WEBVTT 0:00:15.360 --> 0:00:21.360 Programming throwdown episode 186 becoming a manager Take it away, Jason. 0:00 --> 0:00:30.030 Hey everybody. This is one last programming throwdown of the year. Pretty wild. 0:00:32.270 --> 0:00:41.790 I, I got a plastic welding kit a while back and I actually. Okay, so check this out. So I had about three days. 0:00:43.550 --> 0:01:46.670 Alone. So. So we talked about this a little bit before the show. So my wife and her mom and the kids all went to Dallas. And so I was in Austin. My, myself, I did a ton of cleaning up around the house and getting things in order and all that. And one of the things I did is i3d printed a thing to hold the keys. So it's just. Imagine it's just like a little a frame and on the top of the A frame there's hooks that come down and you can put the keys on the hooks. Pretty simple. But it came in, in several parts and you're, you're supposed to like friction fit it together. But my 3D printer is like not that great. And so, you know, it didn't, it didn't fit that tightly. And so I was like, oh, you know, I don't want to throw this out because it's pretty big. Let me just see if I can sort of heat it up because I've heard of people doing this. Heating the plastic and then it just melts. And then you melt it, you weld it, right? You melt the two pieces of plastic together and then when you stick them together, they're sort of like chemically joined or there. I'm probably not saying this right. 0:01:48.270 --> 0:03:20.380 So I pulled out this plastic welding kit and it's basically like a gun, kind of a glue gun. And when you press it, it's just the end gets super, super hot. So in that sense it's like kind of like a soldiering iron, which is like really powerful. I feel like it's way more powerful than a soldiering soldering iron. But I could be wrong about that. And the end effector is, is like this flat plate. So, so you can kind of rest them and it just heats that area up. And so I was able to do kind of like a side to side motion over the gap first. I clamped the parts together, right. I did like a little side to side motion over the gap and it, it, it melted it. It totally melted it. I didn't do a good enough job that you can't see the seat. So you know, I definitely also probably this would work a lot better if it was a hundred percent infill. You know, in this case it's not. And you Know, the grain doesn't line up right, et cetera, et cetera. But the important thing is, like, structurally, it works. Yeah, I wouldn't throw it on the ground or anything, but, like, it held everything together. And that part of it was kind of cool. So it also. The kit I bought comes with pieces of metal, and you're supposed to push on the metal. Think of it as like a staple, but it's like a W shape, and you push on the staple. And because, you know, metal transfers heat so well, the whole staple gets super hot, and then the whole staple kind of melts into the plastic. And so now you have a piece of metal kind of holding it together. 0:03:22.140 --> 0:03:50.450 I didn't try that because I wanted to, you know, just my first time using it, but maybe I'll do that next. But the thing I took away from it is, like, yeah, plastic welding is pretty easy. And the kit was very cheap on Amazon, so it's pretty cool. I mean, if you ever have a broken piece of plastic. I actually have a piano key from a keyboard that we have that has plastic keys. I'm thinking about welding that together now that I've kind of got my feet under me. 0:03:52.290 --> 0:04:36.200 I have many thoughts. We didn't talk about this before, so I have so many thoughts. Okay, I'll try to be brief. I have never used. What you're describing sounds a lot like a soldering iron, but we've come to the same conclusion. Something similar. We have. My children have those 3D printing pens that you just, like, hold. And they are exactly like a 3D printer without the motion. So you just, you know, click a button and it. It squirts out hot plastic. So it's kind of like a hot glue gun, except it's temperature controlled for PLA or abs. And so we'll use that in a very similar. You can just use the tip to iron, but you can also, like, squirt plastic and then use that because it, you know, melts into it and is, I guess, actually like welding is or. Or soldering. 0:04:36.600 --> 0:04:36.960 Yeah. 0:04:36.960 --> 0:04:39.960 Where you're melting a. A similar metal. The other thing is. 0:04:42.360 --> 0:04:54.840 You said, like, solid info, but you can do more outer walls or lower walls or upper walls, depending on the orientation. When we've done this, we've found that. So you get a thicker shell, basically, so you don't accidentally cut into the hollow interior. 0:04 --> 0:05:02.920 If you're using pla, there is like a glue solvent, but it's really nasty. I wouldn't recommend. 0:05:04.820 --> 0:05:27.780 Abs and others can be stuck together with a solvent and Actually like weld. But then you can also buy super glue with an accelerant if you don't know. Super glue when you put it on normally takes like kind of a while to harden. But you can spritz like an accelerant which is some sort of catalyst and causes it to rapidly harden. Um, and it does weaken it, but only like a trivial amount. So that can also work really well. 0:05:28.900 --> 0:05:33.620 Then if you're super glue would be better than plastic welding. 0:05:34.260 --> 0:05:36.180 I don't. It depends. 0:05:36.180 --> 0:05:37.220 Why would it matter, right? 0:05:37.300 --> 0:07:13.750 If you have two flat surfaces that are reasonably large, I mean super glue is going to be amazing. But if you have like depending, right. Like if it's. If they meet narrowly or something, it could be difficult to kind of get enough super glue or super glue in there. Right. But the last thing is, and I was thinking about this a little right now we're recording it's, it's still close to the Thanksgiving holiday, so stuff's on sale but it goes on sale fairly regularly. But this came about earlier today. The like current generation of 3D printers is really good and reasonably cheap. Like 200, $300 for not to show but like Bambu Lab. So if you have a printer that isn't really good, like definitely consider like the A1 or A1 Mini. They're very cheap. And if you're using like personal thing, it's a huge step up from like the ender 3 style of like if you're using something like that and you haven't moved to a sort of next generation one, definitely consider if it's something that you do. I know 200, 300 is still kind of a lot of money but you know we have that newer class and it, the, the software is better. If you're really, really into open source, look for an open source one. But if, if you're sort of more flexible on sort of what you're doing. The Bambu Labs are pretty like just plug and play. Um, they have an app and you can literally my. It's been game changing for my kids. They just go in the app and select what they want and it just prints it. Um, and they don't have to go to a computer and slice it. You still can, but you don't have to. And so they're much more into 3D printing than they would have been otherwise because they can browse the app and just send it to the, to the printer. So. Okay, I have a lot of other thoughts but I'll just stop there. 0:07:13.830 --> 0:07:20.070 But yeah, I mean call to Action. You know, if you have a broken piece of plastic, you could try this out. It actually works surprisingly well. 0:07:21.610 --> 0:07:25.530 Soldering iron, plastic glue, hot glue. Fix your stuff. 0:07:26.490 --> 0:07:34.010 That's right. Yeah, we, we build everything, we, we code up everything, even destruction. 0:07:36.570 --> 0:07:37.450 Time for the news. 0:07:38.410 --> 0:08:15.620 All right. Okay, I have the first news. Parse bot. Turn any website into an API. This is like a bunch of people are doing this browser use, which is another company has a thing called Skills that does this. But this is pretty interesting. I mean the idea is they crawl a website and you're kind of counting on them, I guess, to crawl it really well and, and click on every single button and everything. But assuming that, that they do that once they've crawled it, then they keep kind of breadcrumbs. It's like, oh, for me to like. 0:08:18.340 --> 0:08:28.820 You know, for me to look at your bank balance, probably terrible example you have okay for, for me to check your Starbucks gift balance. 0:08:29.060 --> 0:08:29.780 There you go. 0:08:30.420 --> 0:08:48.060 You have to log in. And so, you know, here's like a recipe. Like the first time, I guess it's like using like a bunch of really heavy AI. It's like, here's a screenshot. What do I do? And the AI thinks a while, right? But then it kind of like leaves breadcrumbs and so. 0:08:49.820 --> 0:09:37.320 And so once it's sort of logged in, then it's like, okay, these are the things that AI typed. Next time, you know, we could just type them. Now of course, you know, if the website changes, you're busted. So you have to, you know, there's basically like, think of it as like a cache caching system. It's like I cached the flow to log into Starbucks. Now if Starbucks changes the website, that's kind of like a cash miss. And so you're gonna have to go back to the AI and the AI is gonna take a long time. But you know, Starbucks probably doesn't change your website that often. And so, and so hopefully you get a lot of cash hits. And so, you know, if, when it works, it's like super magical. It's like it just boom. Like you're logged into Starbucks, you know, in like a few minutes or something, or even less than that, like maybe 30 seconds. 0:09:38.840 --> 0:09:49.160 But yeah, I think this is really cool. You know, the web kind of like email, it's not going away. I don't think ChatGPT is going to delete the Internet or anything like that. 0:09:50.679 --> 0:10:02.520 The Internet's going to be here for, you know, as long as, as, as email has been around, which is decades and decades. So, so having AI that can like interact with the web seems super powerful. 0:10:06.010 --> 0:10:29.850 Yeah, I think the. I, yeah, I'm curious how that ends up working. There's a lot of stuff for anti scraping, I mean a lot of stuff is, you know, not maintained. I'm curious how well it handles, you know, stuff that I guess like people don't want you to scrape and so they do various things to try to detect or evade that or, you know, mitigate it, I guess. 