Putting My Copilot CLI To Work - Principal Engineering Manager AMA

Cool stuff. All right. Looks like Looks like Reream has rebranded inside my own stream. That's cool. Um, welcome to the live stream. Uh, I'm doing this one a little bit earlier because I plan on actually spending probably all the time just being in C-Pilot. Um, but if you're familiar with these live streams and I I basically try to do an AMA type of thing. So, I'll be uh, you know, answering questions and stuff, but I'm I'm going to structure this one essentially just to show how I've been using Copilot CLI and uh, really just kind of jump in between a million things. Um, yeah, because that's just how I've been approaching it. So, this is entirely stuff that's like outside of work for what it's worth. Um, so I'm not showing you how I do things at Microsoft. Um, I'm showing you how I build things outside of work. Uh, because I I don't write software at at work. So, I'm just doing a check here because I am running a lot of co-pilot instances and I want to make sure that my computer is not like totally nuked right now. Um, which I mean it's using it's using all the rams. Let's see. Can I kill anything off? I'm just worried that I'm going to start like getting into things and then it's all gone. So, let me close a couple things off. I have a I'll explain I guess a little bit more as we get into things, but I have a separate machine that I'm trying to to run um like build pipelines and stuff on. So, uh I I noticed that when I had too many things going, even just like compiling code was like cranking my my memory usage and my CPU. So, I'm like I got to I got to structure things a little different. So, get that out of here. close just a couple more things off. Okay. So, um the challenging part with this is that I can't exactly show you the full view of what I got going on. And that's because when I'm streaming, so just as an example, let me flip this over. Get a couple things out of the way. So, let me do this. when I stream, just so you're aware, when I go into this full screen mode, right? Oh, sorry. It says full screen on my side. This is actually a a quarter of my monitor. So, um to have like a 1080 stream, it's just one quarter of what I I have going on. So, uh I'm not saying that to be like, oh, it's cuz I'm so cool. I'm saying that because in order for me to stream this stuff to you, um, I kind of have to move a bunch of things around and bring it into view. So, uh, let me let me go ahead and do that because the the whole point what I want to do here is show that, um, I'm I'm very much like just going between lots of things. And, um, it's I don't know. I would argue like it's probably not optimal. It's probably not optimal, but this is how um this is just like how I like to build things uh outside of work. And I think having AI has like enabled like this I don't know like this disfocus problem for me. So let me um try to explain. So um there is a problem right now with uh PowerShell and Copilot that when at least for me when these things are running so you can see like I have a session running here um if I go to dock something else into it. So, if I take this one and dock it in there, um it will crash and it will not only crash and kill the window, it will actually kill uh all of my co-pilot windows and um I have a lot of them running right now. So, that is not going to be okay. Um, so unfortunately I'm gonna have to like kind of do this thing where I um I don't know like you know bring over some other sessions and stuff like usually you can see this one has a bunch of tabs open. Um this is usually like how I'm building and then I have all of my co-pilot tabs in here. And so what I've been starting to do more recently is just like chunking them up a little bit kind of like you might with your browser tabs or something, putting them in the groups. And then I I'll have like this one's primarily just like a whole bunch of uh I don't know almost like developer side projects for me that I I build up. So, I'll go through them in more detail, but like for example, Genesis, if you watch some of my other videos. Um, Genesis is one that I use for um, you know, building up instruction files and things like that, reusable templates. So, I mean, to kick things off, like why don't we just do this, right? So, um, in this session, I'm just literally building out um, more templates for in in this case for cell. So, I wanted to make sure that I had um some reusable templates. I'm going to have to go through and add the instruction files and stuff. But what Genesis lets me do is when I'm working with Copilot and say I want to start a new project or um even if I'm building out some new part of an existing project, if I have reusable pieces, I can basically ask C-Pilot to go look at Genesis. I have to update one of the skills for it, but it can basically scaffold out a like an entire base project with um all of the things that I like to use. And when I like what I'm doing right now in this session is basically adding more things into that scaffolding so that the next time I want to go build something like for example this one says