You've Been a Bad Agent

CGPGrey Video on Yearly Themes

Creators and Guests

Host
Matt Carey
agent and mcp at Cloudflare
Host
Wilhelm Klopp
building @kolo_ai

What is You've Been a Bad Agent?

Wil and Matt discuss tech, startups, and building really cool things with AI. Sometimes joined by (actual expert) friends.

Wilhelm (00:06.653)
we're back. Happy New Year.

Matt (00:09.716)
And we're back! Happy New Year, TT.

Wilhelm (00:12.125)
Wait, I figured out a riverside feature. Listen to this.

Matt (00:27.18)
Holy shit, how did he do that?

Wilhelm (00:29.243)
We have an intro, hey! It's so fun when you can listen to it, because if you just edit it in afterwards, you don't get the hype of the intro, but it's a very hypey intro, so.

Matt (00:32.856)
That's amazing.

Matt (00:42.72)
You know this is probably the first time that anyone else has heard the intro apart from us. Because I'm pretty sure you couldn't hear it before.

Wilhelm (00:47.213)
Yes, yes, it's been half a year in the making but we finally get there. That's exciting.

Matt (00:53.752)
Yeah. Yeah. We've crossed, I think we've crossed a boundary dude. Like the fact that we've done, we've done enough pods now. I mean, we have like three more that haven't been published, but we, I like, have done, we've done a fair amount. You know, when we first started, we're like, oh, it would be like massive to do one and then, it would be like massive to do three. And then it'd be like massive, insane to do five. Holy shit. I think we've done, we've done a bunch now.

Wilhelm (01:04.567)
Hahaha

Wilhelm (01:10.609)
Mm-hmm.

Wilhelm (01:18.619)
I think we should really pat ourselves on the back. it's true. And I was looking through this and I was like, wow, I can't believe we haven't done them. I can't believe we have done so many over the whole past year. Yeah, 2025, big.

Matt (01:28.632)
Yeah. And people actually like them. People listen to them. Like 2025 has been pretty cool.

Wilhelm (01:35.345)
Yeah, I agree. It's nice. It's nice. And it's always like you wouldn't expect it, right? Like someone comes up to you at an event and is like, yeah, yeah, I listen to it like all the time. Yeah, big fan of this medium.

Matt (01:50.178)
Yeah, no, it is cool. Because for us, it does kind of sound like you're shouting into the void a little bit. Well, we're just shouting at each other, aren't we? But like, I can imagine people who do this by themselves. Yeah, dude, it's nice to see you. I feel like, sorry, okay. I haven't seen you in ages. Well, I saw you in London when you were sneakily there. And then...

Wilhelm (01:56.006)
Yeah, yeah, yeah.

I'm just happy to catch up with you.

Wilhelm (02:07.805)
Yeah, yeah, yeah, that was really fun actually. That was a cool day as well because London in the past, so I was there for a day in mid December and then for a week in like early January over New Year's Eve. And that week, man, it was so cold. Like it's so cold in London now still, I think. But in December it was very nice, very pleasant, like good, line bike temperature. Line biking around London is just, it's just so good.

Matt (02:33.199)
I was on one last night on the way back from London Bridge and I swear it was like bent it was like really broken but like not in a way that you could see when you just like walked up to it so it looked fine and then I got onto it and I was riding and I was like I swear I'm going sideways like like I'm going sideways like something is off yes that was fun yeah line bikes are great

Wilhelm (02:50.371)
Hahaha

I think someone said there's like a mental checklist that they go through whenever they look at a line bike. It's like, okay, is it charged? Okay, is the, like, do both the tires have air? Yeah, has it got both pedals? Because, yeah, that's the most frustrating thing when you unlock it or whatever, and you start getting charged like by the second. And then you realize, shit, yeah, it doesn't have a second pedal, which makes it quite hard to ride a bike.

Matt (03:04.576)
it got both pedals.

Matt (03:23.01)
basically impossible. Although there is a guy around me who has one leg and I've seen him on a line bike and I have no idea how that works. I don't know how he does it. Yeah, I literally don't know how he does it. But yeah, I've seen him on a line bike a bunch.

Wilhelm (03:24.967)
Basically impossible, yeah.

Wilhelm (03:30.886)
Wilhelm (03:34.247)
we should have him on to discuss that.

Matt (03:42.424)
Crazy, crazy. Anyway, how's SF? Is it warm?

Wilhelm (03:42.429)
That's wild. Okay. Um, mate, have so much to talk about. feel, yeah, it is really nice. It's warm and the sky is blue. Well, it's not warm, warm. It's like 10 to 12 degrees, which is like comfortable walking weather. Like, I think I've become such a wimp because I think the London winter didn't use to phase me, but I was like, I don't want to be outside. I don't want to walk. I would do, Oh man. I had a couple of things where I had to buy something. Um,

Matt (04:01.484)
No, London Winter is freezing.

Wilhelm (04:11.644)
in London and I was like in Piccadilly Circus or something and I had to go to like John John Lewis in Oxford Street and I'm like first of all I don't want to go there at all but then also I don't want to walk there one because it's freezing but and also because there's so many people so I would do these like awful awful change thrice tube journeys in central London just so I would avoid walking which that feels pretty sad

Matt (04:16.12)
Over Christmas is awful.

Matt (04:38.86)
Yeah, that's tragic. No, I remember... Yeah, that's actually tragic. I remember like a couple of years ago, you you still live in Camberwell and cycling up from Camberwell, I remember being in like a full alpine belay jacket, like the thickest thing known to man, with gloves, hat, thermals on, and I'm freezing, like actually freezing. And I got off the plane earlier today.

or yesterday and the cold in London, it does hit different. It's like, it's not actually that cold. It's like minus two or minus three Celsius. But it just, like, it goes, I don't know, it goes in your bones, dude. Like in France, was minus 14, minus 15. And I went for a run. Like I went for a run and it was fine. Like it was a bit cold. And like the day when it was like minus many teens.

Wilhelm (05:13.244)
Hmm.

Wilhelm (05:21.872)
Ghosts to the bones.

Wilhelm (05:27.861)
wow. Seriously.

Matt (05:37.356)
That was freezing, like felt really, really cold. But if you wrap up warm enough, it's kind of okay. I don't know, I think it's a dry cold. It's something about like the permanent moisture of London that fucks you up.

Wilhelm (05:37.841)
Mm-hmm.

Wilhelm (05:49.092)
I see, see, I see. Yeah, yeah, yeah, yeah, interesting. I buy it. I also had to look up where Canberra was just now. It's quite central to London, yeah. It's right next to Peckham.

Matt (05:55.09)
actually you've never been it's quite south yeah it's it's actually turning into one of those places with like really good food yeah like kind of bougie like weirdly yeah anyway you know where it's not gonna be sorry sorry what'd you say

Wilhelm (06:11.558)
You know, I used to tell people I was from Peckham.

I used to tell people that I was from Peckham.

Matt (06:19.234)
Did they believe you?

Wilhelm (06:21.264)
rally

Matt (06:22.978)
Yeah.

Matt (06:26.949)
I don't know where you were going with that. I don't know if that was a joke or like...

Wilhelm (06:28.998)
you

No, no. Go on, go on. What were you going to say?

Matt (06:32.792)
What? I was going say, do you know where it's not as cold?

Wilhelm (06:44.154)
Ooh, I think I know, but please, I don't want to steal your thunder here. Where is Not as Cold? Where is Not as Cold and where is the city most like San Francisco in Europe?

Matt (06:49.718)
No, no, no, no, go for it, steal my thunder.

Matt (06:59.379)
Lisbon! Lisboa! Yeah, so I guess like this is like time to break some news that I'm moving into Lisbon. What? Mate, so you have a bridge and I just wanted my own bridge, you know? So I couldn't come to see your bridge. I needed my own bridge and that's why Lisbon came in. Yeah, so it should be fun.

Wilhelm (06:59.874)
Lesbian, what?

Wilhelm (07:10.374)
What?

Wilhelm (07:15.92)
But in London that's way more precious.

Wilhelm (07:26.65)
Super exciting news, yeah, huge, huge year for us. It's been in the planning for a little while. You have a lot of, I think, wealth from crypto and you need to tax shelter that. Hence the move to Lisbon makes a lot of sense. Tell us more.

Matt (07:39.864)
That is not the reason I'm moving to Lisbon. That's ridiculous. what do I say to that now? Yeah, so there's good food, there's good wine. In what coin?

Wilhelm (07:45.198)
Of course, of course, that's what I would say too. But, please do elaborate.

Wilhelm (07:55.814)
Just tell me it wasn't all in fart coin.

Fart coin? That's a thing, right?

Matt (08:01.849)
farts coin. okay. I don't want to move on conversations too quickly. But no, Lisbon is going to be awesome because there was like a connecting thing to that. This is going to take us on a massive tangent.

Wilhelm (08:07.375)
No,

Wilhelm (08:13.114)
No, no, we love tangents. We love tangents.

Matt (08:16.298)
No, no, but this is like a whole tangent that's gonna like take the rest of the episode. Okay, so, so, Lisbon's gonna be really fun. Super nice weather, Cloudflare have an office there. That's basically it. Like, yeah, I'm really excited. It's gonna be fun.

Wilhelm (08:20.507)
Okay, okay, fair.

Wilhelm (08:31.998)
a really nice office as well from what I hear.

Matt (08:34.616)
A beautiful office. Yeah, when are you going to come co-work?

Wilhelm (08:38.01)
Wow, that's a good question. Maybe in the summer.

Matt (08:43.682)
Yeah, cool. We're having an MCP maintainers meeting in February in Lisbon. I can now finally offer nice places for people to come. So yeah.

Wilhelm (08:48.746)
no way.

That's cool. I've seen these MCP maintainer meetups, I think, on Twitter whenever they happen, and it looks really cool. It looks like an awesome group of people.

Matt (09:02.104)
Yeah, I think what you're probably talking about is like the spec teams, which are like, yeah, they're like big brain stuff. We're just hacking around with random bits of JavaScript. So we're like, we're much more small brain, but the spec teams, they're doing the full stuff, the full bias.