0:10:31.280 --> 0:11:18.410 Yeah, yeah, I mean it's, it's like all the problems of scraping and then now other problems. So it's definitely not a trivial task. Which is, which is, I think one of the reasons why it's so interesting. My guess. So admittedly I haven't tried any of these. I have tried browser use and Sky Vern and a bunch of these tools that, that browse the web with AI. And it is kind of how I described initially, where it's like AI pauses, takes a screenshot, looks at all the HTML, thinks for a while and then clicks and then the whole thing repeats. And those actually surprisingly work terribly. You would think with all the time to think and the screenshots and everything that it would work well. 0:11:20.410 --> 0:11:25.230 But there's just so many things that, that, that don't work well. Like for example. 0:11:27.870 --> 0:11:39.390 You know, when you look at the HTML, you see all the things that you can't actually click on except. Oh, but you can, because if you use like JavaScript, you can click on. 0:11:39.390 --> 0:11:42.710 Yeah. When the mouse is in this hitbox, it dynamically change. 0:11:42.710 --> 0:12:39.760 Yeah, yeah, exactly. So like for example, a website might have a pop up and they're expecting you to like answer some questions. But you could write JavaScript to click a button behind the pop up and unless the website defends against that, it'll let you make that click and you could even browse the site behind the pop up. But what happens is now the website's really confused. And so I've seen like major websites like just crash because you're doing things that the developers, you know, didn't really expect. So, so I've tried a lot of these programs and my conclusion is that it's a super hard problem and now we just added this other layer of, you know, breadcrumbs and all of that. So, but, but these kind of problems are also the most interesting and lucrative if you can pull it off. So I think it's, it's a space to keep an eye on for sure. 0:12:42.160 --> 0:14:17.380 Yeah. Putting aside privacy for a minute and getting a little off topic from what you're saying, but, but I'll try to make it brief is I think the sort of like AI powered web browser in the sort of middle term, short term could be a really cool thing. And I don't think I've seen a good implementation of it yet. But things like you're saying like, here's my bank, my bank doesn't offer, you know, a good API, like searching at a bank is terrible for transactions. I know I made this or whatever. And so I want to log in so that I'm owning the credentials and I don't have to give them. But then I want you to like browse around and answer, like go through all the pages and just interact like a human would temporarily and like answer questions for me or hey, you know, I'm trying to find coupon codes for this website and like, can you go find coupon codes and try plugging them in and see if any of them are active or not? You know, an email search as an example still sucks horribly. So like something that could like, you know, not have to integrate with every single email provider, but just be like, hey, this, this is my inbox. Go through my inbox and find this thing like super, super useful. You know, I think there's tons of optimizations long term, hopefully we just ask and it happens. But midterm, before all those magic integrations, I, it does seem like this sort of adaptation layer, but I can't imagine building a business in this space. Like, I just don't feel it's defensible. I don't feel like it's gonna last long term. But props to people trying to, to kind of eke out a living doing that. 0:14:18.340 --> 0:14:23.940 Yeah, it's kind of, yeah, it's, you're kind of, it's interesting. I think if you. 0:14:26.100 --> 0:14:33.940 If you wait four or five years, all the big players will have something. And so you, you have kind of like. 0:14:35.490 --> 0:14:59.570 Like building a company around something like that is kind of like breaking into a bank or something. It's like, okay, you know, I have 60 seconds. So like, you know, get this company off the ground and get some kind of traction and get escape velocity and then, and then maybe you can take the money or take the user base and pivot it to doing something totally different. 0:15:01.020 --> 0:16:13.870 Speaking of AI companies, my next news topic is the recent release of Google's LLM Gemini 3. Or I guess it's more than elements multimodal. I don't know what they classify their AI. We'll just use that term, I guess. And there's a couple of really interesting things about Gemini of course. Like seems to have moved to state of the art forward. Benchmarks are just terrible in AI right now. It's just so subjective. Like it's very difficult. Lots of the benchmarks say they did better, but everyone's not really sure. Is it meaningful? Like it doesn't change the space. I think it's in my mind how far we've come in just a couple years. But now like the Turing Test is just like nobody thinks it means anything anymore. I'm not a big like. But it just seems these, these chatbots are like so good now that a lot of stuff that they were doing really bad a couple years ago, they don't do anymore. But yet it still hasn't like made the world stop turning. And so aside from that though it does seem like a big improvement. I don't think it's been formally announced but the, the sort of speculation on the Internet seems to be that they did training and, and they have been doing inference but now both on their TPU hardware as opposed to gpu. 0:16:13.870 --> 0:16:14.550 Yeah, that's right. 0:16:14.710 --> 0:18:07.190 Which makes a, a really big shift. I tried to look, I didn't see any like concrete confirmations that, that this is true, but people seem pretty sure that this is, is what's happening. Of course Nvidia has been, you know, a big story for a while. Lots of people would love to sort of break in into that space. But I think also TPUs tensor processing units as different than GPUs, graphics processing units, although not sure how applicable that name is anymore, but are much more in theory power efficient or could be. So for Edge, sort of compute robotics, that kind of stuff. It would be really cool to sort of unlock this in my mind. I'm, I'm excited but no one seems to be definitely pressing against it. Where we have things that aren't just static, but actually training and inferencing and sort of adapting themselves and rolling forward so the things are learning. I can have, you know, something on my Edge device that is improving itself, learning my habits and not just like in a sort of out of band kind of way, but in a sort of like ingesting new data and updating and sort of refreshing as opposed to the sort of release of new checkpoints. That is kind of what happens today. And so definitely a leap forward. Lots of cool stuff happening in this space. Lots of competition. Also really interesting. We were talking just, just the previous, you know, about companies and having, you know, rob the bank, but it's funny how dynamic the situation is with top players still. So Deep Seek and you know, meta Google, OpenAI anthropic. Like every, like, it seems to be still a very open playing field for, you know, all of these providers and no one seems to be carving out a specific niche yet. And they seem to just all be kind of generally competing. So lots of speculation about how that's going to develop. 0:18:08.390 --> 0:18:37.460 Yeah, yeah, that makes sense. I think that the, the challenge, you know, tying a lot of what you said together is, is that we haven't seen a lot of what I call value capture. I'm not sure if I'm using that word correctly, but, but basically. So, okay, look at like search engines. So you know, there was altavista, there was Ask Jeeves, right? There were all these different search engines, but they didn't really capture a lot of value. And there were attempts to say like, oh, if you. 0:18:40.210 --> 0:19:27.290 If you pay a certain amount, then you can be kind of in like a, a certain tier or like there's a certain section that was premium and that didn't really take off. And then Google really captured value by just saying like, anyone can pay to be at the top of any search result. And here's sort of like a mechanism for doing that. And then Google captured, you know, a ton of value and it continues to do that. And so for AI, it's like there's a lot of. Yeah, as you said, the Turing Test is totally blown out. I mean, this is a little embarrassing, but my dad thought that Meta AI was an employee. He's like, you know, I talked to Meta AI telling them I was having problems with my account and they couldn't really help me, but they were so nice. I talked to them for like 15. 0:19:29.290 --> 0:20:05.060 So the Turing Test for, especially for, for, you know, older folks is totally solved. Right? I mean, younger folks continue to get more and more savvy, of course, but, but, you know, Turing test totally blown out. But, but, you know, that's a good example. You know, like, did Meta create any value there out of that 15 minute conversation? No. Right. So, you know, there is customer support, representatives, representatives, that's Sierra. And a bunch of companies are doing that. That might be the first place where they're capturing significant value. Although even there it's. 0:20:07.700 --> 0:20:38.660 It'S, it's a cost savings versus, like in the case of Google, they created a new market. So cost savings is okay, but you know, real value captures when you create a new market. Like for example, a robot that, you know, can mow your lawn and do your laundry with one robot or something like that. Now it's like, okay, there's a whole new market that you just opened up. And so until I think AI captures value, it's going to be really hard to benchmark. And. 0:20:41.060 --> 0:20:53.860 This is, I guess, gets into the whole AI bubble, you know, Can AI capture value rapidly and at a large of magnitude to escape all of the debt that it's accumulated? Nobody really knows yet. 0:20:58.910 --> 0:20:59.150 But. 0:21:00.910 --> 0:21:16.270 But yeah, Gemini 3. From a technical standpoint, Gemini 3 is very impressive. I don't have access yet, so I'm only going based off what I've read that Internet reviews. Yeah, exactly. Rotten tomatoes gives it 97%. 0:21:16.830 --> 0:21:20.890 Have you tried the image one part of it, though? Nano Banana have. 0:21:20.890 --> 0:22:17.060 I've done a lot with Nano Banana. Nano Banana is very good, very fun. Yeah. Folks out there try these things. If you haven't tried them yet, they are super fun. Okay, my second news topic is depth anything 3. So, you know, there's a bunch of these anything models in computer vision. So the most famous is arguably a segment anything model. And what, what Sam does. And there's now SAM too. You give it a picture. No training, nothing. Like, first you download the model. The model's enormous. Right. But then you give it a picture and it segments the picture for you. So, for example, let's say you are working in a factory and you have a bunch of photos of a conveyor belt and different things moving along the conveyor belt. You know, you could like start doing background subtraction and all sorts of computer vision stuff if you know what you're doing. 