⁓
Ian Landsman (41:23)
yeah, it's good. want to know this is important. This is important for your token usage too, you know, maybe you're on medium.
Josh Pigford (41:34)
like rate limit issues and I'm currently set it's currently set on high. I don't know if I'd reduce that in the past couple days because I kept getting rate limited, but regardless, that's what I just leave it there. I don't I don't otherwise I don't change it. It's not like I'm like, this is gonna be extra difficult. Let me bump it up. Don't really think about it.
Ian Landsman (41:38)
Mm.
Okay.
You're cranking on high and it's going alright.
Right.
Yeah. And then you never dropped the sonnet for any implementation stuff, right? You just opus the whole way.
Josh Pigford (42:04)
I in
my head, if I if sonnet just means it's going to screw it up. But that may not be right at all. It might be fantastic, but like I d I just I don't trust it for whatever r I mean, 'cause it's not the latest and greatest. So
Ian Landsman (42:10)
Right. Yeah.
Yeah.
Right.
Yeah. I want to play with it a little bit more with, I've talked about this on here before, but I use superpowers, which is like a skill set and like kind of structure some stuff. And one of the things that's designed to do is structure everything in the planning phase so that like you can use the lower model in the actual implementation. I always override it and just tell it to use Opus for everything. Cause like, I agree. just wanted to like, give me the smartest brains you got and let's like go after it, but maybe there's something there.
Josh Pigford (42:37)
Sure.
Yeah.
Ian Landsman (42:43)
Maybe if it's like, if it's all set up enough, the sonnet really will be sufficient, you know.
Josh Pigford (42:47)
I it's some people talk like, you know, say trying to use different models from like a speed perspective. But again, like I'm spinning up so much stuff in a given day that the idea that I need that to go faster, no, I don't I don't think my brain could wrap my head around like that much more code. ⁓ I'm I feel like I'm reaching my limits.
Ian Landsman (42:52)
Right?
Right.
Right. Well, this is an interesting,
an interesting thing of working on a number of things simultaneously, right? Is that you don't even care how fast it is because it's like, yeah, because you're going to go off and do other things. And then at some point you'll come back and it'll be done. And if it took five minutes or 20 minutes, it doesn't really matter because you were doing other stuff in there anyway.
Josh Pigford (43:14)
I don't. No.
Well and I think there's an argument to be made that like ⁓ the downside of this sort of rapid fire, you know, you can spit out a whole thing in minutes or whatever, means that you you never stop to second guess if you should have done it at all, or maybe a different way to handle it from a UX perspective. I don't know. So it's like this it almost forces this sort of breather to to sort of gut check like, maybe that's not a great idea. Whereas I could have otherwise if it's too fast then I'll just do it, ship it, then
Ian Landsman (43:42)
Mm-hmm.
Right.
Josh Pigford (43:54)
Don't
think about it again. I don't know. There's all sorts of ways to spin that.
Ian Landsman (43:57)
You've
got to wonder if someday there'll be also some like token pricing. That's like, Hey, if you can take an hour for this, we're going to give you, it'll be half the cost or whatever, which I mean, I would make that trade off a lot. Like if it's like, you just want to fit it in when you can. And that's helpful. And you give me a discount like that'd be fine. like, uh, yeah. So, all right. That's interesting. Um, it's, we got a little feedback here real time. Sir Claude Boris said, uh, to always use the latest model. Yeah. You know, I think that's true. That's how I approach it right now. But I also think.
Josh Pigford (44:05)
Right. I for sure.
Ian Landsman (44:27)
Right? Like a year from now, Opus 4.8 level will be the cheap model, right? Like whether it might be called Sonic, but like that level of capability will be what it's doing. So you do wonder at some point if the K if you're shipping all this stuff with this model, then in like a year from now, if you're still okay, if you're okay right now, like you're probably that's you're probably doing all right. Even dropping down a tier of the model potentially. I mean, we'll see if we feel that when we get there, but, in theory.
Josh Pigford (44:50)
Potentially, yeah. Yeah, I am I am curious. Like
I already think if I had to drop down to four point six, I'd be like, no way. Ca four point eight is so much better. I have no idea. You know, it may spit out the exact same code.
Ian Landsman (45:02)
Yeah, right. Yeah, there is a whole psychological
thing that's quite interesting there as well. Yeah. All right. Well, that's the show. Thank you so much, Josh, for coming on. Thanks to Bento for sponsoring the show. Aaron will be back next week and we will see you all then. Thanks everybody. Have a good one.
Josh Pigford (45:13)
Thanks for having me.