Wilhelm (09:06.799)
yeah.

Wilhelm (09:13.687)
cool, cool, cool.

Hahaha

Wilhelm (09:22.0)
And speeds become big. Wait, okay, but walk me through more of the reasoning of Lisbon or what you're excited about. I know like the surfing is a big part of it as well,

Matt (09:30.774)
Yeah, like the water being just there, the surfing, like a bit more nature as well. And they've got an insanely cool tech scene that's just getting bigger and bigger and bigger. And yeah, I wanted to, chance to explore that. been in London for a few years now and just see something else. Wanted to move to Europe as well and Lisbon seemed like a good place. Yeah, that's pretty much it. I'll be back in London, like a fair amount.

Wilhelm (09:55.184)
Hell yeah.

Matt (09:59.385)
because we're still going to be running AI demo days, just organize the February one and the March one and potentially an April one in your side of the pond. But like it's very, very much like initial steps for that one. Yeah. But this is as early as I've ever been, like almost I'm like two and a half in advance. How cool is that? Yeah. So yeah, but I will be like flying around here, there and everywhere, but mostly back to London a few days a month, definitely for that.

Wilhelm (10:02.14)
Mmm.

Wilhelm (10:09.276)
Ooh, really?

Wilhelm (10:16.219)
That's very cool.

Wilhelm (10:20.548)
I see, see. Yeah, yeah, that's awesome. That's really cool.

Wilhelm (10:27.46)
Nice. I remember, I think I mentioned this to you, but one of the groups I really enjoyed hanging out with in London was the indie beers community, like the indie hacker community led by Charlie. And they had a very early offshoot to Lisbon. And I think there's now a thriving like Lisbon indie hacker community.

Matt (10:37.142)
Yeah, yeah, Charlie, yeah, yeah.

Matt (10:45.986)
think there is. Well, you know, like Levels.io lives in Eressaea, I think is how you say it. Like literally like 30 minutes up the coast. So yeah, I'm gonna find out where he lives. No, no, no, it's not. The Algarve is in the south. Eressaea is like, you have Lisbon with the the river coming out of it. And then if you just go a little bit north on that western coast, if you go too far north, you end up at Nazaré where the big waves are.

Wilhelm (10:51.702)
yeah, yeah, yeah.

Wilhelm (10:57.668)
Is that the Algarve? Is that what it's called? Or is that somewhere else? Okay.

Wilhelm (11:08.592)
Yep.

Wilhelm (11:16.028)
Mmm.

Matt (11:16.312)
And if you end up going even further north you end up in Porto, but like in between Nazaré and Lisbon is Eresséa. Eresséa, which I'm probably saying massively wrong like yeah, but yeah, the Portuguese is gonna be a whole whole new adventure.

Wilhelm (11:22.288)
Gotcha.

Wilhelm (11:28.08)
And that's where Peter levels lives. my God, yeah. So what do people usually do in terms of learning Portuguese?

Matt (11:39.521)
night school, I guess?

Wilhelm (11:41.702)
But so you're planning to learn Portuguese? Like that is like a, no way.

Matt (11:43.64)
Yeah, like I'd like to. I know people who've moved to Lisbon and not learn Portuguese and that seems kind of sad to me. So yeah, if we're talking New Year's resolutions, which I think we should, one of mine is to be a functioning Portuguese speaker.

Wilhelm (11:53.436)
Mm.

Wilhelm (11:57.528)
We definitely should.

Wilhelm (12:03.898)
Wow, by the end of the year.

Matt (12:07.574)
Yeah, I think so. Like, yeah. I mean, I heard it's a very tough language to learn and I've been dipping my toes in the last few weeks and it is seemingly very tough. But yeah, that's the hope. That's the dream.

Wilhelm (12:08.122)
That's exciting.

Wilhelm (12:21.859)
It's fascinating because you hear it, when you hear it spoken, I feel like it sounds almost more like Russian than like Spanish.

Matt (12:29.686)
Yeah, they have like a sh sound, which sounds quite Russian-y.

Wilhelm (12:34.681)
Yeah, maybe that's it. Man.

Matt (12:36.692)
It sounds cool though. It sounds cool. I started watching this new Portuguese TV show. It's like on Netflix. It's called Gloria. Yeah, would actually recommend. It's like spy thriller thing. Yeah, but it's all in Portuguese. So starting my adventure. Gloria, Pássia da Natas, loving life. Yeah. Yeah.

Wilhelm (12:45.082)
Okay.

Wilhelm (12:56.902)
my god. Yeah, They're just so cheap in Lisbon, aren't they?

Matt (13:03.01)
They're just so good. Like, I don't care how cheap they are. They're just so good. Yeah. No, it's gonna be fun.

Wilhelm (13:06.582)
Hahaha

That's awesome. Obviously, I'm imagining like it's kind of like B for indefinitely for now. But what are you thinking? is it just to you? Does this feel like a sort of two, three year thing or like a five, 10 year thing?

Matt (13:24.661)
I honestly don't know. honestly don't know. Yeah, can't give you much more than that. My girlfriend's French, so at some point she's probably gonna want to go live in France, as French people do.

Wilhelm (13:30.191)
Ha ha!

Wilhelm (13:34.895)
I didn't realize that that's Juliette's long-term ambition. I also didn't realize she was French French. I thought she was British French.

Matt (13:42.506)
Nah, she's French French. She just has a mad British accent because of university and stuff. Yeah, but...

Wilhelm (13:44.293)
French French.

Wilhelm (13:49.023)
really? No way. So she was born in France, but... and all this stuff.

Matt (13:53.593)
Uh, so yes, her dad's, her dad's Scottish, but her dad, lived in France for like 25 years. Um, but her mom's French, French, family's French, French. Yeah. But yeah, she grew up in France. Like I was literally just at her parents' house. It's so fun because she lives, they live in the mountains in the Alps. And so it's just like, yeah, in Chamonix. So it's just like, it's insane. It's super fun. Amazing mountains.

Wilhelm (14:00.188)
God, okay, no way. Yeah, I thought she was British. Like, yeah, yeah, fair, fair. man.

Wilhelm (14:17.669)
How long were you out there for? You did tons of skiing, right, over Christmas?

Matt (14:21.72)
I was just there for the week, like from New Year until yesterday. And not tons of skiing, because mostly they do like ski touring and like walking and stuff. So you just end up walking for ages to ski like five minutes. Basically is what you end up doing. And the snow wasn't like, isn't really good. Actually, as we were leaving, like it was bucketing it down.

Wilhelm (14:26.758)
got it.

Wilhelm (14:36.827)
Mm-hmm.

Wilhelm (14:42.779)
I see, see, I see.

Matt (14:50.038)
And so I think the next week is gonna be really, really good. If anyone is out in the Alps, they're gonna, yeah, very jealous.

Wilhelm (14:56.805)
That's awesome. Yeah. I know Chamonix is like such a triathlon trail running paradise. Like everyone. Yeah. Yeah. Right.

Matt (15:02.07)
Yeah, in the summer. Yeah.

Yeah. No, in the summer, like the UTMB goes from there. So like the ultra trail in Mont Blanc, like it goes round Mont Blanc. It's like a massive running race, like 170 kilometers or something. And that's like the super famous one. And you'll see UTMBs like all around the world now, but it started in Chamonix. And then I've actually, I've actually got a running race this summer. The Chamonix marathon, which is like, I watched Juliette do last summer and now I'm going to do it this summer.

Wilhelm (15:18.541)
mental.

Wilhelm (15:30.458)
Whoa.

Matt (15:35.948)
Yeah, it looks really good fun. It's kind of like a third or no, it's like less than a quarter of the UTMB, but it's not the same route. It's a bit more, a little bit more technical route and yeah.

Wilhelm (15:38.159)
That's epic.

Wilhelm (15:46.533)
Yep.

Wilhelm (15:52.421)
and it's still a marathon in the mountains.

Matt (15:55.225)
Yeah, it's like 57 ish K, I think at the end of the day. They're always longer. They're always longer than what they say they are. Yeah, it says it's, sorry, no, it's like 47 K. So they say it's like 42, but then when everyone does it, it ends up being more like 47. I think it has two and a half thousand meters of elevation in it.

Wilhelm (15:58.203)
so it's a lot more than a marathon.

Wilhelm (16:06.458)
47, got it.

Wilhelm (16:15.086)
Yeah, that's crazy. When is that in the summer?

Matt (16:17.016)
But it's that in the end of June, I think, there's actually a really fun thing that like, because I got like an eco pass. They because like last year, I've been trying for last couple of years and never got a spot. And it's like, yeah, there was like a lottery and it was like one in 12, then one in 15. And like, I was like, never getting a spot. And this year is probably more but this year they released a bunch of bunch of spots.

Wilhelm (16:21.165)
Okay, cool.

Wilhelm (16:31.787)
It's like a lottery system.

Matt (16:44.44)
but you have to, that's like very like eco-first. So you have to go by public transport. And I'm like, yeah, I'm well down for going by public transport. Kind of thinking I was still going to be in London, which makes it very easy. You just get the train there. And now I'm like, now I'm going by public transport. Yeah, yeah, yeah, yeah. But it means I got a ticket, No, I don't think I'm playing.

Wilhelm (16:56.088)
Wait, how? you have to, travel to the race by public transport. That's the condition of the ticket. Interesting. I mean, a plane is public transport, isn't it? Like it's not, unless it's your private plane, I would consider that public transport.

Matt (17:13.096)
Nah, I don't think a plane counts. I don't think a plane counts. so yeah, I think you could do like car shares or trains, maybe ferries, which is kind of questionable about how good they are for the environment ferries. But yeah.

Wilhelm (17:27.29)
Let's take a cruise from Lisbon to Monaco and then happy days.

Matt (17:30.072)
Yeah, literally, literally, I was considering something, but maybe not that, maybe train to Barcelona. And you can also do car shares as well. So could do like a road trip along the bottom. Could be interesting. Anyway, sorry, we chatted way too long about this, but this is my exciting news. I until it's running a big old race in the summer.