0:22:19.450 --> 0:23:45.180 Or you could just run Sam, and Sam will segment out the conveyor belt and the object, which was one line of python, which is pretty remarkable. So, course is not perfect, but it's pretty darn good considering it wasn't trained on your use case or anything. I've used it for a bunch of different things. It's very impressive. So. So, you know, going along with that trend, we now have depth anything. So you can give it a picture of anything and it will generate a depth map. So for each pixel, it will tell you roughly the distance those photons had to travel to go from whatever it is you're looking at to, you know, that, that lens or your eyeball or whatever it is. So, so for example, if you run this on a picture of a tennis ball, you can imagine like this sort of radial image where the parts on the edge of the tennis ball are the furthest away and then part of the tennis ball that's in the center is going to be the closest to your eyeball. Right. So. So depth anything. And I just think these anything models are just super interesting. I mean again, it's like, it feels like there's just so much value you can capture there once you have the ability to just get a depth estimate for from, you know. Pretty. Pretty wild. 0:23:47.180 --> 0:23:53.940 Yes. I've been looking at this for a project I'm wanting to work on where I. I don't ask. 0:23:53.940 --> 0:23:54.340 Very weird. 0:23:54.340 --> 0:24:10.100 Anyways, having like two web cameras and stitching them together and I thought it'd be really easy to like stitch into basically like a wider field of view with an arbitrary number of cameras. Like I assumed this is like a solve the problem. Like open CV would just have something. 0:24:10.500 --> 0:24:10.980 Yeah. 0:24:11.300 --> 0:25:32.210 So it turns out I like digging into it. Like there's some pay software you can. But basically you can kind of like understand the transforms between them. You can kind of do some of it, but, but basically and I didn't really think about it. The one of the problems unless you have a very, very carefully set up system is understanding the distance away of the various objects. Like an insta360 or something where it's like very well controlled, the positioning, but just setting up something in an arbitrary arrangement and then having them stitch everything's at different distances and you end up with parallax. And so I kind of got into thinking about how you would integrate something like one of these depth anything to understand the depth estimates, do the stitch and then maybe you would not be able to run it in real time. So maybe you would run it like every so often and like sort of update in the background the depth. And so it wouldn't be exactly perfect for you know, moving objects, but for static objects it would be okay. So definitely very powerful though because you know, whatever. Just a year ago, a few years ago it would have been incredibly difficult to like I have an image, how do I like produce a depth map that you know, isn't just atrocious because they were not very good. They were very like flat plane before based on sort of a segmentation. And so these new ones are much, much more impressive. 0:25:34.060 --> 0:25:45.580 Yeah, that makes sense. I mean you're. You're trying to calculate like the trans. The 3D transformation of each of these cameras. And so you could do that at a low rate and then just use the most recent transformation. 0:25:46.620 --> 0:25:57.740 If there were infinite time, I would have done this as a project over the last week. But unfortunately I only got as far as understanding where the problem was and and having an idea of what I would build. So. 0:26:00.250 --> 0:29:19.790 Speaking about projects I finally got around to, though I know we've talked about on the podcast before Run Pod, but I finally said I had not tried any of the AI video generation tools and I wanted to not just like pay for it, was like you pay for it. I didn't want, I'm, I'm, I'm cheap. I didn't really want to pay for it and I wanted to kind of learn it myself. And so my, my kids had a picture that they wanted to like add an animation to. In this case it happened to be, they were, they were with an animal and they wanted the animal to do something funny, like something hilarious and you interact with them. But it was just a static picture and of course nothing actually happened. They just were, you know, like at a zoo, just standing next to the animal. But we wanted to animate because we see these videos on the Internet. And so I said I'm gonna, I'm gonna finally figure this out because I, I, I should, I should know how to do this. And went through the process of using WAN 2.2. I don't know that's particularly relevant, but finding a template, finding the way to get it to auto download on runpod and, and I was assuming I was going to have to sort of give it my credit card and risk it, but actually I was pleasantly surprised. Shout out to them. Unlike, you know, AWS or something else you basically buy. In this case, I bought I think the minimum like $25 of credits worth of runtime. And if you're careful to select a GPU that is not like super state of the art, it's very reasonable. So I think in total I've made, you know, four or five, five second videos and I think I spent like two or three dollars of my credits for like bringing up, downloading all of this stuff. And if you were to generate a bunch while the session is up, it would be way cheaper. But I, I don't because I like do it like I'm going to figure out. So I take it back down. So I'm not using time. But if you're careful with it and you're interested in these things and you're like me and you don't have a like modern GPU that can run any of this stuff locally by the way, which is really expensive. Even if you do, um, Definitely check out RunPod, but also check out some of these like bigger models. I've run locally like stable diffusion type models before, but the video Generation just doesn't work very well. But if that's been holding you back, definitely check this out. And it is very interesting to learn how to use like comfy ui, how to like kind of do all of this stuff that, that you kind of, at least for me, you see in YouTube videos or people. How are they doing this? My family likes to watch these like Christmas videos with music on the TV in the background. But it's obvious they're all AI generated because people do weird stuff in the background. I was like walking around to like a Christmas market and a kid is like flinching between like, you know, wearing a jacket and wearing a coat and wearing. It's like. Anyways, people are out there doing this and turning it into like money, which I'm not interested in. But I feel a little left behind that I haven't been making funny AI videos for my family. So I have this thing where I want to go back and take like pictures of my older relatives or even people who have passed away. Sort of just making kind of that Harry Potter style, like subtle movement, you know, animation for some of these and sort of like providing that as like a sort of like a gift into the family, you know, or like funny moments, you know, Here's a picture of my father at the Thanksgiving table and the turkey, you know, blows up or something, you know, and then, you know, people will get a laugh out of it. It's very cheap humor, but I'm here for it. So I'm trying to figure this out. This is my quest. 0:29:20.030 --> 0:29:24.910 This is. You came to the right place, Patrick, because I've been doing this for a while. 0:29:25.150 --> 0:29:26.030 Oh my gosh. 0:29:26.430 --> 0:29:51.500 And the website I use is Foul AI, which stands for features and Labels AI. And they have like warmed up all the models for you. And so you just charge. You're charged per call. And so for like image to video, you're charged maybe 8 cents or 25 cents or something. Recall. And so the most recent one, there's two. 0:29:53.660 --> 0:30:15.750 So my wife's pretty paranoid about the kids safety. I mean, this is like very common, right? And so I took. I take the kids on these hikes and sometimes she gets nervous and I'll sen. I send pictures. So I sent her this picture of the kids standing over this rock ledge. And underneath is a. Is a creek and there's actually a. 0:30:17.510 --> 0:30:44.160 There was a. What's it called? It's a black snake, a water moccasin. There's a water moccasin in the creek now. Now like it's like maybe a 10 foot drop off this rock. So like, you know, the water moccasin is not coming up to us or anything like that. We're just sitting there on the rock. It's a giant boulder, you know, eating sandwiches and watching nature, watching water moccasins or whatever. I sent this picture and. 0:30:45.920 --> 0:30:52.960 And my wife gets really nervous and I was like, I was like, oh, check this out. So I sent a video where. 0:30:54.560 --> 0:31:01.440 Basically the water vacuum leaps out of the water, turns into like a dragon and eats all of us. 0:31:03.600 --> 0:31:27.300 And I was like, oh, I guess you're, you're right the whole time. The other one I did was there's a model called X boson which is, you know, kind of like the CERN particle thing, but it's a voice cloning model, totally open source. And that one is because it's just audio. You can run it on a reasonable GPU or even on a cpu. 0:31:28.980 --> 0:31:41.500 And so I cloned some celebrities voices and I had them wish a happy birthday to some friends of mine and I purposely picked celebrities that I know my friends hate. 0:31:41.740 --> 0:31:42.300 Oh. 0:31:43.740 --> 0:31:45.900 And it's like, hey, I know you're a big fan. 0:31:49.340 --> 0:31:59.740 So good, so good. And you know, it's everyone I sent it to, I knew that they would find it funny. So yeah, I didn't send it to anybody who would be genuinely offended. 0:32:01.930 --> 0:32:14.490 I saved that for my family. But for these people and they thought it was absolutely just drop dead hilarious. They were asking me how I made it. They want to make it for their parents, for their friends, all that stuff. 