this GitHub issue 126 is uh going to add into the the templating and component system generic object storage and then one of those initial implementations will be using uh forcell right but there's also like Azure blob storage and there's also like uh like S3 three. Yeah, the the Amazon one. I don't know what Google has, but anyway, like I can basic I think even Cloudflare has one. It's probably I think Amazon backed, but anyway, I can just add these in so that when I'm spinning up a new project, um if I'm already working in something that is, I don't know, like based on Azure, then I can more readily select those types of things. So, I'm just going to kick this one off and tell it to go build out 126 cuz I've already had it ahead of time go research and split out a bunch of work items for me. So, this one can just chip away at it. But, I have that I have um I build like a I don't know if people depending on how many people watch these and and tune in for other videos, but I've said for uh for many years that one of the things that I always go back to building is a role playing game. So, I'm just having Cop-ilot basically rebuild this game for me. And so, um, it's just like a little thing on the side. And if you're paying close attention, I have it building out like some of the game engine, uh, in here. And then in this other tab, I have like the lore being built out, right? So I can just have conversations with co-pilot in between things so that this role playing game that I started like 20 years ago that keeps you know I'll build it for a bit and then take a break and then come back to it a couple years later and I'm like I have to rewrite this whole thing. Um right now Copilot's just doing that for me. So I've not I've not seen any of the code for it which is pretty crazy. So, um, I had it I don't know if I showed it on stream a little while ago, but I had it like, uh, building out some of the Unity pieces and then I was flip-flopping back and forth between, um, I don't know, like trying to go 2D versus 3D and I wasn't really happy uh, with either of them. So, kind of what used to happen is I would do that, I would try something, get frustrated, and then be like, I'm just giving up on this, right? I went and coded all of this stuff. I'm no longer happy with it. I'm just frustrated. Going to change. It's a pain in the butt now. Whatever. I'll go do something else. And this time I was just like, "Okay, co-pilot, like get me back to some other part in my history. like take a snapshot of this one, but like um we're either going to go back in time on this branch or we're hopefully if we structured the code properly and we we did this time um I said like everything in this folder, you know, get rid of it. And so that's what we did. And so co-pilot's chipping away building out parts of my game, I'll just move that out of the way. Um I have a bunch more. Um, I recently, speaking of like refactors and being kind of frustrated and stuff, um, I talk about one of these projects I have called Needler a lot. So, if you watch code commute, um, I I give updates on that. I've made a couple of like YouTube tutorials on it. Richard, I know the feeling. Yep. Um, so in on code commute and on those YouTube videos, I talk about needler, which is, um, it's like an opinionated way that I do dependency injection for, um, for scanning types for building plugins. And so, um, I do this kind of thing basically with every C project I make. And I started building this out because I said like, why why keep doing it over and over? I might as well just make it reusable. Um, and I've built it enough times where I'm like, this is the pattern I think I like. So, uh, I built Needler for this purpose. And then as I've been building a lot more, um, like aentic workflows and things like that, trying to use LLMs inside of the things I'm building, I was like, okay, well, the next sort of progression here is I want to have uh, these types of things work really well with with my dependency injection approach. and just kept building. And um the other day I kind of had this realization. I'm like this needler, you know, repository that I have is it's no longer just dependency injection. It's actually becoming more like agent workflows and stuff. And I was like, this is not it's not okay because the whole point of needler was to be dependency injection. And so, um, I had co-pilot over the weekend basically split it all out, right? This is just another example where in the past I probably would have gotten frustrated and been like, man, I've put so much time and effort into this and it's going to take so much time and effort on like the most boring stuff for me. I don't want to go moving code around trying to get it to compile. Like to me that just feels like I'm, you know, pressing pause and then taking, you know, a 100 steps backward and hoping that when I unpause again at full speed. And so Copilot pulled this out. I moved it into something called Foundry. Again, I'm I'm kind of doing a bit of a rapid fire thing here to walk through some stuff. And the point is just to show like on a regular basis like on on a Friday after