Wilhelm (17:46.66)
That could be fun too.

No, man. It's huge. Still with Cloudflare, same team, nothing changing on that front.

Matt (17:56.793)
Oh yeah, still like smashing out work on MCP. I've got some really fun, yeah, really fun bits for agent integration coming up. We're gonna make code mode awesome. We're gonna, there's some other stuff I can't talk about, but yeah, generally it's gonna be really fun.

Wilhelm (18:11.202)
I love it. Big news. And this is happening like soon, right?

Matt (18:16.278)
Yeah, like probably by the time this goes out, I will have moved. That's why I'm talking about it. Hopefully. Otherwise the next one. Sorry.

Wilhelm (18:23.874)
Yeah. So are you in like big moving stress at the moment or is it more chill?

Matt (18:28.862)
I mean, I just called up a bunch of movers and they all, a few of them gave me some quotes that weren't insane. So should be fine. There's a bunch of like weirds, paperwork you have to do as with any moving of country. And Portugal is quite funny in that a lot of them you have to do in person in an office, which is, you have to book an appointment for. So I have a friend or actually a colleague who moved to Lisbon with CloudFlare.

Wilhelm (18:41.978)
Mm-hmm.

Wilhelm (18:48.658)
Matt (18:54.834)
and ended up like without wifi, without electricity, without heating for like two weeks because you need this number to do anything. Yeah. So I'm going to try to avoid that and get the number as soon as possible.

Wilhelm (19:03.466)
Interesting.

Wilhelm (19:10.298)
Smart, get that number. OK, cool. So we start talking a little bit about news resolution stuff. Let's talk a bit more about that. I imagine for you, like, yeah, there's a lot of, like, Lisbon adjacent stuff. Yeah.

Matt (19:19.788)
Yeah, big one for me is learning Portuguese. Yeah, big one learning Portuguese shipping. There's some really fun stuff with MCP, like V2 is going to come out that sort of brings the SDK into like, sort of like current generation of JavaScript tooling. So it should feel really nice to use. It should feel like very modern. No, the MCP SDK, the legit, the official one, try and bring all that into like the one.

Wilhelm (19:38.222)
Does the agents SDK.

Matt (19:46.391)
Here we're going for like a monorepo approach. So if you're just using the server, you don't need all the stuff from the client. We're like doing deployment options. We're doing like middleware for different frameworks. We're going to have, we're going to have all of it, you know, it's going to be, it's going to be awesome. And like this, yeah, every API is going have an MCP server because it's going to be like one line in whatever framework you're using.

Wilhelm (19:52.121)
Mmm.

Matt (20:10.488)
That's the plan.

Wilhelm (20:10.906)
That's awesome. That's cool. Wow. No way. So you've kind of taken over the JavaScript MCP SDK.

Matt (20:19.416)
No, I haven't taken over. There is like four of us kind of working on it or three and a half of us kind of working on it. And I've been a little bit out of the loop in the last few weeks because I've been on another project, which yeah, which you'll see sometime soon. That'll be kind of fun when it comes out. But this one, back to open source, which is what I really enjoy.

Wilhelm (20:30.65)
Mm.

Wilhelm (20:41.208)
Yeah, yeah, awesome. Yeah, I was trying to figure out some New Year's resolution stuff as well. Are you familiar with the CGP Grey-inspired yearly themes?

Matt (20:53.61)
No. What's that?

Wilhelm (20:55.64)
Maybe we can put the video in the show notes for it, but he has this whole spiel about like, new year's resolutions, like, especially when you make them like very specific, they can be just like, kind of tough to follow because it's like, it's usually like a kind of somewhat ambitious goal and...

Matt (21:07.0)
Mm.

Wilhelm (21:15.106)
like, I don't know, work out every day or whatever. And then it's both kind of hard to hit and then also whenever you don't hit it, you kind of beat yourself up over it and you feel kind of bad. And then also it doesn't really account for all the other stuff happening in your life and like how the year might develop. So...

Matt (21:17.592)
Mm.

Matt (21:24.0)
Yeah.

Wilhelm (21:32.684)
So he recommends this idea of picking a theme for the year. Like for example, the year of reading would be like a common example. And then the idea is that in every decision that you make sort of throughout the year, you kind of consider the theme. like if you're standing like in queue, like, so it's not like I need to read a book every week. That would be like a...

kind of like a news resolution where you know easy to fail that easy to beat yourself up over but if you do like the year of reading it's both like yes i do want to read books a lot but like if you have more busy times and you don't read a book for like two weeks then like you haven't failed the theme or like if you're standing in in line to get a coffee at like a coffee shop instead of like pulling out twitter and scrolling on twitter you might decide to like read like an article or whatever on your phone so it's kind of like a bit more forgiving

Matt (22:12.993)
Yeah.

Wilhelm (22:28.013)
and a bit more broad and a bit more high level. And it can kind of orient all the various decisions that you make in life. So every time, I think the way he frames this is like every time there's a fork in the road, you can like pick the side of the fork that is like more aligned with the theme.

Matt (22:31.361)
I like that.

Matt (22:46.648)
That's cool. That's like sort of along the idea of trying to be sustainable with your emotions, right? Like don't go on crash diets because if you want to like, you want to crash diet, you won't be sustainable. then like one point you're going to be like, I really want that chocolate bar. have that chocolate bar and then you're, shit. Well, guest diet's over now. Let's fucking go. So yeah.

Wilhelm (22:57.785)
Mmm.

Wilhelm (23:10.239)
Exactly, Yeah, yeah, yeah. Well, I already ate the chocolate bar. I might as well like eat everything else today. Exactly. So the theme for me this year will be the year of embracing the American way.

Matt (23:16.738)
Yeah, might as well go to Mac Ease.

Matt (23:27.096)
Wow, you're going to invade Greenland?

Wilhelm (23:29.369)
Yeah, I appreciate that there's a lot of American news at the moment.

Matt (23:40.384)
of bracing the American way. You're just gonna turn up outside Downing Street, walk inside, grab boss man out by the scruff of his neck and take him back to the US.

Wilhelm (23:53.85)
Exactly. Yeah, it's funny. I've mentioned this in the people and some people are like, ooh, I hope not too much or whatever. But no, think to me, like what this means is just like approaching stuff with like optimism, like being very like positive. think like something Americans are very good at is like striking up conversations with strangers in public. Like to me, that is like, like that would be an example of like a specific thing I would like to do.

more, like be more loud, like share more stuff like in public, maybe get a driving license. I think that is like very American way. How do you not know this? No, I don't. So listen, I can drive. just choose not to. Until this year.

Matt (24:31.576)
Do you not have a driving licence?

Matt (24:36.696)
Can you actually not drive?

Matt (24:44.77)
Dude, that is a great New Year's resolution. That is a tick box exercise.

Wilhelm (24:50.073)
Do reckon? It feels kind of daunting to be honest, like getting a car, parking, all this stuff, but I do think it is embracing the American way.

Matt (24:51.03)
Yeah, get driving lessons.

Matt (24:58.892)
When you look at the absolute specimens of people who can manage that and still function, you'll be okay.

Wilhelm (25:08.345)
By specimen, do you mean yourself, or what do you mean? Oh, I see, I see. You're saying there's people, there's all sorts of people who can drive. I understand.

Matt (25:11.754)
No, just like, you'll be fine. You'll be fine.

Matt (25:21.25)
There's all sorts of people who manage to drive and don't find it daunting and are able to get about like a daily life, which includes driving. You'll be fine. Dude, you need to learn to drive. Open up some doors, man. That's a crazy amount of freedom you're missing.

Wilhelm (25:40.085)
Exactly. And that's another good word for embracing the American way, freedom.

Matt (25:45.944)
freedom.

Wilhelm (25:47.192)
I think another thing is interesting. actually, I don't know how much of a point I to make around this. But this is also a little weird to explain. But I think another part of the American way is when you have financial...

Hmm, how do best frame this? I think Americans solve problems by making more money, not by being, by saving more money. If that makes sense. You do? Okay, nice. Man, one of, so going back to Europe over... Go on, go on.

Matt (26:20.472)
Yeah, I know exactly what you

Matt (26:27.882)
It doesn't work for everything though.

Matt (26:34.76)
I think evidence, yeah, I'm just, watching a really funny TV show, which, it's not funny. It's not meant to be funny, but I'm finding it absolutely hysterical, which kind of embodies that, like, if you have an issue, just go harder. And at some point you might be able to dig your way out, like through the bottom or something. And I'm not sure, yeah, I kind of get it. I think about this with like the...

that UK has an insane amount of public debt and like we have a huge rising deficit and all this sort of stuff. And if we just made more things as a country and we had more tax dollars, then we could fund things that could make more tax dollars. And like, it's that like self perpetuating loop of like you do better things. Yeah, of growth. Yeah. And being like very growth centric.

Wilhelm (27:16.597)
Mm-hmm, mm-hmm, right.

Wilhelm (27:22.521)
Totally. Of growth.

Matt (27:30.144)
I haven't really heard that many people talking that way, especially of politician age in the UK, until... Have you seen Matt Clifford's video?

Wilhelm (27:36.569)
Mm.

Wilhelm (27:42.297)
Oh my god, yeah, wait, actually, I'm not sure I've seen this video, but I'm a fan of Matt Clifford's. I think he's done some really, really incredible things, especially on the political side. Well, what, he brought out a video recently?

Matt (27:51.244)
Well, do you think you'd be the next prime min- not that recently, like a few months ago. But genuinely, like I saw that and I was like, Matt Clifford, prime minister. If there was a polling market right now, I would bet on it.

Wilhelm (28:01.259)
I'd vote the hell out of Matt Cleffert. Yeah, okay, maybe we should make one. No, yeah, he's the real deal, big fan. And it was interesting going, okay, actually, I'll just give a simple example. say I wanna like get a car and...

Matt (28:05.965)
which is not financial advice, but yeah.