0:32:16.570 --> 0:32:37.550 Yeah, I think the SORA video Generation from OpenAI seems to. I start to see some videos of these getting posted that seem pretty cool. But I don't know, the, the way that they want to scan your face as like the entry fee into like allowing you access to that always kind of creeps me out. So yeah, I haven't done, I haven't done it. So. 0:32:37.630 --> 0:32:49.150 Same. Yep, exactly same. I mean the ones on, on foul, they're open source, they're decent. They're not great. They're definitely not, sorry, quality. But they're, they're decent. I mean for a laugh, they're good enough. 0:32:49.470 --> 0:33:56.290 Perfect. That's all I need. All right. My book of the show isn't, isn't even a book, it is a video, but it's a documentary. So I'm going to call it sort of cerebral. And therefore my book of the show, which is the Thinking Game, which came out recently on YouTube and is sort of a, a DeepMind documentary and a little bit behind sort of, sort of their story and it's, it's getting a lot of really good reviews. Something that if you haven't heard about, definitely worth checking out. I think it is very sort of, how do you say, apropos, like very of the moment. I think it's like an hour and a half sort of long. But you know, one of those sort of well shot sort of documentaries going through the story of a lot of history there, which is crazy to think they've been at it for a while, but just the, the progress, like even. Which now we're back to the side, but even this video generation. Like I saw someone regenerating Will Smith eating spaghetti, which I know we've talked about and I was like, oh, that was like 10 years ago. Like nope, nope, that was like two and a half years ago or something. It's like we've gone from that to that's incredible or something like that. 0:33:56.290 --> 0:34:03.810 I mean now it's like flawless. I mean now if you ask for that video you just get it and it's like you have to squint really hard to find the difference. 0:34:04.450 --> 0:34:42.500 Yeah. So two years ago the YouTube video was AI Will Smith eating spaghetti. Yeah, spaghetti Sally. Yeah. Anyways, this crazy and the, the DeepMind folks, you know, AlphaGo was like an early headline that was like 10, 10 years ago. A little less than 10 years ago. So I mean crazy, crazy how far, you know, this stuff came sort of. But of course people have been at it for a lot longer really thinking about it, really working hard on it. So definitely something, something worth checking out. Not, not like sort of super deeply technical. But sometimes it's okay. Sometimes it's good to see sort of like an artistic view of the stuff that we all maybe take a little bit for granted. 0:34:44.100 --> 0:34:49.300 Yeah, totally. My, my book of this show is a classic. It's Plato, the Republic. 0:34:50.980 --> 0:35:09.620 This is like a bunch of volumes. It's a, it's a large tomb. But I found it super interesting. I don't know if we've, I think we've probably talked about this but, but you know, one of the things that, that, that they focus. I mean it's a massive book but one of the things they focus on is. 0:35:12.480 --> 0:35:24.480 You know, have you, you've probably heard this thought experiment, Patrick. So let's say somebody loans you their gun, you borrow their gun because you're going hunting or something. 0:35:24.720 --> 0:35:25.280 Okay. 0:35:25.280 --> 0:36:11.550 And then yeah, you kind of forget that you borrowed it. They come back months later and they're in a rage, you know, and they're maybe intoxicated. They're intoxicated. They're super mad. They're like, I'm so mad, I need to take out Bill. Bill's gonna get it. I need my gun back. So, you know, like it's his possession. And so if you don't give it back or you lie and you say you don't have it or you threw it away or something, you're kind of stealing, you're doing something immoral, right? But if you give it back, that person is going to go on some rampage. And so it's bad for the world, right? And so, you know, Plato and, and, and Socrates. And there's this other person who plays the. 0:36:13.870 --> 0:37:25.570 The person on the other end of the debate, but he always loses. Basically, Plato wins every time, right? So there's this poor guy, I forgot his name. Terrible with names. But, but he, you know, he's just kind of a stooge, you know. But you know, he says, well, you know, yeah, you have to return it because, because it's his property and you're stealing it if you don't do that, you know. And then Plato says, well, but you know, it's for the greater good for you not to do it. And the conclusion after like all this debate they come to is, is that you. There's really no absolute laws. You know, everything kind of has a caveat. And you, you know, there's what, what you should true do instead is try to come up with like proper values so like you, you value life. So if you know this person's going to go on some rampage, then the lives are worth the not telling this person the truth or whatever. So it gets, it gets. But, but it's just like such an interesting question and I'm sure if there's like a philosophy major out there, they are just like completely like, you know, tearing their hair out right now as I try to explain this. But I'm actually going through. 0:37:27.570 --> 0:38:07.250 Instead of just reading it because I feel like I am just. It's kind of like, it's like it's almost to read a reinforcement learning research paper in Europe's next month or something. It's like, it's like there's just so much background, right? So instead of doing that where I knew I wouldn't really get a lot out of it, I paid for this, this basically class where this person who, you know, has a master's degree in like literature and philosophy kind of reads it and then kind of like an audible, but audible plus plus, right. So this person reads it, then they stop and then there's a discussion and then they read the next section. And so. 0:38:08.930 --> 0:38:32.900 You know, a lot of people, like, I don't know if a lot of people do this, but I, I used to often get the urge to like read the classics. You know, it's like, oh, I've, I've tried to give a good example here. I mean a lot of the classics are pretty relatively easy to read. Like Moby Dick is not that hard to read or whatever. But, but like the Republic's an example, you know, the Divine Comedy, you know, these books that are just like. There's so much. 0:38:34.420 --> 0:39:14.810 Knowledge you need to have, especially about the era and the time times and all that to really understand them, that it's just not tractable for people like us. This is like a great way to read those books and understand them. So, so I'm about, I think halfway through they're also skipping some volumes because I guess some volumes are just so specific to that era that they couldn't really get enough value out of them. You know, without, like again, without. If you study enough, you could do anything. But they, they know their audience so they're like, we're going to read volume one and then we're going to skip to volume three, et cetera, et cetera. But I've been having a blast and I recommend folks try that. 0:39:16.810 --> 0:39:34.090 Yeah, I don't, can't say I've ever read any of the hard to read classics and never done a philosophy. So I won't say much because I too feel there's a whole thing where people like to play at understanding it and then like, I do respect that there is a whole field of study there and I know nothing about it. 0:39:35.590 --> 0:39:44.230 Yeah, I feel like I know a little bit more than I did. But you know, so. Patrick, I'm not giving your gun back. That's basically. 0:39:48.710 --> 0:39:51.350 Well, I. Okay. I take offense that I'm on a. 0:39:51.350 --> 0:39:52.630 Okay. Anyways, we're just moving on. 0:39:53.030 --> 0:40:03.440 My tool of the show is a game. It is. I thought for sure I must have talked about it before, but I haven't. Which is Core Keeper. Core keeper is a 2D top down. 0:40:05.040 --> 0:41:14.880 Game in the sort of style of Minecraft or Terraria where you kind of mine blocks in a, in a subterranean environment and you are interacting with creatures and you know, fighting. What do you call mobs I guess is the Minecraft term, and kind of going around and automating some stuff and you know, gardening and doing. I guess it's like a life Simulator meets some kind. I don't know. I never got into the Sims, but definitely have enjoyed Terraria and Minecraft and this is sort of in that same vein. Which is Core Keeper. It's available. I don't think it's on mobile, but it's on PC, Switch, Xbox, PlayStation, so lots of places. It's not a new release by any means, but if you haven't noticed yet, I'm way behind on playing like new games, but I've been playing this one and enjoying it. And for those folks who remember from last episode, I am proud to say I finally launched a rocket in Factorio. I forced myself to sit down and do it and then I had the inclination to keep going and I was like, nope, I am stopping and I've got to switch. So now I've been playing Core Keeper, play some other games. But yeah, is it, is it co. 0:41:14.880 --> 0:41:16.720 Op or are you playing by yourself or both. 0:41:17.600 --> 0:41:48.620 So I think the kind of default mode is played by yourself, but there is a co op mode as well. There unfortunately isn't a split screen, which would have been amazing, but yeah, I think it's a sort of between, between systems. Co op can take place, which would be super cool, just not something that's feasible with my, my setup and my family. So it would have been awesome if it was a split screen, but you know, whatever left wanting there. But definitely still your game. It should be relatively cheap because like I said, it's. It's been out for, for kind of a long time. 0:41:50.810 --> 0:41:59.370 Very cool. My, My tool to show is also a game. It's workers and resources. Soviet Republic. 0:41:59.610 --> 0:42:01.930 Oh dear. Sounds like. Papers, please. 