work and then Saturday and Sunday um like when I'm at my computer I'm kind of hopping between all these things. And so the other thing is of course I build something called brand ghost. Um so I have like these are usually all docked together but like I said I was spinning up these PowerShell windows when the sessions were in progress. And uh I'm not I'm not going to lose it all. So um I was using uh Copilot a little bit earlier for um one of the workflows I had running and something went wrong. And based on what I was seeing, it it was doing something that I wasn't expecting. And so instead of me like just sifting through logs and then trying to comb through stack traces, um I just pointed it at my Grafphana and logging. So I think that's Loki on top of Graphfana. Um and I think I think now I'm just using the Graphfana MCP server, which seems pretty great. And just said like here's um what's it called? uh like basically the the job ID that was running and then had this weird behavior. And so it went and looked it up and then and basically went through all of my code and figured out what happened. Um and so if I were to scroll up in this chat, one sec. I just I should have done a sanity check before recording this. I curse at co-pilot a lot and I don't really want to um I don't know like convey that. I want to be as professional as I can. So um here right so I said look at graphana and uh Loki logs to see what's going on. Um there's some user identifiers and stuff here. Uh these are temporary things anyway. And so it went looking and it said like it found that it did the right thing and then it just pointed out to me like by the way you were running this on a different backend instance. So um the front end was pointing to two different backends and I purposefully like I have not built out distributed caching yet. So this whole thing actually worked. It's just that um the cache was like 60 minutes long. So, I was like, why isn't the stuff showing up? Why isn't it showing up? It was there. Um, I probably should have looked in the database, but uh it's nice because when I'm jumping between all these things, what I can do is actually, um, I can go ask co-pilot to go do that because I don't actually care in the in the most um nice way possible, I guess, right? Like, I don't care to go do that. I know what has to get done. I don't want to do it. I would rather do something else. And I know co-pilot can go do it. So, let me just go delegate it and go do something else. That's kind of the theme with all of this is like let me just focus on the parts that I want to, right? Um I'm not, by the way, encouraging people and saying like this is this is the best way to do it. Um but what I am saying is that uh like I've been programming for a long time. Um, I've been programming for over 20 years. Um, I'm not saying I can't learn new things. I learn new things all the time. But, uh, the I I think I've learned for myself. The part that I enjoy about software development is things being created. And, um, the parts that I don't enjoy are some of the really mechanical parts that I'm like, I'm I don't know, like putting boilerplate code together. Like that's why I'm building some scaffolding libraries for myself. There's a lot of pieces that I don't love spending time on and historically they've been very necessary, but now they're less necessary for me. So I if I have the option, I'm not going to do it, right? Got to add bots to the block list. Um Richard says, "Absolutely. I feel like I'm creating evil AI when I yell at Co-Pilot for doing something I didn't want." Yeah. Um, it's funny because when I use C-Pilot at work, I catch myself because when I'm at home, uh, I talk to Copilot probably in ways that I would never talk to a human being. Uh, and I I mean, if I if I were miked up and talking to it, I think that I don't I don't think I could say that stuff out loud because I get so frustrated with it sometimes. Um, and so when I switch over to my work machine and I'm and I'm actually working with Copilot, I catch myself where I'm like, don't don't say that. Uh I don't I'm sure all the telemetry and stuff for my my messages I've probably opted into a bunch of stuff especially being a Microsoft employee um and and having uh you know certain GitHub connections that I I do with my tooling that I'm sure I've opted in for like yes you can read all of my uh my messages going through it but still I'm I at least feel like I can say hey look that's outside of work and the work stuff I got to keep professional Okay, so let's see how to navigate some of this stuff. I think probably the more interesting part is going to be uh maybe less about Brand Ghost. I want to talk about Foundry and Needler a little bit. Um I'll talk I haven't talked about this one at all. Um this is the the kind of side questing that I'm doing now because um I can build it with co-pilot. So I for a friend when I moved out here to Seattle um their family had uh an issue with recovering stuff from uh their sort of like computer that they had like the family computer when the kids were growing up. So, uh they they they had mentioned that like