Wilhelm (28:24.344)
lived a car-centric life and embrace all the expenses that exist with a car, right? Maybe a car costs like a couple of grand, I don't know. I don't really know how much a car costs. But then also like parking, especially in San Francisco, can be really expensive. Like you can pay like 400, $500 per month just for a parking spot. And like, is kind of, right? But then I think the way you can approach that is you can either be like...

Matt (28:45.772)
That's outrageous.

Wilhelm (28:53.13)
Okay, like I could, I could like cut back on other things in my life. I could like, like if I don't want to increase my overall spending, I should forego some other things in life, like cut back on this, cut back on that, save up some money, then afford the parking or kind of figure out a way to like move to a different neighborhood where there's actually free parking. Like some neighborhoods of San Francisco that are maybe a little less convenient in other ways have like

more space for cars so you get like more affordable parking. Or I think the American way is like just figure out a way to make more money so that you can then you have an extra thousand dollars a month or whatever that you can then use to pay for your car and car related expenses. And I think that's it sounds kind of outrageous. I think this is like you can you can make decisions in that way or I think you can.

So this year will be the year of embracing that. But yeah, I think this is just like much less, much less common like in Europe.

Matt (30:02.444)
Yeah.

think how we would think about it would be like priorities, right? Like you're like, I want a car. So I should prioritize my, that goes up my like priority list. And so when I think about like moving into a place, I think about, like my brother did this. I didn't, I sold my car when I came to London, but my brother kept his and he managed to find a place still super central that just has parking on the street. Like, and like,

Wilhelm (30:09.271)
Yeah.

Wilhelm (30:20.728)
Mm.

Wilhelm (30:33.196)
Right, right. Yeah, yeah.

Matt (30:35.178)
London's like way more densely populated than San Francisco. I don't think you'd like, when you learn some of the hacks, I don't think you'll find a problem with parking. Like it will be very easy.

Wilhelm (30:48.492)
Yeah, and I don't want to say don't be smart. And I also don't, not trying to like knock Europe or anything like that. I just think like, like to me, the more American approach is like, I have a problem. I can solve it with money. You can also solve it with other ways. You can solve it with time or... One of the most memorable things for me in this past couple of weeks in Europe. So we spent two weeks in Germany and a week in London.

Matt (31:02.86)
Yeah, it's, you could solve it with a bazooka, yeah.

Matt (31:16.642)
Yeah.

Wilhelm (31:17.11)
over Christmas, which was lovely. But one of the things that was kind of crazy to me is like how, what an amazing life you can live on really not that much money in Europe. Like the grocery prices are just wild. You can pay like two euros and you get a really nice Parmesan, better than any Parmesan I could probably buy in San Francisco. And it's just like, whoa. On the train journey, we took like a four hour train at some point in Germany. And I just saw

a guy, like maybe a guy in his like mid thirties, and he had put like a yogurt, he had made his own yogurt and put it into like an old big jar of pickles and like screwed the lid back on so he could eat the yogurt on the train. And I'm just like, that is like, he probably paid like $2 for that yogurt overall, because it's just like, yeah, lower less.

Matt (31:58.113)
Nice.

Matt (32:06.572)
No, less.

Matt (32:10.449)
Way less dude. Two, like two dollars and I can buy a massive thing of yogurt from the shop.

Wilhelm (32:16.564)
Hahaha

Matt (32:19.064)
Dude, no, like San Francisco is expensive. San Francisco is expensive. Like you could also make yogurt if you wanted to, but San Francisco is expensive. And so you'd be like, oh, now I need two liters of milk. Oh, two liters of milk is gonna cost me $6. Why wouldn't I just buy a $3 yogurt?

Wilhelm (32:19.074)
That's what I mean though. It's just a very different way. Exactly. I think...

Wilhelm (32:32.48)
I think, right.

Wilhelm (32:40.226)
So I think at the fork in the road, when you see, OK, things are expensive, you could either decide to play the game of figuring out how to save at various points and be smart about it. Or you can take the path of, OK, let me just make more money, solve the problems with money. So I think the more American path is that one. Anyway.

Matt (33:01.314)
Yeah.

Is this like you motivating yourself to make more cash? Is that what this is? Is this the level that we've got to? It's like, you're like, I do not want to have to make yoga and put it in a pickle jar. know, I wanna get, for fuck's sake. yeah. Jesus. Yeah, I mean, whatever motivates you, dude.

Wilhelm (33:09.688)
Yes!

Wilhelm (33:20.28)
Exactly!

Haha.

But also, yeah.

Matt (33:32.578)
There is a beauty, okay, okay. I'm gonna, no, there's a beauty the other way around of living on not very much and being, having more free time, which I think Americans don't necessarily understand. Like I read something like bringing it back to vaguely like what we tend to talk about, like Jason Lou.

Wilhelm (33:34.508)
There is a beauty. There is a... There is a... 100 %

Wilhelm (33:59.256)
Mm.

Matt (33:59.393)
really cool, the cool rag guy, guy who made instructor, we've probably spoken about him before. He just got a job, he just stopped consulting actually and got a job at a lab starting very, soon. And he wrote a little like basically goodbye post to all of his, all of his kind of fans, like people who'd signed up to his consulting newsletter. And he also wrote a post about like things that he would, at least I think it was a post or maybe it was just a long tweet.

Wilhelm (34:10.284)
Mmm.

Wilhelm (34:15.767)
Mmm.

Matt (34:27.724)
about like things that he like kind of didn't really, didn't really make sense to him at 20. And it's like money that you save at 20, because like there's two very different schools of thought, they're kind of both right. And it's like all about like whether you save any extra cash you have or whether you use it for like an experience right now, because hopefully the ideal is that the money you make at 30 or 40 will like be inconsequential.

Wilhelm (34:33.908)
Okay.

Matt (34:56.044)
Like it will make the money that you've like, like scrabbled together at 20, 25, like inconsequential. And so his thought was like, and he expressed it in a very, in kind of the way that you're imagining, but he was kind of making the other argument was that like, he was like double working shifts in college to put $2,000 together rather than now he spends

Wilhelm (35:21.975)
Mm-hmm.

Matt (35:25.74)
4,700 on a business class flight to Japan when he's bored. And it's like, actually he didn't go on a grad trip because he was like working double jobs and trying to save that cash when actually he should have just spent that cash and gone on a grad trip because actually he was gonna make money afterwards. But like, I don't know how much I get behind that because like the whole like...

Wilhelm (35:29.354)
Right, yep.

Matt (35:51.639)
vision of what you're trying to do and how you're trying to live your life. really does probably has meant he got to that point where that sort of money is inconsequential to him. And I think that that also is important, but there is definitely a thing about like, you do earn cash to spend it and you should do that, especially while you're younger and able to spend it on things that are active and vibrant and travel related and all of that sort of good stuff.

Wilhelm (35:58.935)
Mm-hmm. Yeah, totally, yep.

Wilhelm (36:17.367)
100%.

Yeah, I completely agree with this. I feel like there's this meme which is prominent in the Reddit fire community, the financial independence. And it's like people forever scrape by or save everything and then eventually they hit their number where they never have to work again. And then I feel like you see a ton of posts that are like, okay, I hit the goal.

Matt (36:31.032)
Oh yeah, that community is weird. Yeah, yeah, yeah, I know that one. Yeah, yeah.

Wilhelm (36:50.039)
I've retired, I've quit my job. What do I do now? Like what's... This was the biggest goal in my life. And wait, how do I spend my day now? Like what am I doing? So I think like I'm a big fan of not pursuing that strategy because I think you end up with that conundrum of like what's the point of life and like what's the meaning of everything. And you actually should just figure that out on the way.

Matt (36:53.548)
Hahaha!

Matt (37:02.807)
Yeah.

Matt (37:15.798)
Moving this back to what we normally talk about, I am kind of interested how you think that interacts with where AI is going and the future of... Because I think in... I I told you when Opus 4.5 came out, this was a meaningful step in the model's ability to connect together various...

Wilhelm (37:20.565)
Yeah, yeah, yeah.

Wilhelm (37:27.297)
God, mate, I have no clue.

Matt (37:44.185)
actions, inferences, like inferences in terms of like extracting data from two different data sources in a way that's like not obvious. And like its ability to like put together like a meaningfully accurate piece of work, like small piece of work consistently, like every time. So that now people can do things like Ralph, you know, have you seen Ralph?

Wilhelm (38:06.954)
Mm-hmm.

Wilhelm (38:14.222)
Mm-hmm. Yeah, we should talk about Ralph, eh? Um, Ralph Wiggum. Yeah, yeah, yeah.

Matt (38:15.535)
Um, yeah, but before we talk about Ralph, before we talk about Ralph, like what, do you think this actually means? Because like, yes, sure. A lot of this advice people give based on their like preconceived notions, like my, my dad's advice would always be to like save as much cash as you can now, because like money in a pension at 25 is going to be a worth a lot more than money in a pension at 50 or like trying to put it in at 60 or like whatever. Um, and

Like it's very easy for me as like a younger person to be like, oh no, but like there's plenty of time for that, you know, like all of that stuff, but maybe there, maybe there isn't like maybe the work thing is, well the work as we know it is gonna like, well it is gonna fully change and like, what is it gonna change? Yeah, what is it gonna change to? Yeah, yeah, there's that meme as well, which I find hysterical. Like I see that one on Twitter all the time. It's like, you have six months.

Wilhelm (39:00.95)
The permanent underclass. How much time do we have to escape the permanent underclass?

Matt (39:14.518)
to trade your way on Polymarket to a gazillion dollars so you will never be part of the secondary class of this new society that's coming and I'm just like, wow.

Wilhelm (39:15.221)
Yeah.

Wilhelm (39:24.022)
Yeah, and there was a big like Dworkash post about like insane inequality that we might get to. Yeah, I find this very hard to reason about. I don't know, but I do think that the predictions people make often turn out to be like very wrong in shape, I guess. Like for example, right, a few years ago, I feel like in like 20, I don't know, 21 or whatever, lot of people...