0:42:02.570 --> 0:42:16.260 Yeah, yeah, yeah. So it's, it's set in a, you know, a member of this, you're a member of the Soviet Union probably in like the 80s or 90s, if I had to guess. Early 90s. Early, early, sorry, 70s or early 80s. 0:42:18.020 --> 0:42:36.660 And it's, it's a, it's kind of like a city skylines kind of thing where your job is to set up tenements, set up factories. There's a supply chain. The thing that's pretty cool, I think that this game did a good job of is you can choose how. 0:42:38.500 --> 0:43:36.410 Okay, so, you know, in Civilization, you can choose to have a manager manage any of your cities or all of them. Right. And so at least when I play Civilization in the beginning, I start micromanaging my one city because I feel like I need to get the avalanche going faster than everybody else. So I need that tiny snowball to start Right. But then once the avalanche is going, and especially if I'm, if I'm winning, you know, I'll just turn all the cities on managed mode and then just focus on my army and just winning the game. Right. And so this has that mechanic in it where you can choose the level of simulation and so you know, you can actually go all the way down to like each individual person needs to be fed and have all their needs met and you even need to like bus them to their job. So you need a bus route and all that. 0:43:36.560 --> 0:43:36.800 That. 0:43:37.600 --> 0:44:20.180 But as your city grows, you can kind of like up the level of simulation and then it will start like automatically trying to do the right thing there. And so in the end you're kind of running like a pretty large city and you're kind of managing all these different AIs and trying to make more strategic decisions. So I'm still pretty early in the game, so I haven't uncovered a lot of it yet, but it's very cheap. It's on sale, which the sale's probably going to be over by the time the show airs, but you can put it on your wish list. It'll go on sale again. But I found it a lot of fun actually and it's kind of a twist on the city building I haven't seen. 0:44:23.140 --> 0:44:23.380 That. 0:44:23.380 --> 0:44:29.620 Sounds cool. Off the trail. Although my initial search for it tells me people saying it's a very hard game. And so I'm nervous. 0:44:30.540 --> 0:44:33.660 Okay. Yeah. To be fair, I have been playing on very easy. 0:44:33.820 --> 0:44:34.460 Okay. 0:44:35.180 --> 0:44:42.540 So I'm still sort of exploring the game. We'll see if it's become so hard that to play on normal is just not forgiving. 0:44:42.940 --> 0:45:18.430 I saw something I. I used to feel bad about this and then I saw a quote keep popping up I think on Reddit or somewhere where the, the person who made Elden Ring and some of the other like really difficult games which I. Which I've not played, points out that even though they're really popular for making these like very hard games that they actually kind of suck at video games and they, they use like everything they can to just like make the game easier for them to play through. So they build these games that are super difficult, but yet they themselves play it on basically like give me mode and I'm like, oh, okay. I feel a lot better than even the like, I'm not, I'm not alone so I shouldn't feel bad for doing this. 0:45:19.790 --> 0:45:50.680 Yeah. The only game I really play on hard is, is are these turn based kind of tactical games. Like, I got the Final Fantasy Tactics remake on Switch, and I played that on the hardest setting. And, you know, the thing about those games is once you have the patterns down, then you just kind of know what buttons to press. So it's. It's. I feel like I don't know that much about Dark Souls, like, games, but my guess is that those people who. Who play those games have, like, just a lot of muscle memory. 0:45:53.280 --> 0:46:02.800 All right, time for the topic of the show, which is either going to be a love it or hate it topic, but we're here for it, provide our own hot takes, becoming a manager. 0:46:04.240 --> 0:46:11.360 So let's give a little background before we start. So Patrick is a manager. Do you manage managers? 0:46:11.680 --> 0:46:12.160 Yes. 0:46:12.240 --> 0:46:18.080 Okay, so Patrick is a mom. I call those people moms. Patrick is a manager of managers. 0:46:19.870 --> 0:46:48.890 Obviously, Patrick has been a, what I call a line manager, which means a manager of pure. All individual contributors. And obviously Patrick's been individual contributor. So on my end, I have more or less done the same. So I've been individual contributor, a manager, a mom. I think one time I managed a manager, a manager or something like that. I don't know. There's one more level there, but it was. It was like. 0:46:50.730 --> 0:46:55.690 Like one part of my organization, so I barely dipped a toe into the next level. 0:46:57.690 --> 0:47:28.230 And I've been in IC sporadically throughout that. So. But Patrick and I have kind of done the thing, so to speak. So, yeah, we're not. We didn't do a ton of research going into this. We never do. I feel like it's more like it's better for us to just give our own experiences rather than tell you to go read someone else's experiences. So you're going to get kind of, as Patrick said, our hot takes. But, you know, there are takes of people who have kind of been there and done that. 0:47:28.870 --> 0:48:05.330 And I think to Jason's point, it'll come from the point of view of people who have been specifically engineers in our cases, because that's what we know. And then being asked to, or having the opportunity or wanting to become a manager. I think there are other people who come up in sort of a more direct path to a manager or from product management roles or other things. And we're not. I don't think we're going to have much to say except by incidental sort of discussion about those things. And similarly, we're not going to talk about, like, what it means to manage an engineer if you've not been an engineer yourself. 0:48:06.620 --> 0:48:24.700 Yeah, good call out. Yeah, there's Also, as you said, there's, there's blind spots. But you know, knowing our audience, I feel like we probably have a lot of people out there who are either thinking about becoming a manager or they've kind of followed sort of the same trajectory and, and hopefully this will help you folks out. 0:48:26.860 --> 0:48:29.580 So what is a manager, Patrick? 0:48:29.900 --> 0:50:47.340 Well, I was going to start, I don't know, I'm not going to attempt to do the executive summary. I hate executive summaries. I asked ChatGPT what is a manager, but a manager is someone who helps to organize the tasks of other individuals. And I think for this discussion, for the most part we'll focus on managing individual contributors so people who are themselves doing technical work, while you may be doing technical work yourself as a manager as well, either contributing, doing peer reviews, perform sort of like code, code checks, you know, helping to sort of steer direction architecture and that kind of stuff. And I think a couple of I guess like high level observations before we get in is the first thing is like opportunity. So unless you are kind of like starting a business and doing a startup for the most part, if you're individual contributing somewhere, there is likely going to be a time, at least in my experience, if you're doing well as an individual contributor, where someone will come along and ask you if you would like to manage another person or if you know a manager is leaving, they may ask you to take over a team. I think maybe we'll get to it. But there are differences between sort of starting your own team versus taking over a team that are pretty distinct. But it is something that if you've been an individual contributor for a while, you need to kind of think about whether it's something you want to do. And in most good companies they try to provide paths for people who don't want to become managers and who do and should try to generally keep the pay scale equivalent between them. So that is, it shouldn't be strictly a financial decision at your current job and salary change to become a manager versus stay an individual contributor, although results may vary. Um, that can highly depend on where you are. But that opportunity will come and it may come more than once. It may only come once it made come lots. It may be something that's talked about if you transition to a new company, but definitely something you want to think about before you get into that situation and sort of not, not any specific time. Right. You never know. But kind of thinking through if that's something you want to take on or not. 0:50:50.370 --> 0:50:51.490 Yeah, that's a good call. 0:50:51.490 --> 0:50:51.810 Out. 0:50:53.410 --> 0:51:00.530 I would say that, that the big difference between a manager and an engineer is as an engineer you. 0:51:02.770 --> 0:51:42.840 You'Re expected to meet some sort of key results. So you say we're going to get clicks on Instagram to go up by 1% this quarter and you code, code, code and you run tests and, and you know, you build models and you replace old models with new models and then boom, you get one and a half percent and you exceeded your expectations. Right. As a manager you also have to add retention to it. And the way the results are measured is pretty different. But, but the, the, the sort of categorical difference is that you have to add retention. 0:51:44.800 --> 0:52 And the thing about retention is it's like a massively laggy indicator. Right. So, so if half your team quits, well like you've like you or your manager have like really messed up because they should have seen the retention problem like way further in advance. But. 0:52:07.600 --> 0:52:29.090 But you know, to be a little cynical, you know, if people are willing to stay for no pay, let's say people are just so passionate they're willing to stay for $0, well that's what the company should pay them. Right. I mean, I mean I know we're being kind of cynical here but, but, but you know, retention is the way that companies especially retention of, of key talent and, and. 0:52:31.170 --> 0:52:46.650 High rising folks. That's the way that companies kind of measure the health of their leaderships. So, so, so you have to add that dimension to, to yourself to, to be a manager while continuing to deliver good results. 0:52:47.690 --> 0:54:54.620 Yeah, actually you bring up a great point. I'll mess up your alliteration of results and retention by saying I think we're talking about managers as thinking about both tasks and people and there are strictly people managers which are not responsible for technical sort of tasking and key performance indicators, but just of sort of personal growth, happiness, retention. And there are also whatever you want to call like technical managers who are you know, managing projects and sort of like technical teams and tasks without being responsible for the growth of individuals directly or retention as you're kind of saying. And I something that maybe is a personal philosophy but I try to communicate to people on my team as being responsible for both, you know, I'm responsible for both tasks and people and I try to tell them that as they're growing that one of the things that's not, is a must have, not an optional, is leading that leadership is something that's not optional, that whether they become a people manager in addition to their technical work, they need to be leading Technical work, regardless, that is driving good architecture, managing the code base, helping organize other people behind key ideas and sort of rallying folks towards their cause and being a winning personality, like whatever you want to kind of say, as sort of soft skills, but that those things are not really negotiable. But if you aren't willing to sort of lead and help demonstrate a way that other people can come behind you, whether you manage them as people or not, then it is going to be very difficult to rise and to get promotions and that kind of stuff. That's my personal approach to it. There may be other ways, but in the way that I do it, I, I think it's a good way as an individual contributor to think about it. You're doing your own thing, but then you are also doing it in a way in which other people could follow, mimic you. You're setting good patterns, establishing practices, and in that way you're leading whether or not the people whom you are leading, like, are under you in the organizational structure. 