I can't remember how it was, but like basically haven't haven't seen like childhood uh photos and videos because they're just on this machine uh in like you know in the dad's garage or whatever and like computer doesn't turn on and I was like hold on and this is going back a few years now right when I moved out here I was like hold on like I actually I used to work in digital forensics building digital forensic software. I'm like, I don't have the digital forensic software anymore, but I am a software developer. And I'm pretty sure if I have your hard drive, I can probably recover your your family's pictures and videos. And so I did. I built um I built some, you know, mini digital forensics software to recover pictures and videos. And I was able to get that family back. Uh all like they didn't care about the other stuff that was on there. just they they didn't have their their kids uh photos and videos from growing up. So, um it's kind of crazy to think that, you know, this family had taken childhood photos and videos and had all these memories and the kids had never been able to see it once they grew up, right? Even though they had them, they they couldn't they could never go back and look at their memories. So, uh that was a that was kind of cool. that was kind of special for me to be able to help out with like a I don't know a bit of a niche skill set building software. And so um in this case I was like hey like I actually came back to this it's been I don't know a couple years now like I built this thing and like I could just have co-pilot chip away at parts of it. So um I have it building out like um I don't know like a bit of a digital forensics like recovery tool. Um, and so you can even see in this, right? If uh you're reading some of the text on my screen, it completed a pull request, whatever. Uh, sorry, it completed a GitHub issue and opened this pull request. Um, and you can see like it's saying needler now generates the SQL 8 factories and providers. So like I had it go build stuff with Needler, which is my dependency injection and uh scanning framework. Um, it this one also uses like uh some of the scaffolding and instructions from Genesis. So, I'm going to I'll kind of uh go back over those things, but I figured like talking about this one would be kind of fun. Um, because it is a side quest. Like I what I get to do with this is like tell it how I want things built and what comes next. And for me that is so cool. Um, and I I'll I'll talk about this more when I get into like some of the Genesis pieces um and some of the I mean we'll see it in Needler too. But um these if you're wondering like well Nick why why do you have like the all like so many little side things going on? Like isn't that super wasteful? Like isn't that a waste of time and like focus and energy? And the answer is perhaps maybe yes. Uh but one sec. This is not an advertisement, but if someone wanted to sponsor my videos, like here you go. Do it. Um so, um sorry, question in the chat. How do I like using C-Pilot? Uh CLI Copilot on Windows. Um love it. I am a who knows maybe I will keep converting because this has been the year of change for me. So uh I love using copilot CLI. I used claude code a lot last summer and then kind of moved I was going between like claude code but I'm not like a I don't like terminals which is ironic isn't it? Um I don't like terminals. I'm you know C developer love being in Visual Studio. So I was using uh the like co-pilot agents in the cloud like heavily last year trying to use claude a little bit and like the problem that I was having was that the visual studio agent tooling and I'm talking about like purple visual studio not VS code um it was just behind it was so frustratingly behind that I was like I want to be in Visual Studio know, doing my Visual Studio things and I have to go spin up um, you know, VS Code on the side. I have to go I have to spin up Claude to go do something. I have to spin up cursor and I'm like this is so silly because sure there's co-pilot in Visual Studio and I would I don't know ex is my stream still working? Sorry. On on on my side, the stream that is showing me what's going out was like there's no connection. So, I hope I'm still here. And I know the chat's really delayed, so apologies, but um yeah, hopefully I'm still here. So, it was probably the beginning of this year when I was trying out uh Claude a little bit more. I tried Copilot CLI and I think it's just been long enough for me that I'm not sitting here saying C-Pilot CLI is better than Cloud Code or the other way around. I don't care. Um, but it it got good enough for me that I don't use Visual Studio anymore. There's no point. So, I use Copilot CLI and then it's right to GitHub so I can go read pull requests. Um, and I haven't tried codeex, man. Why do I I keep I've been saying since like March gonna gota get off the co-pilot CLI to force myself to try uh codeex. Still no. Um I'm interested in trying it. I'm not a like I do work at Microsoft. Yes. Um I like using C. I like Visual Studio. I like the Microsoft things and I work there. So, I realized that me talking about Copilot