I was hanging out with lot people in London who were very worried about AI safety risk and whatever. their thinking was like, or the way people would talk about AI risk is like, we have this super intelligence that kind of takes over control of the world and like Skynet or Armageddon or whatever. And in reality,

Matt (40:12.504)
Yeah.

Matt (40:17.558)
I think a lot of people still think that is a reasonable thing that might happen. Yeah.

Wilhelm (40:23.265)
Sure, sure. But I think the path towards that can be just like very... Yeah. The way I think this stuff plays out in reality is... Okay, I'll give two examples. This is the first one. The way we actually see AI influence things today, at least, is you have these subtle things. Like, for example, what was that report that MPs in parliament in the UK now give like...

Matt (40:27.33)
Well, not reasonable, but like a possible thing that might happen.

Wilhelm (40:50.583)
there's way more comments submitted by them or the way they structure their speeches is like very different because like they say things now like I rise to speak, which is apparently a very American way to like make an argument in a like parliament type body. But now all these like British MPs are doing it because they all writing their speeches and all their comments in like chat, chat, chat, chat, chat, chat, chat, chat, chat, chat, chat, chat, chat, chat, chat, chat,

Wilhelm (41:20.179)
Stuff is changing and stuff is changing in like weird ways, but it always happens like a little bit differently to how like we think. The second example is, I think you look at the very beginning of LLMs or whatever and you see like, yeah, okay. This will massive, like all software engineering jobs will go away. That might be like the prediction. And then in reality, know, software engineering is changing and the juries have been out on that. But

What you also see is everyone talking to Tachi BT as their personal therapist or their personal doctor, right? As like a first line defense for that. And then again, think like that, like no one was predicting this stuff that like these things would actually be like useful, capable therapists and doctors that you can speak to. So I think like stuff is changing for sure a lot. But in reality, it always plays out differently to these predictions people make, I think.

Matt (42:19.608)
Can I tell you one thing that I'm a little bit worried, actually two things I'm little bit worried about this. The first is actually what you were kind of what you were alluding to with the MPs. When I see like Chak Chibuti and the big LLMs, sort of like a distillation of the internet with some clever steps in them. And what's quite sad about that is like the majority

of their training data is in English and promotes like a Western, like a Western American and to some extent British culture, but mostly in American culture. I do think I have like a couple of worries about this in that there are going to be some parts. First of all, we have like a homogenization rate of all people using these LLMs where they're going to feel

or think or be told something in a very, similar way, which is kind of interesting. And people talk about the psychosis of LLMs where people go into these doom loops where they all start saying the same things. And is that like an LLM related thing, like an LLM related psychosis because everyone's being fed the same stuff. And you see this to some extent with social media.

Wilhelm (43:38.4)
Hmm.

Wilhelm (43:42.08)
Right.

Wilhelm (43:46.742)
Mm.

Matt (43:47.446)
And the crossover of social media and LMS is kind of interesting also where we have like Grok doing really weird shit. But you have like this.

Wilhelm (43:53.001)
Yeah. And you have people adopt these strange speech patterns in real life, right? Like the, it's not X, it's Y. Like people actually say that like non-ironically more now than they used to because it's such like a common thing LLMs say.

Matt (43:58.231)
Yeah.

Matt (44:08.748)
Yeah. And I do worry a little bit about like people, like especially parts of the world losing some, some cultural reference, not cultural reference, but some like, there is like a loss of individuality there. Like, like you and me as individuals as well. I mean, we're not writing novels, but we're writing code and there's definitely like, but there's definitely a loss of individuality of our like, of our code syntax, right? And if you try and extrapolate that towards like,

Wilhelm (44:25.451)
Hmm.

Not yet. Who knows?

Matt (44:38.584)
other types of knowledge work, even other types of like the arts, then potentially there's a loss of individuality on like the individual brush stroke on the individual word choice on that type of thing, but maybe like a greater autonomy on the overall idea. And I saw something where it was like people like the distillation of LLMs is like from money you get execution and

Wilhelm (44:57.611)
Hmm.

Matt (45:07.672)
yeah, like you give money and you get like some execution from LLM and previously, what was it? It was, it was something quite funny. It was like, you pay money to get skills was that previously you had to pay workers with skills to get money. Like it was something like, it was something like, like the wrong way around. But, but that is, it's kind of by the by that. That's the first thing, with LLMs where it's like this homogenization. But then the second thing is like, do you worry about like societies that maybe don't.

Wilhelm (45:12.16)
Mm.

Wilhelm (45:25.707)
I see. Interesting. Yeah.

Matt (45:36.384)
are not gonna share about this, not gonna share from this. And for one, I think that these societies are like countries where English isn't the most common language, which is like most countries, countries that like English speaking is actually very, very low, or countries where their language isn't very well written down, or people generally just don't write it down. And they'll like potentially lose a lot of that.

Wilhelm (45:40.736)
Hmm.

Wilhelm (45:49.94)
Mm.

Wilhelm (45:58.134)
Mm.

Matt (46:05.335)
like culture and that I worried about. There's a guy in Malta at the moment who's doing some work to try and make Maltese like spoken language data sets because they don't really exist like at all. And that's like a big ish problem because like that's that whole like you have customer support bots, have like hotels like with the customer support agents and like.

Wilhelm (46:15.67)
Right. Yep. Totally.

Matt (46:31.286)
that will all be done in English because technically English is a national language of Malta, but they'll lose that spoken Maltese over time. And that's like a small thing.

Wilhelm (46:36.437)
Mm-hmm.

That sounds like a smart thing to do from that guy. There was a, like I think it was the Dworkash Gwerne interview, like maybe last year or 2024. And I think he said something like, yeah, we should all be like riding more and putting it out there more because by doing that, by putting our own thoughts and thinking out into the world, like you can influence the LLMs even just a little bit. And...

Matt (46:48.3)
Yeah.

Matt (47:00.332)
You influence, hugely. Dude, even putting your thoughts on Twitter, you can influence whole companies. Like, having a legacy so lasting that you influence like generations and generations of language models and Gwen's definitely done that.

Wilhelm (47:15.37)
Totally, yeah. And he goes even further and says, like, if you don't do that, if you don't put your thinking and writing out there, then you're sort of like...

submitting to the LLM. Like you choose not to steer, you're happy to take a backseat, which is kind of interesting. So I think that's like one actionable thing one can do. But I mean, the models, they do speak other languages too, right? Or like, isn't there these stories about like DeepSeek sometimes goes and thinks in Chinese because it's like more efficient or something.

Matt (47:46.275)
Yeah, I don't know. Other models, especially the deep sea models and the Chinese LLMs, they're not going to have a problem with this because they have such a vast collection of training data in Chinese. It's going to be fine. I think about this with companies as well. Maybe it's a good segue to talk about the Tailwind saga.

Wilhelm (48:05.012)
Yeah, yeah, yeah, yeah. I was kidding. I have that on my list. Let's do it.

Matt (48:09.858)
Well, like for people who don't know, like Tailwind is a front end library that basically, as far as I'm concerned, fixed CSS on the internet, which is like styling on the internet. And a lot of like true design, like yeah, like a lot of people, like front end is a very weird and wonderful place where people have very, very strong opinions. But as someone who doesn't really care that much, Tailwind has been great.

Wilhelm (48:24.828)
Agree.

Matt (48:38.978)
has the utility to let me just get out the way and let me style things as well as I possibly can, you know, like it just has all the bits. And so it was kind of, yeah.

Wilhelm (48:42.409)
Yeah, huge.

Wilhelm (48:54.441)
And Adam Wathen, should say, he's one of my all time heroes. It's incredible to have made such an incredible contribution to the world. Every single, or so many bloody websites, especially new ones, use Tailwind. Like, chatgpt.com uses Tailwind. The Versel website uses... Yeah, basically... Right.

Matt (49:12.28)
Well, everything that's built now, because anything that uses Shad Cien and React, everything that uses, like, that comes out of v0 uses Shad Cien and React. And Shad Cien is entirely built with Tailwind. So, yeah, like...

Wilhelm (49:26.037)
Yep. it just, the tailwind approach to styling just, I think fits much better to how developers think about styling. And it's a, a, yeah, it, me, I agree. It made me a lot more productive. It feels like it fixed something broken with like the CSS approach.

Matt (49:41.579)
I think it goes very well with, like we can stop shouting about how great Tailwind is, but it goes very well with like, as far as I'm concerned, the React way of thinking with like components and like encapsulating particular pieces in certain places. I can see how like CSS and generally like having style sheets for like when you had massive JavaScript files and massive HTML files, that kind of makes sense almost.

But like when you have the react point of you then like tailwind, it's just so nice. But maybe we can talk about like what happened. Do you want to explain?

Wilhelm (50:19.733)
Yeah, so I think I have a decent overview of this. So Tailwind is an open source project, but there is a team behind it. The team until recently, I think, was like seven or eight people or something like that. And then the way they fund that is they have various paid components to...

that they have various paid components as kind of part of their offering. So the core of Tailwind and all of its features are very much open source. But what they've done in various ways is essentially have these component libraries or pre-made Tailwind websites that you could purchase. And that's how they were funding the development of the whole effort and paying those like eight people's salaries.

So nothing in those paid component libraries, you couldn't recreate yourself by just using Tailwind very well. But this is like by the Tailwind designers, or the Tailwind creators who created the whole framework, who are incredible designers, who know how all the different Tailwind components should fit together. And they would then hand you this like ready-made component library where you can speed up even more for making your own website.

those and I've actually kind of, think I've bought all of them over the years. They've, in my mind, they were always quite affordable. And they were also, which I think is quite a meaningful part to this whole story. They were always a sort of one time purchase thing. So for like 200, 300, $500, you could like,

have a one-time purchase to buy all of the stuff, all of these components, and then use them like in perpetuity forever in your own website. And I should say, sorry, also crucial part, the way they would advertise their own paid offerings, like these paid component kits, was through the Tailwind docs.