0:54:56.230 --> 0:56:10.480 Yeah, I mean, one thing that I would say is we've, we've worked at companies before where, yeah, there were pure people managers, they're called, like functional managers, I think. But generally speaking, most companies, most managers will be responsible for both. Now, it doesn't mean that that manager will be delivering any of those results, but they are sort of at the end of the day responsible for those results. And that allows them to, to decide sizing. So based on sort of how, how, how much of the sort of key results they're responsible for, they can use that as one of the indicators to decide whether to hire more people. Or on the other end, if, if, if, if. There just isn't that, that many results that they could be responsible for. So here's an example of that. Let's say you're on the security team and there just isn't that much, that many changes to the code base. And so there just isn't that much for your security team to review. Well, that kind of puts a hard limit on how many folks you can hire. So, so there's kind of like this connection between. 0:56:12.720 --> 0:56:17.900 You know, between sort of the people management and the technical. They do kind of overlap in that area. 0:56:20.140 --> 0:56:54.130 But, but yeah, a big part of management is sizing and growing and shrinking the team. And so the, the way you grow the team is through, is through, is through hiring. So as a manager, you will become very adept at looking at resumes, quickly getting to, you know, quickly assessing someone's, you know, both their current competence and their Ability, their future competence, you know, quickly in an interview. 0:56:55.890 --> 0:57:17.890 And ultimately like kind of deciding, you know, whether to kind of go, no go on hiring folks. That's. There's usually a hiring committee, so you're not making that decision by yourself. Um, but. But you're going to be, you know, a very big voice in, in that committee because you're directly responsible for that person. 0:57:18.850 --> 0:58:41.480 Yeah, I mean, I think that's a. We. We talked a lot about interviews on our podcast before, but now we're talking about a little bit from. From a different side. I think it is one of the not talked about but key things. And I see it in, in new managers or even experienced managers, which is in reality, many times the call on who to hire, like a specific person is your call, and you're responsible as manager for making that final decision. Now, there could be things that prevent it. Right. You know, you're not able to get the compensation package you want. The person doesn't accept what, whatever. But at the end of the day, if that person joins your team and they're bad and they impact themselves and the rest of the team, like, that's on you getting them up to speed. That's on you. Like, you know, and it's one of those things that you have to be very thoughtful about, about picking the right person. And to be honest, like, I've done many, many interviews, seen many people hired or not hired. I've never seen what I would call is a perfect candidate. A candidate that was just like, there is absolutely no thing that we're concerned about. This is absolutely, we should go all out and, and do everything possible to hire this person. If there was, I'd probably be really scared because it'd probably be very expensive and then itself would be a sort of struggle. But, you know, almost always there's some downside. 0:58:41.720 --> 0:58 Those candidates, then you're like, are they a flight risk? Why are they taking this job? Are they going to demand a promotion? So, yeah, there's. There's never a perfect candidate. Yep, yep. 0:58 --> 0:59:14.490 And so that's always comes down to at the end of the day, just being comfortable with that risk and eating it yourself, which is, in my opinion, different than individual contributor. The number of things, as a manager you end up having to, what I call, eat the risk. For it's your risk. Your manager's gonna blame, your manager's gonna blame you if you get it wrong. The people under you are gonna blame you if you get it wrong. And there's really no one else to turn and point To. 0:59:15.290 --> 1:00:59.350 Yeah, that's an amazing metaphor. I wanna just pull on that thread so, you know, I'll connect it to money and sort of like the Soviet Union and marks and all that. So if you think about a worker kind of works, you know, let's say an hourly worker, you know, works that hour and gets that salary. But what actually is the reason why the owner should get anything? Let's say the owner doesn't actually manage the convenience store or whatever it is. Well, it's because the owner took a risk. Right. The owner put in a bunch of capital into that convenience store that they could have put into the stock market or, or whatever, or just held it in a treasure chest or something, but instead they put it into this convenience store. They lost all of it, you know, on day one, and then they built it back up and now they're getting dividends. So. So the owner is getting dividends on this big risk while the, the worker is getting kind of paid for their labor. And so as a manager, you know, even if you don't own the company, I think that same kind of metaphor holds where, where, you know, I carry the burden of all the bad decisions that I've made as a manager. And as Patrick said, a lot of these decisions take years and years to reconcile and maybe they never get reconciled. Right. And so you just have to live with that. That is something that is pretty unique. I mean, as an engineer, you might have this when it comes to picking a programming language or an architecture or something, but by and large, you don't carry that same kind of cloud over you as you do a manager. That's just something that you learn to. 1:01:00.950 --> 1:01:01.750 Thrive in. 1:01:04.230 --> 1:01:15.880 So talking about how to manage, what are some of the pieces that come down to kind of the day to day, the month to month aspects of accomplishing the job as a manager. 1:01 --> 1:01:32.240 Yeah, I'll cover the first one, your job as a manager. And this is true whether you're a middle manager, an executive, et cetera. If, if you're an executive, there might not be anybody above you per se. The board of directors is kind of a totally different thing. 1:01:32.240 --> 1:01:32.920 Stockholders. 1:01:33.240 --> 1:01:58.860 But yeah, you're, you're, you're kind of a council. And so you still have to advocate. There's still sort of like a, you know, if, if you want something and none of the other executives want it, well, then you're not going to get it. And so they kind of collectively are your superior for that one argument. Right. So. So the process is more or less the same all the way up the chain where. 1:02:00.620 --> 1:02:28.410 The company has a goal. Let's say the goal is, you know, the company makes, uh, it's an orange cooperative. The company grows oranges in orange groves. And what your job as a manager is, or management, let's say collectively, is to break that goal down. So I guess tech is a better example. Maybe the goal of the company is to. 1:02:29.930 --> 1:03:45.550 To, to, you know, run this sort of ad service and this run like a Google search engine and the ad service, that's the company. But somehow that goal has to be broken down into such a fine grain that you can justify like upgrading Python. Like why should the company move to Python 3.14, right? So someone's got to draw like a causal link from, you know, the overall goal of the company all. And some of those goals are sort of like, I guess, spiritual in a sense. Like, you know, maybe if like Google's motto, you organize the world's information, it's kind of like more of like a spiritual or inspirational goal. But you could kind of, you have to draw a line from that all the way down to, okay, we're going to abolish Python 3.7, right? Or something like that. So, so your job as a manager is to do that and also size the opportunity and the risk. So if you say, sure, me and my team of two people are going to upgrade all of Google to Python 3.11, well, like you just took on a goal that is just not going to happen. And so you're, you're going to fail at that goal, people are going to get burnt out and. 1:03:47.390 --> 1:04:24.260 And it's just gonna be very frustrating, right? Or you're gonna have to hire a ton of people very quickly. On the flip side, if you say, if you make a goal but you don't kind of draw that line and correctly, or it's just not ambitious enough, then even when people accomplish the goal, you know, they won't be able to get recognized for that accomplishment because it's just not tied back to the company in a way that's influential. So, so that's a, that's, I think the biggest part of, of being a manager is, is getting, getting that part right. 1:04:27.060 --> 1:05:03.520 And I, I think a lot of that comes down to trying to set up frameworks so that the people under you can make those decisions. Like, the higher you are up, you don't want to be, you know, minor versions of Python to be something brought to you. You want to establish the framing so that the people under you understand the key concepts of maintainability. Of you know, getting new features of hiring of those kinds of things and are able to make those decisions and give you the high level, sort of like hard or arbitrary things with that that the framework doesn't cover. 1:05:05.280 --> 1:05:07.520 Yep, yep, yep. That makes sense. 