Wilhelm (52:27.367)
So in the pre-LLM world, the way you use Tailwind is like, you're like, I've learned about Tailwind. It's great. Let me make my website using Tailwind. Something, I need to look up how something in Tailwind works. Let me go to the docs. Let me search for it. Let me see what the pattern is. Let me see how I like, I don't know, make something underline or how does like CSS grid work in Tailwind or how do I make, like use one of the color schemes. You would always go to the docs. And then in the Tailwind docs, you would,

among the information you're looking for also see an ad for their paid offering which obviously might even be what you're looking for but either way like it was a link to the paid offering. Now what's happened with LLMs is LLMs all have Tailwind adjusted they generate the components for you in Claude in

Chachi BT, wherever, you no longer have to look to the docs or have to look at the docs way less often. And then you also, as a result, never see the ad for Tailwinds own paid offerings. So I think their website traffic is down. What was it like 80 % or something like quite substantial.

Matt (53:30.112)
Yeah, but I think there's more to it than that even. So their web traffic is down, or I don't know if was there like ARR was down, but they were still getting...

Wilhelm (53:40.94)
Well, that's the thing, they don't have ARR. This is my main point with this, that it's not recurring. It's all one-time purchase, which is insane.

Matt (53:44.225)
Yeah.

Matt (53:51.725)
think what, what, okay, and this all came to a head when someone tried to make a pull request to the Tailwind docs to expose an LLMs.txt, which is just a simple file which collects like all of the, well, I think this one was actually in the style of like an LLMs full. So it was every single docs page in Markdown just exposed on that endpoint.

And so what people could do here would they they could load that into like various search MCP servers, they could load that like they could just directly give that to an LLM, let it search through, let it like discover all this stuff. And these are pages that are available publicly anyway. But it's just an easier format and markdown for the for an AI agent to like, to grab and it's like a very common pattern. Like most every Mintlify doc site, for instance, serves in LLMs.txt.

as a feature, like, Mintlify shout about that as a feature. If you go to any Mintlify docs site and you just do .md on any of the pages, you will end up with a markdown file. And if you go to the root and do lms.txt, you will get the lms.txt. If you do lms-full, you will get the full lms, optimized markdown of the whole docs site. And that's like a feature. what...

someone pushed a PR to do that. Kind of rudely, I'm not gonna lie. Like the PR thread is a bit weird. He's also like advertising his own like TikTok video as well. It's like kind of strange. But this guy did make the change and the change looked like it worked. I didn't test or anything. And after sort of like 25 odd comments, it kind of got.

Wilhelm (55:25.523)
Yeah.

Wilhelm (55:29.619)
Ha

Matt (55:44.801)
shut down and closed by Adam the maintainer, one of the maintainers saying that they had to lay off 75 % of their staff, their product wasn't making money. And one of the main reasons for this was the LLMs could grab their docs so easily. And actually this was something that they wanted to move to a paid offering was the ability for LLMs to grab their docs. Yeah.

Wilhelm (56:10.302)
So the thing that's insane, right, is that like more Tailwind is being used than ever before. Tailwind is more popular than ever before. We're all making way more software than we used to. We're all using Tailwind way more than we used to. But because of the way their business model works,

they're making less and less money. And Adam published this very raw, like 30 minute podcast, just talking to himself, him going for a walk, running the numbers over Christmas and basically realizing, wow, we have like six months left to live as a business. We need to like lay off now in order to survive. Otherwise, like, and try to turn this around. So that's just this like stark kind of reality.

Matt (56:54.026)
It's insane. They were getting a hundred million clicks on their site a month. No, sorry. They were getting a million clicks on their site a month. Like it's insane that they can't monetize this. And I think actually what's killed them, which is really wild is ShadCN. Like ShadCN is pre-made components in Tailwind that use Tailwind, but they're free. You basically copy and paste them into your own site.

Wilhelm (57:13.362)
Hmm. Interesting. Yeah. Free for anyone.

Matt (57:22.882)
There's a whole CLI for doing that really nicely and easily. They have theming packs that sort of like fix some of the more rough edges of Tailwind. It just fixes it and makes it a cohesive product, but it's free. It's made by Versel. It's like completely free. Like it's open source. It's like a gift to the community. And because that is free and it uses it, it's like Tailwind is sort of like an integral part of

this ecosystem. So I think Vercell have two choices now. And this must have been chatted about previously. Vercell have a choice to make an acquisition, which they often do of these types of teams that they make an acquisition.

Wilhelm (58:10.177)
I don't- I don't think Adam would go for- would go for an acquisition.

Matt (58:14.102)
No, well then Adam needs to find a better way of making cash because selling code is not basically not going to fly in this AI era. Like that is, that is that I think for me, that is the lesson is you will not make money selling code anymore. Like that, that boat has gone.

Wilhelm (58:18.686)
Yeah.

Wilhelm (58:23.453)
Yeah.

Wilhelm (58:28.594)
It's interesting because they, to me, the way Adam and the Tailwind team have monetized has always seemed a little bit like playing the game on hard mode.

because it is this one-time purchase thing. I think people who have not been in, done this themselves, it's a bit hard to appreciate, but recurring revenue is an insanely good business model that unlocks so much. If SimplePall had always been one-time purchase as opposed to recurring revenue, I don't think I ever could have quit my job. I would have never made more than my salary. So.

And then obviously Tailwind has always been like one-time purchase and not recurring revenue, even though we all get like so much value for it. We probably all would like just want to pay for it. It doesn't even have to be like in a very smart business model, like just a way to pay for Tailwind on a recurring basis and get access to some stuff that you don't get access to for free. think so many people would jump on that. And then a way I've always admired the Tailwind team for choosing the like really hard path and still make so much money. Like I think they had like

over two, maybe $3 million of annual revenue based on this model, which is like so, so, so impressive. Like to get to the numbers that high on a non-recurring model, like that is a really, really incredible entrepreneurial accomplishment. But I do think it's playing this game on hard mode and figuring like embracing a recurring model will like make them a lot more money. So I hope they, yeah.

Matt (59:35.724)
Jai

Wilhelm (59:59.642)
I would imagine they figure out something on that front. obviously a lot of people are kind of coming to the Tailwind team with like offering help, offering assistance, offering more sponsorship because it is just like such a beloved like part of our lives.

Matt (01:00:18.38)
Yeah, I think it's just wild that... I get what you mean about hard mode. It's kind of like the indie hacker of 2012, 2015 mode where like, or maybe even later, maybe 2020. Like, I guess like the last person that I saw kind of doing this who got rinsed was Mark, I don't know his last name, but it begins with an L, Louvin, Mark, like...

Wilhelm (01:00:35.027)
Hahaha

Matt (01:00:48.076)
The guy that made ship fast. Ship fast?

Wilhelm (01:00:48.219)
Okay.

I don't know if I know it.

Matt (01:00:53.462)
And it was like, it's basically a template for your SaaS app. And it's like 40 bucks or whatever. And you buy this and it's connected to Stripe already. like, you can literally just sell whatever from this. Like it's just like a template. And the idea being that everything is spun up already or not even spun up, but you can just then deploy this to best sale and you're done or something along those lines.

Wilhelm (01:00:57.948)
Alright.

Wilhelm (01:01:04.837)
I see,

Wilhelm (01:01:20.061)
Yep, yep, yep, yep, yep.

Matt (01:01:23.608)
I just don't think that flies anymore when you can take an app that you don't have any of the source code for and just point a model at it with some screenshots and be like, based on this minify JavaScript bundle, can you build me the whole app, please? And I mean, you see like the Ralph guy building his whole software development lifecycle from scratch, like including GitHub and everything and like...

Wilhelm (01:01:41.191)
Yeah, yeah, yeah, yeah.

Wilhelm (01:01:50.663)
Wait, which guy?

Matt (01:01:53.045)
Jeffrey whatever his name is. Yeah, Jeffrey Humley. Yeah, I don't know. Speaking of which, everyone, there's a lot of people got very excited about get this over the holidays, me included. yeah.

Wilhelm (01:01:54.636)
Jeffrey Hunt. Huntley?

Wilhelm (01:02:01.201)
Yeah, we should chat about this.

Wilhelm (01:02:08.197)
Yeah, yeah, wait, yeah, let's chat about all this. Yeah, any closing thoughts on the tailwind stuff? I mean, I think that there'll be a way to figure this out. Like, yeah, I think it's an incredible product. And now, you know, they just need to figure out the business model.

Matt (01:02:19.714)
Yeah, it's really, yeah, know, you can't help but like do a slight, I know that comparison is like the anime and everything, but like a slight comparison to to Bun, where this is what Bun could have been in sort of like five or six years had it not, had they not looked for an acquisition.

Wilhelm (01:02:33.939)
Mm.

Wilhelm (01:02:41.191)
But Bum never made any money.

Matt (01:02:43.39)
No, no, no, no, of course not. As soon as you, you know that Silicon Valley thing where, it's not Silicon Valley. It is Silicon Valley. Yeah, the TV show where it's like, you can't show revenue. As soon as you show revenue, you are fucked, because they'll want more and more and more and more.

Wilhelm (01:02:48.967)
Yeah, yeah, yeah. No, no, it is, is, yeah, I don't know, man. I feel like they're quite... But like, Tailwood never raised any money. They... I don't know. I don't know. I feel like they're...

Matt (01:03:07.586)
Yeah, no, you're not wrong. They're very different. They're like hugely different. But they're both free dev tools, which are like becoming more more and more used and seem to be very native to to AI, right? And you see one of them getting acquired for a princely sum and the other struggling to survive. And it's like what...

Wilhelm (01:03:13.928)
You

Wilhelm (01:03:18.033)
Yeah, it's true.

Wilhelm (01:03:31.579)
Yeah. No, for sure.

Matt (01:03:34.86)
Like you can't help but make comparisons. Like what is the difference between these two things? And yeah, one of them was on hard mode, definitely. One of them was on like an indie hack here. We're gonna build this like very sustainably, like for ourselves. We're gonna charge for our skills. We're gonna try and like basically, we're gonna try and let everyone have a piece of our skills for a very like affordable price. And the other was like, we're not gonna charge anyone for anything.

Wilhelm (01:03:40.007)
Yeah, yeah, yeah.

Matt (01:04:03.779)
We're just going to build really cool tech and hope that one day someone sees value in it.

Wilhelm (01:04:09.552)
Yep. Yep. Yep. Yep. No, I think they'll figure it out. I think Adam is an absolute legend. Such a role model. Making something like Tailwind, just creating something like that for the world, like that's just so, yeah, it's just legendary.

Matt (01:04:21.88)
So cool. Yeah, so cool.

always makes me laugh when people get like, I think the, you know, Ricky from React, he has like these odd fights with people on Twitter. And it's just like, dude, like, and Adam Wolf as well, people have like random fights with these two people on Twitter. And it's like, guys, just like lay off. Like what these guys did for software engineering and like generally you all Muppet programmers who are now able to spin up websites, like there's no tomorrow.

Wilhelm (01:04:36.627)
yeah.

Hahaha

Wilhelm (01:04:48.423)
Mm-hmm.

Matt (01:04:55.576)
It's them, guys. It's them. It's always been them. Just lay off, guys. Just fucking go find someone else to annoy. Yeah. I think they do deserve some gratitude as well as a bit of admiration.

Wilhelm (01:05:05.874)
That's funny.

Wilhelm (01:05:11.812)
Agreed. All right, should we talk about, well, Ralph, Gastown, and Git?

Matt (01:05:19.376)
wow. Okay, let's,

Wilhelm (01:05:22.182)
What do you want to talk about most?

Matt (01:05:24.44)
what do I want to talk about most? Well, I want to talk about my Git project.

Wilhelm (01:05:27.378)
I probably need to run in like five ish minutes. Yeah, yeah, let's do that.

Matt (01:05:32.194)
Okay, okay. Just before Christmas, Monday before Christmas, we had a demo morning for the agents team and holy shit, Cloudflare, and holy shit was it cool. There was three of us who'd built something and then Sunil came along to watch and it was completely impromptu. We were actually just on the call because it had been canceled and we didn't realize it been canceled. And so we were just like, fuck it, who's got anything to demo? And so we demoed.

Wilhelm (01:05:56.658)
Hmm.

Nice.

Matt (01:05:59.639)
Like it was the coolest demo series I think I've ever seen. I can actually share some of it now because I think Steve just published his thing. So we'll talk about his. He made like an agent system, like deep agent system where you can connect multiple agents. You know in core code, you have multiple agents. They're all defined through JSON files. He's made this so that you just push that JSON file to like to a S3 bucket and then

Wilhelm (01:06:21.626)
Right, yep.

Matt (01:06:29.322)
all of the agents exist on Cloudflare as pieces of infrastructure. So that lasts forever, they're all durable objects and they last forever. And now you can do things where these individual agents can talk to each other and you can ask something or you can create like schedules for something on like a main agent or on a specialist agent. And it has the ability to like basically negotiate the swarm.

Wilhelm (01:06:34.354)
Mmm.

Matt (01:06:57.57)
to get the answer back or to do the thing or to engage with the API or to call an MCP server or to do whatever. he's, it is the most wild sort of multi-agent architecture I've seen. Everyone is gonna copy it because it actually works and it's sick. And it doesn't rely on like a lot of stateful durable execution and things like that. And it doesn't rely on something like temporal to run.

Wilhelm (01:07:02.747)
interesting.

Matt (01:07:26.968)
It's called Agent Hub, I think.

Wilhelm (01:07:28.178)
What's it called?

Matt (01:07:34.592)
Yeah, and Steve's GitHub name is deathbyknowledge.

Wilhelm (01:07:40.722)
Death by knowledge. Okay, cool.

Matt (01:07:43.16)
Yeah. Death.

I'll find

Wilhelm (01:07:48.041)
and talk about what you built.

Wilhelm (01:07:54.128)
haha.

Wilhelm (01:08:14.251)
yeah.

Matt (01:08:16.148)
not in the terminal, you can have that running as well. So you can like open that on your phone and like message your sandbox that's running in the background, just like going away, doing its thing, maybe ralthing. And like your sandbox can just be going forever. And yeah, like, like it is so cool and it's so slick and so easy to set up. And yeah, Neresh has absolutely killed it with the sandboxing recently. Cause like the networking from the sandboxing is actually the hard thing and making that feel easy.

Wilhelm (01:08:16.562)
Mm-hmm.

Wilhelm (01:08:29.872)
That's very cool, yeah.

Wilhelm (01:08:43.29)
Yeah. Interesting.

Matt (01:08:46.456)
Yeah, so you killed that. And then, right, those are the two you con.

Wilhelm (01:08:48.45)
I random side note on this because we didn't have to talk about this, but one of the things that's really been popping off recently is Claude bot, with a W C-L-A-W. Steinbergers, Steinbergers. Yeah. I yesterday bought myself a Mac mini just so that I can have it running nonstop like at home running Claude, because it's definitely nicer to use.

Matt (01:08:58.37)
Yeah, that's Peter, Peter, I can't say his last name. Yeah, that guy.

Matt (01:09:07.298)
Did you do it? You did it! Wow, that's cool.

Wilhelm (01:09:17.014)
Claude code, like locally than the web version, the web version, you just don't have much, like, it's just a bit, like annoying, friction full or whatever. and, but it sucks. Like I've had so many moments where like Claude was running and we were like somewhere and I had to close my laptop and leave. I'm like, I don't want you to stop working Claude. So now I can just, we'll be able to shell into the home Mac mini and, have Claude can keep running there. Anyway, back, back to you.

Matt (01:09:47.107)
Yeah, no, so that's of like handing off between Terminal and Sandbox and all that stuff we're like thinking about a lot, which kind of brings me to the next thing, which is what I built, which is called Zaggy. And it's kind of trying to be a better Git. And it's like a feature. I don't know if it's cool yet, but it's like a feature. Yeah.

Wilhelm (01:10:08.625)
It's so cool.

I need to start using it.

Matt (01:10:14.314)
It is pretty, I'm trying to ship tasks today. So I'll tell you how that works in a second. But basically it is Git. So it's the Git CLI just aliased to something called Zaggy. And then I actually alias it to Git locally. So when I'm running Git, I'm actually running Zaggy. And I've implemented all of the popular commands like status, log, commit, push.

Merge, a bunch of others I've implemented. Any that I haven't implemented, they go and just use system git anyway. And all the commands I've implemented are somewhere between one and a half and two and a half X faster. They use like half the amount of tokens on average, which is like, they're just like kind of minor things, but they're really cool things. They're just, they were like the impetus for doing it. And now like I realized that...

Wilhelm (01:11:02.522)
Mm-hmm.

Matt (01:11:09.376)
If you wrap Git, you can enable so much cool shit. Like there are so many APIs in Git that no one uses. Have you ever heard of Git notes?

Wilhelm (01:11:15.983)
Yeah. Only because you told me about it a few weeks ago.

Matt (01:11:20.588)
Yeah, so Git notes is an API that you can add some metadata to any commit and you can collect them and then squash them and then add them to a singular commit, add them to a PR, add them to loads of other stuff. And I have no idea where you see them in GitHub. Apparently you can, call into Claude, but I have no idea where you see them. But they are, they're cool. They're just like another piece of information you can add, but you can add it in a structured way. So for instance, I added this...

Wilhelm (01:11:38.834)
really?

Wilhelm (01:11:46.801)
Mmm.

Matt (01:11:50.061)
this option to Zaggy where whenever I do git commit, if I am in an agent, if the CORDCODE variable equals one or the OPENCODE variable equals one, or I've set the Zaggy agent variable to be one of those, it will enforce that one, I can't do any horrible commands like force push or check out.

over my local staff or like do anything like that that agents like really love doing that's like pretty like breaking. So I can like stop all of that like in its tracks because I can just like disallow that. But I can also add some really clever auditing stuff. So when you do a commit, I have a dash dash prompt and that instructs the agent to include the like verbatim prompt of the user that the user made.

Wilhelm (01:12:24.411)
Ha ha ha.

Wilhelm (01:12:35.441)
Mmm.

Wilhelm (01:12:44.771)
Hmm, nice.

Matt (01:12:46.88)
in order to get that outcome. So what you get now is you get the user's initial request, you get the full commit message and everything, and then you get the code. And then from that, you have a more of a cohesive thing to PR. There's also another feature I'm working on at the moment where you'll also get the whole transcript of what you talked about. And that will be included as a Git note in the PR.

Wilhelm (01:13:14.267)
Cool. Yeah. So can you have multiple notes per commit or how does it work? Damn. That's so, that's very cool.

Matt (01:13:16.417)
And so like...

Yeah, Yeah,

Wilhelm (01:13:36.209)
Wilhelm (01:13:39.663)
Nice.

Wilhelm (01:13:43.599)
Yep.

Matt (01:13:47.775)
into Claude code. Fuck that, man. Just use Git. I may build in the Git APIs that no one ever wanted to use, that agents would never use before. So now I just say, write a plan out with Claude, and then I'm like, Claude, could you just convert this plan to tasks, to Git tasks? And Claude goes, yeah, sure I can, having no idea what Git tasks are, but just knowing that Git is a thing that people trust and love. And so it writes.

Wilhelm (01:13:55.533)
Yeah, yeah.

Matt (01:14:17.324)
Git, it writes out, git tasks dash help, or whatever it writes out. My CLI goes, this is how you use Git tasks. Git tasks add task number, and then description. That adds a task. Git tasks delete. The agent can't actually delete tasks. It can only add or append to tasks. And then, so the agent writes out all your tasks. Super, super cool. And then I just do git, and then I'm like, Claude, could you like, git agent run?

Wilhelm (01:14:35.856)
Mm.

Matt (01:14:47.56)
over my tasks. And this is where it gets really funky. So now I start an inner loop of Claude, basically like an inner instance of Claude over each one of those tasks. And that my friends is Ralph.

Wilhelm (01:14:50.225)
got it. Yeah.

Wilhelm (01:15:02.799)
Nice, yeah, that's smarter than Ralph almost, I feel like.