1:05:08.560 --> 1:06:29.880 So another thing about managing is thinking about, you know, a role as sort of a coach that you have the players on your team, you have these goals that you're working towards the game field and rules have been set and, and now you need to get everybody to, to kind of play together and develop the individual players and I guess to kind of make up the word there the, you know, thinking about yourself as a, as a coach. And so now you start to think about things like having one on one meetings with the people on the team. And some people hate one on one, some people love one on one. Sometimes they happen weekly, sometimes they happen never. The race can kind of vary where you individually work through all of the people on your team and sort of talk about the various things about how they're doing, about what they want, making sure that you think about their retention, about their growth, about their happiness, whether they're burning out, you know, if there's other tasks, you should be sort of giving them opportunities for growth. And you know, sometimes in many companies I've been at, the concept of career planning is something that happens like right around performance review time, but it really should be a, a constant discussion. Hey, you did really great at this. Hey, this is an area where I expected you to try to do a little more or hey, have you considered also helping out on this thing? 1:06:29.880 --> 1:06:30.160 Right. 1:06:30.160 --> 1:06:41.120 It's a sort of ongoing continual thing to help that. And it seems like an overly formal word, but just having the conversation with people where you're trying to help them be better. 1:06:42.720 --> 1:06:46.610 Yeah, yeah, that makes sense. Um, yeah, I think. 1:06:48.290 --> 1:07:20.530 I think you know, one on ones of kind of. My strategy has changed over time, but the current strategy I have is basically, you know, I want them to be like relatively frequent, so weekly or bi weekly, but pretty short. So, so you should be able to say here's a piece of good feedback, here's something that was lacking. You know, why don't you tell me what's on your mind and hopefully you can get that done in 30 minutes or even a little less than that. 1:07:22.530 --> 1:07:25.010 But you want it often enough that people. 1:07:26.769 --> 1:07:31.090 Don'T forget something and then, and then you only get half the story. 1:07:31.970 --> 1:07:33.170 Oh, that's a good point. 1:07:33.250 --> 1:07:33.730 Yeah. 1:07:37.490 --> 1:07:37.690 Yeah. 1:07:37.690 --> 1:07:59.810 I think another big part of managing is the performance review system. And so this is this is something that I think a lot of engineers don't understand. The performance review system is ultimately an incentive system. It's a system that's there to drive incentives. And so. 1:08:01.650 --> 1:08:11.970 Because it's to draw, it's designed to drive incentives by nature it has to change because the incentives of the company change over time. So for example. 1:08:14.290 --> 1:08:19.890 This is way before I joined, but Facebook had a reputation for like. 1:08:21.890 --> 1:08:44.059 Poor. What's the word? Like poor uptime. Particularly like messenger. I don't know if you remember this. I mean, I remember this as a user messenger. I'd go to send a message and sometimes it would just fail. It's like, how can IM fail? You know, so, but, you know, just, it was, it was a reliability crisis. This is, I don't know, maybe 2007, 2008. 1:08:44.059 --> 1:08:44.459 Ish. 1:08:44.939 --> 1:08:45.339 And. 1:08:47.019 --> 1:09:04.900 At some point it leaked that Facebook had one site reliability engineer for every two engineers, software engineers. So it's like the system was just constantly falling apart. And, and so they made a push to. 1:09:06.660 --> 1:09:43.800 Again, all this before I joined, but they made a push to like do stable infrastructure. And they bought the company that made hbase. They bought a bunch of these open source projects or built them from the ground up, but they hired like a ton of people who are really, really good at solid infrastructure. And they added to the performance management system a new axis called engineering excellence where it would, people would talk about how, you know, they built something that, you know, was defendable, that, that reliable, robust, et cetera, et cetera. And so. 1:09:45.480 --> 1:10:35.440 You know that, that, that axis was added to push the company in a certain direction. Right. And so people, especially junior folks, will look at the performance management system as if it like was handed on stone tablets, you know, and then, and especially if things change or if it's interpreted differently, you know, one year versus the next, they're super discouraged. Right. And so as a manager, you know, your job is to, is to, you know, both like, see the trends that are coming up, you know, be a part of the change so that you can look around corners and just communicate well, you know, to your team, you know, of sort of where the company, like where the puck is going. 1:10:39.440 --> 1:10:39.760 Yeah. 1:10:39.760 --> 1:10:50.480 I think that as individual contributors, you look for like, what is the checklist of things I need to do to get the highest raise. And I think to your point, Jason, it's just like not how it works. 1:10:50.640 --> 1:10:51.120 Yeah. 1:10:51.280 --> 1:11:04.320 And it's, it's sort of a mis expectation if you think that's how it works. And I, I've seen bad, I've Seen less bad. But I don't think anyone sort of goes around saying they have, like, solved how to do performance management. 1:11:06.160 --> 1:11:25.150 Right, Exactly. It's kind of like no one can perfectly balance on a tight rope to where, like, half of the atoms in their body are on one end of the tightrope and it never moves. Right. Um, so. So, you know, it's constant adaptation. Yeah. 1:11:25.150 --> 1:12:32.080 And, I mean, I think it sets up. And I'm not trying to be, like, a negative person about it, but there's a natural tension where the company wants to retain people for the least amount of, you know, pay increase and least amount of salary because it's a cost and still get what they want out of it. But they know that they can't not do it, and so they must do it, but they want to do the least of it. And from an individual standpoint, you want to maximize how much you're being paid. So it sets up this natural. And there's all sorts of philosophies on how to measure if your system is working or not working. Retention numbers, you know, improvement numbers. And there's flaws with all of it, but ultimately it's just not. It's a sort of unstable thing. It's a. It ebbs and flows as the company is doing well, as the stock market is doing well, as your division in the company is doing well. You know, all of those things change. Yeah. Investor sentiment, what they're looking for. Are they looking for cost cutting or are they looking for growth? Like, all of those things impact what ultimately feels like a very personal thing. 1:12:33.290 --> 1:12:57.210 Yep, yep. Totally agree. Um, I mean, the one sort of solace here is, you know, Patrick and I are both, what, mid-40s, right. Early to mid-40s. And, I mean, I feel like there's still a ton of career left, but I also feel like I'm kind of done sort of chasing promotions, and so. 1:12:57.530 --> 1:12:59.130 No, no, no, you're not allowed to say that. 1:13:01.190 --> 1:13:03.030 Don't negotiate against yourself. 1:13:04.150 --> 1:14:19.370 Well, I mean, my point is, to junior folks out there, if you didn't, you know, make senior this time around, it's not that big a deal. Right. And so, you know, it's no reflection on you. It could just be the company, you know, really cared about stability, and you're working on, like, the new version of TikTok or whatever, and the companies didn't care about it that much this time around. It's very possible. Very, very possible. It could be like, everything was perfect, except your boss, like, didn't, like, say sort of the right Things or didn't have enough influence or you know, and it, and if you, if you look like around it could be someone else had the opposite where her boss said all the right things and they got the promotion and how did they get it? They're half as good as you or whatever it is. You know, I see this so much. Um, and, and like the, the thing to take away is like you have so many years in your career and if you just work hard all of those years, you will eventually get what you want. I mean, at least statistically. I mean not every single person to a person, but statistically it will average out. 1:14:20.670 --> 1:15:09.960 Yeah, there is a tension there between like you're saying short term and long term. And I do think, I do see that often where, especially when you're junior and I understand it's very difficult to say, hey, just have some patience. But I will say like you said as a manager in general, over long time horizons, things tend to work out and it tends to be okay. What that does leave out though is opportunities. And we talked about this thing last time for asymmetric opportunities. So you know, leaving to a different company or a different set of opportunities, like those things do need to be considered and our options and not really part of this conversation. As a manager you're trying to like, you honestly want the best for people and sometimes that is for them to leave. But in your role as a company representative, that's not something you want to be like encouraging people to do. 1:15:10.600 --> 1:15:45.660 That's a good point. Although I think, you know, this is maybe less a manager IC thing, individual contributor thing and more of just a senior junior thing. But, but as you move further in your career and build influence, you now kind of have to live a dual life. Like if Jeff Dean went on Twitter. So Jeff Dean, for folks who don't know is I think he's an individual contributor now, but he's, he's been like managing all of Google AI at different points in time. Time and whatever. So someone with like a ton of influence, if he gets on Twitter it says, you know, honestly. 1:15:48.460 --> 1:16:39.390 And I'm just making this up, I just going to pick a random team. You know, Waymo is just such a trash team. Like all the other teams at Google are just so much better. No one really should go to Wayo from Google and we shouldn't take any of their people. If he posted on Twitter it would be devastating. Right. So, so first of all, that's not true. Don't know. That's just a thought experiment, hypothetical. But but, but my point is, like, so Jeff Dean has to kind of live two lives. Like, he has to live his personal life. And maybe he's got an inner circle of people. And in that circle he can say, look, people from Waymo are transferring to Google AI and then failing out. You know, like, we have to, you know, put the kibosh on this. Right? Like, we have to figure out what to do. Um, but then externally he has to say something different, and to his reports, he has to say something different. And so. 1:16:41.390 --> 1:16:59.650 I think that just. I don't know if this really goes into whether you should be a manager or not, but just as you mature as a employee, you start to realize that you do have these two masks that you have to wear. And that's just part of. Part of having influence and responsibility, I think. 1:16:59.810 --> 1:19:09.030 Yeah, the extension of what you're saying is definitely part of choosing to become a manager is you are taking on more of an explicit company representation role. You're the one who has to deliver the bad news. You'll find out the executives love to deliver good news and they love for the managers to roll out bad news. And so lots of the bad news comes in team meetings and is your responsibility to give out. Um, and good news comes from other places. Right. And so that, that's just a inevitability. But there are, there are. You know, we, we've set some native. I mean, there are really good opportunities for managing as well. If you love interpersonal communications, if you like helping shape the team culture. I feel it's not the sole responsibility, but a manager is a huge portion of setting team culture. And personally, I take great joy in helping people and watching them sort of grow into better engineers. And pointing back, like we were talking about AI, you know, hey, five years ago, remember when it was like, this was a struggle or that, and now we're doing X, Y and Z, like, look how awesome that is. And then also getting to hear a little bit more of, like, people's story and a little bit of a different perspective. And I think it's important to realize a manager isn't a dictator. And so you have to build that trust as well. I can't just tell people what to do. I have to work with them to hear them and motivate them to resolve conflicts. And if you derive value in that, being a manager is like a great way to have that be something that counts for your performance review. I've been in situations where I did a lot of that stuff, and I wasn't a Manager. And when it came time to do my performance review, I got criticized because it was like your technical contributions weren't as high as other individual contributors on the team. And I pointed out I do all this other stuff, you know, lubricating interpersonal conflict and resolving issues. They go, yeah, yeah, you do, but that's not your job. And it's like, oh, this is really frustrating. So in that moment, for me, being a individual contributor, that was probably unique to that situation, but moving from individual contributor to manager allowed those things to be claimed on my performance reviews as part of my job. 1:19:11.280 --> 1:19:34.640 Yeah, I mean, this is sort of like tying a couple things together, but this is why, hopefully it's not always like the best coder who becomes the manager. So. So I've heard. I've had people say, I've never actually seen this in real life, but I agree with it. If it was to ever happen. Like, the single worst reason to become a manager is so you can win coding arguments. 1:19:35.440 --> 1:19:37.890 Oh, gosh, I don't know of anyone. 1:19:37.890 --> 1:20:24.770 Who'S ever thought that other than to make fun of it. But. But, yeah, like, this idea that, oh, you know, when I become the manager, then, you know, people will have to write in Haskell or. Or people will have to, you know, do this video boosting idea that I have that is like an unfathomably bad reason to become a manager. The reality is, is when you become a manager, you often lose a lot of your technical credibility. And. And. And that's just. That's just a fact. So. So, yeah, that's. That's the other edge of that sword. But, you know, to Patrick's point, you know, the best relationships I have, work relationships I have, are people who I used to manage. 1:20:26.450 --> 1:20:35.960 And. And especially, you know, people who, if you've managed them and then they've gone on and done amazing things, that's like the best relationship because. 1:20:37.640 --> 1:20:59.960 You know, because you are their manager and. And. And they have that kind of relationship with you, but you're not managing them anymore. And especially if they are a rocket ship, they also have kind of like a peer relationship with you at the same time. And so you're getting kind of honesty but respect at the same time. It's. It's an amazing relationship. 1:21:02.110 --> 1:21:24.510 And I have a friend who. I won't name her or anything, but, you know, she was on my team at Meta and then went on to OpenAI and these other companies, and I was talking with her last week, and it just was like, one of the most fulfilling, like, work conversations. I've ever had. I mean, it was just such a joy and. 1:21:26.110 --> 1:21:36.280 It was such a joy and such an honor to. To have that kind of conversation, which I feel like you would miss out on that if you didn't go the manager route. 1:21:38.280 --> 1:22:01.230 Yeah. I think it is true, though. You know, you do get less time to code. It is less of an important thing when you do it. And to be honest, I think it can be somewhat selfish to spend a lot of time coding and to make yourself a key person, because if some issue comes up, it will take priority. If you have to go resolve a conflict or, you know, deliver slides for a presentation or whatever, and you're the key person on a key deliverable. 1:22:02.910 --> 1:23:13.580 That'S a struggle, and it has to happen sometimes, but it's best avoided. So it is true. And to Jason's point, sometimes you kind of have to lose the argument with other people in order to, like, help the overall team. Whereas when you're an individual contributor, you can kind of grind it out a little longer to try to. To try to win stuff. And then I. I will say I think it is a touchy topic, but being a manager doesn't have to be forever. Like, there are options, especially when you're changing jobs, but even in the same job, although it's much harder to go back to being an individual contributor. So I do think as a manager, it's important to keep your skills brushed up, to keep making some progress in your coding abilities or at least not letting them atrophy. So, you know, make sure that. That that's something you're aware of. But it is difficult within the same team. I've. I've seen it become very awkward before, but definitely, like, with switching teams, it's not something where, if you become a manager, and often when you switch jobs, they don't want you to be a manager. They want you to go back to being individual contributor, at least for a while, and then build up a team. And so I think being aware of that is something that. That it can be a transition. 1:23:14.620 --> 1:23:21.580 Yeah. So, you know, transitioning back to engineer is actually very, very common. It's one of these things that. 1:23:23.420 --> 1:23:37.820 I don't even look at it as a bad thing. As Patrick said, it'll often happen if you change companies, but even, you know, people who are ICs and have management experience, like, bring a whole new dimension to the table. 1:23:40.500 --> 1:23:40.700 Yeah. 1:23:40.700 --> 1:24:29.640 I mean, my mind is kind of like racing with people's names, but I'm not gonna say say any of them, but I know People who, you know, went into management and it just, you know, it wasn't a good fit. Not for them as a person, but just the situation wasn't a good fit. You know, they inherited a team. The team had all these dysfunctions and they switched to IC and that gave them like perspective and time to kind of reflect and everything. And I know one person in particular who now is, is like a really high up person at, at OpenAI, a different person and a director level, you know, and I think that, that it's, it's very common to go from manager to IC and then go back and it, it happens all the time. It's just that there's. 1:24:31.650 --> 1:26:16.790 There'S sort of like a stigma. Like I remember when I switched to manager, a bunch of people congratulated me and it's just, it's just human nature, right? Like I didn't get a promotion, I didn't get any more money, but it's human nature to think, oh, the managers have influence and, and at least in tech they don't really have that much influence but, but the perception is still there. And so conversely, you know, if you switch from being a manager to an ic, people won't want to really acknowledge that. And so people will often not make a big celebration about it or anything. And so you miss, it's like survivor bias. Like you miss all the times people switch to IC and particularly you miss all the times people switch to IC and it was so much better. You know, one of the most famous ones is Somith, the person who is the inventor of Pytorch. He led all of Pytorch and switched to an ic. This is years ago but, but you know, that enabled him to, you know, lead through influence and you know, not have to do some of the management stuff that they found super draining for them. So, so it happens way more often. And I guess the point is, you know, you can switch to manager without it destroying your life if you hate it. So you could always try it, especially if you're kind of just, if it just works out for you to lead a team, maybe it's a team you started, maybe the manager left and you're kind of the natural replacement. It, it behooves you to give it a crack and then find out whether you like it or not. But, but do it knowing that it's a very, very different job and you're to going to have to scale up in a new way. 1:26:21.750 --> 1:26:31.189 Well, hopefully that has been a useful discussion filled with opinions and hot takes. So I'M just kidding. Nothing too spicy. Nothing too spicy. 1:26:31.430 --> 1:26:48.110 That wasn't too bad. I feel like it was pretty positive overall. Cool. All right. This was great. I feel like this is. I can't remember who recommended this, but this is a a really solid topic. Thank you out there. Thanks to the Discord. A lot of Happy Thanksgivings in the Discord, which is really cool to see. 1:26:49.630 --> 1:26:51.950 And we will catch you all next time. 1:27:07.070 --> 1:27:09.150 Music by Eric Barndaler. 1:27:10.730 --> 1:27:11.890 Programming throwdown is. 1:27:11.890 --> 1:27:28.810 Distributed under a Creative Commons Attribution ShareAlike 2.0 License. You're free to share, copy, distribute, transmit the work, to remix, adapt the work, but you must provide attribution to Patrick and I and share alike in kind.