Matt (01:15:04.076)
That is what Ralph is.

No, is, it's the actual implementation of Ralph that Jeffrey was doing with bash scripts. I can do inside a git CLI.

Wilhelm (01:15:16.163)
interesting. Okay, wow.

Matt (01:15:17.952)
And so, and so like all of the stuff that, like Ralph is pretty complicated to set up. Like, it's, it says that you have a big old plan and you have to format your plan in a certain way so that each individual piece can be, can be taken out of context. That's how Ralph works. Right. You have this big old. Okay. I'll explain how Ralph works basically. So, so Ralph, you create a big old plan and each piece has to be like a, a, a chronological ordered, task, right?

Wilhelm (01:15:30.788)
Wilhelm (01:15:35.567)
I thought, okay, interesting. Yeah.

Matt (01:15:47.329)
And then you give the whole plan to Ralph and you say, here is a plan. Work on the next task. If there are any structural things that we didn't know about, then add that to agents.md or called.md and keep a running ledger of structural things that we didn't know about. But basically work on that next task. Only do the next task. Run all your tests. Create tests for it. And then once you're finished, write out this.

Wilhelm (01:15:54.629)
Yeah.

Matt (01:16:14.84)
particular phrase like I am done everything passes I haven't that's Ralph and then you just run that in a loop over your plan

Wilhelm (01:16:24.558)
Right, right, right. Yeah, yeah, yeah. Wait, so this might be slightly different to how the Ralph plugin Claude code works because...

Matt (01:16:30.284)
No, the Ralph plugin works the same. just works without having an outer loop and an inner loop. It just uses the outer loop and iterates over the outer loop. But it actually works the same. Having an inner loop is harder to set up, but it's much better once you've got it set up because now the outer loop, I can still chat to Claude and be like, Claude, like where are we on our tasks? And Claude goes,

Wilhelm (01:16:36.825)
Right.

Yeah, yeah, right, exactly. That's what I thought, Raph.

Wilhelm (01:16:45.721)
Yeah, exactly.

Wilhelm (01:16:50.883)
I see, I see.

Wilhelm (01:16:54.617)
Yeah.

Matt (01:16:56.824)
Get task list and then it says the next task that is it's working on it's like Ralph is working on this task Sweet, and then you can be like alright like can we think about structuring like the tasks that we have like Maybe we might want to split some of them up into streams Maybe we might want to like so we can have parallel Ralph's maybe we might want to like add some more things to the end Can we think of things that more cohesively and then we can do this whole like structural analysis of like our code base?

Wilhelm (01:16:57.434)
Yep.

Wilhelm (01:17:04.58)
Yep, makes sense.

Yeah.

Wilhelm (01:17:19.396)
Got it.

Wilhelm (01:17:23.632)
Yep.

Matt (01:17:25.632)
and like thinking about new features in without a Claude the whole time, Ralph is just chugging through the tasks in the backlog.

Wilhelm (01:17:34.756)
Got it. Okay, I think I understand. Yep. And obviously what you describe is much, much, much simpler than Gastown, which is also, I think, trying to just keep things moving constantly. But it has all these other, it has these other, the deacon and the mayor and all these roles. But okay, that's cool. All right, so.

Matt (01:17:44.492)
Yeah, this is the simplest way of doing it.

Matt (01:17:51.383)
Yeah, yeah, yeah. And if there's, yeah, if there's one thing I know from this age of LLMs is if you keep it simple and everything collapses into the model, at this point, Ralph could be running over anything. Like I literally, you can pass any command you want as the Ralph agent. And so I could pass open code. I could pass like whatever I want. Like if my internal company builds something, I could pass whatever. And I just need to ask an outer agent to call get agent.

Wilhelm (01:18:10.98)
Right. Yep.

Matt (01:18:21.312)
Run. And that's it. Done.

Wilhelm (01:18:24.89)
So if I or someone else listening wants to try out Zaggy, what's the like first thing we should try to do?

Matt (01:18:30.584)
Go to GitHub, go to matzkary forward slash aggie. There's an install script. You install it. You set up like an alias to Git and then just go about your day. Like nothing will break. You'll everything. Just start using it. And then like two weeks later when you've like kind of forgotten you've installed it and you're aliasing some random other person's software to Git, then be like,

Wilhelm (01:18:44.29)
Okay. So don't try to use anything so they just start using it.

Matt (01:18:58.56)
Maybe I might want to use the Zaggy task tracking system rather than the Claude task tracking system. Or maybe, there's some other real niceties. If you make a call load plan, you can do git task import and then plan.md and it will convert all of it to git tasks from your plan.md. So that's kind cool. You can also do like,

like git plan PR and it will write out a description of like every task that was done and like where you're at and like yeah and all like everything that was done and the commits that related to which task and all of that stuff so like initially just like start playing with it and you won't even notice you won't even notice you're using it and if you do it will only be because your checkout and your commits and everything just got faster that'll be the only that should be the only reason you notice it

Wilhelm (01:19:45.882)
Just start using it, okay cool. Yeah, that's.

Wilhelm (01:19:54.82)
Yeah, yeah, yeah, nice.

Matt (01:19:58.521)
And then, yeah, then just like start, yeah, just get started playing with it. And it's one of those things that you should just use by, like, I think I'm gonna, I use it, I've used it by default for the past two and a half weeks, three weeks, and I will continue to use it by default and only use it basically. Yeah, nothing else. And there's so much fun stuff you can add when you control Git. Like, Git is an amazing, amazing protocol let down by a really, really, really shit CLI that's so old.

Wilhelm (01:20:21.05)
Totally.

Wilhelm (01:20:27.064)
A bad UX. Yeah, this thing, this like pattern that you make use of in Zagi where like you change the output of Git to be like better and have additional instructions to the agent. I don't know if it has a name, but in my head, I've called this like intentional prompt injection in the past. I think it's so powerful.

Matt (01:20:28.79)
Yeah.

Matt (01:20:44.204)
Yeah, yeah, yeah. I've called it recipes before, but yeah.

Wilhelm (01:20:47.502)
Because especially because there was a whole thing going around Twitter just before Christmas with Torsten. Well, I think one thing that people don't understand is like, if you just give an agent a tool, it will basically never use it unless you specifically ask for it in the user message. Like if you give it like a refactoring tool or whatever, like it won't use it just by itself. It will only use it's like trained on in-distribution tools. And I think this has been like, this has been like the bina mic systems.

2025 trying to get Cloud Code to use colo like without someone intentionally asking for it. But the way around it is this pattern, this like intentional prompt injection where you get the, you get to put the instructions in, a thing where the agent's already trying to use an existing tool.

Matt (01:21:26.05)
Yeah.

Matt (01:21:32.822)
Yeah. So the good thing about Git is it's like the most commonly used CLI ever. And so by injecting yourself inside of Git, you can be used and so you can have that like platform to use more features. So I think basically what you're alluding to is when in Cloud Code, like a bunch of the prompt says, when you want to commit something first, you should do log to see like

Or first you should do status to check which branch you're on. Then you should do log to see what previous commit messages look like. And then you should commit. If you run log on most repos, you will blow up a context window. And so they have to add in the particular flags that you need to add for log and all of that sort of shit. So that stuff you can sort of mitigate quite easily by making an agent-first design.

Wilhelm (01:22:07.511)
Right, right, right, right. Yep.

Wilhelm (01:22:15.213)
Yep.

Matt (01:22:28.344)
And by keeping the interface the same, like the agent doesn't have to learn anything new. I think for your use case, it feels like hooks is going to be like the beauty, the beautiful thing. But I really want to start like, yeah, I really, I know. I really want to like stay away from hooks as much as possible because I heard about how like, have you tried to uninstall beads?

Wilhelm (01:22:51.535)
I've never tried to install it.

Matt (01:22:53.74)
No, you never tried to install it. Okay. Yeah. Well, yeah. So I've just been seeing on Twitter the last few days about people actually trying to uninstall beads and it like connects, it goes so deep into your like hooks, into all of your coding agents. It just like injects itself and like sets up stuff in random config files. like, I don't want to be like that. That's, I want one line removed basically and a folder, one line in your, your, in your,

Wilhelm (01:22:55.933)
Correct. So I've not had this problem.

Wilhelm (01:23:08.214)
Yeah, yeah.

Wilhelm (01:23:16.823)
Yep. Yeah, yeah, yeah.

Matt (01:23:22.2)
ZSH config file or your shell config and one binary removed in your system and Zaggy should be gone. Like Zaggy is not a virus.

Wilhelm (01:23:31.457)
Yeah, I think there was like a 200 line bash script going around and like a GitHub just to help you uninstall beads. That's yeah, that's a lot.

Matt (01:23:38.85)
Yeah. Yeah. That feels like an anti-pan. That feels awful.

Wilhelm (01:23:44.495)
It does, Okay, man, I have to run. Bloody hell, so much to talk about. I think we can be back, yeah, mean, obviously we've had a bit of a break over Christmas, but things are moving. It's gonna be an exciting year. mean, there's so much happening.

Matt (01:23:47.308)
Yeah, we have to go.

It's been sick, dude.

Matt (01:24:00.674)
Yeah, it's gonna be cool. And I'm get the last couple of episodes out. I don't know if I should just publish this one or just get my act together and publish the previous ones. I might just publish the previous ones this week and then publish this one next week. Would be kinda cool.

Wilhelm (01:24:15.767)
I was just thinking about this. I want to publish this one like today because I think this stuff just goes out of date so quickly. Let's do it. This is the new studio. Yeah, yeah. So we can both, we can both publish it.

Matt (01:24:18.806)
Okay, good. Publish it then, publish this. Yeah, because this one's under, this is under our new studio, isn't it? Whoa. Yeah, okay. Yeah. You get this one done and I'll get the other ones done and we will, teamwork makes the dream work, man. It's lovely to see you.

Wilhelm (01:24:32.985)
Sweet.

Hell yeah. Happy New Year. Let's get hacking. Sweet, bye.

Matt (01:24:38.233)
Happy New Year, Bossman. Bye! Can I say hi?