CLI is like, "Oh, it's just cuz you work at Microsoft." No. Um, this is literally the tool that pulled me out of Visual Studio and kept me out. But I'm not saying that because I think that it must be better than every other tool. I'm not saying that. What I am saying is it happens to be the one that got me to stick. So, I do have to keep exploring the other tools. install a VS Code and Codeex right now and then do GPT 56 soul ultra maybe. But but the one thing the one limiting factor is the token usage. And that is one of the other things that's kept me in co-pilot CLI because I have a little bit of a uh special power if you will um which also allows me to do some of the side questing a little bit more. But um codeex interesting. So VS Code still as like the editor and then code but I might and Google searches are are not not it. Sponsored results are one entire page. That's crazy. I know I'm not sharing my screen. Sorry. Not not on that tab, but wow. Very pretty. I feel like this feels like scary to scroll down. Like it's like, oh, it's very pretty. And then I'm like, what? Where are we going? What's Oh, no. There we are. Um, okay. What do we What do we got here? Oh, this is the actual codeex like app, I guess, right? I'm curious. Uh, Freedom, what is uh what's what is it about Codeex for you that like what stands out? Because it sounds like you're using VS Code as the editor if I if I didn't misunderstand. And so what is it about codeex um and and like how it's operating that really stands out to you? Very curious. Um cuz like like I was saying earlier, right, with C-Pilot CLI, I am not trying to say, hey, it's better than everything else. Just happens to be the one that I use. So I'm curious. I had the same conversation with people around claude code earlier this year because people were saying like, oh, like claude is just better. And I was like, "Cool." Like, "Tell me tell me what you're using Claude for. Like, I would love love love to learn." And then they would tell me things and I'm like, "Yeah, but that's not like that's not a Claude specific thing. Like I could list four other tools that do that." And maybe maybe the answer is Claude does it better. And that's totally fair, but but no one was saying that. I think I had maybe two people that were telling me that they were comparing like planning and then building features out with cloud code and uh and co-pilot CLI at one point and and they were saying they felt that process was better and I think that's totally fair that's fine but other people have been saying oh it's this feature like I remember someone saying like you can do slashcontext in cloud code and I'm like you could you could do that and go with CI too. I don't know what the I don't know like um I took goen and built flutter dart back and made alm local host encrypted chess agent cool that's awesome um codeex models better the actual models trained oh interesting um but you're weren't you saying that okay are you saying codeex but because you were also saying in the chat too like GBT 5.6 Six Soul Ultra. Um because like what I was a few code commute videos I've been doing. By the way, there's a few different YouTube channels I have. So if you're I know I don't think anyone on this chat that's active right now is on YouTube, but um this streams to the Devleer podcast YouTube channel, but I have one called Code Commute. And when I've been doing some AI updates on that, I was talking about switching over to the the 5.6 six soul uh model and honestly like I've been super impressed with it. I was stuck on the Opus models. I think 5.1 for GPT is the last GPT model I used and I was like anything after I'm like ah they suck, man. Like however I'm using them it's not it's not working. And I tried this one and I was like holy crap. Like this is this is pretty sweet. Um the codeex model of GPT56 Ultra is different. Oh, cool. Very neat. And then Richard saying GBD505 can solve system issues that co-pilot could not. But I'm just curious cuz like uh like what what is uh I'm curious what is different, right? Because even Richard saying 5.5 can solve system issues that co-pilot could not. But co-pilot using what model? like using the same model. Again, I'm not my I don't want to sound like I'm challenging and and and saying you're not right or something. I'm just I'm just curious because I I want to if I don't ask the questions, I'm not learning because I'm not doing right. So, it's important for me to ask. Uh I was a dude who built the origins of music models and the chatb's original voice. Oh, cool. That's awesome. And yeah, please uh Richard, uh you said make some videos showing me design my agent stack. Yeah. Yeah, super cool. I think that's awesome. Um yeah, this is cool. But yeah, I really do need to look at codeex, right? Because uh I found last year when I was playing around with different things, at least I was like getting to see how the tools were evolving like and I can't use them all because I have to stick to something for a little bit or else it's just all over the place. Um, but now I'm just so entrenched in co-pilot CLI that I'm like I I know I'm missing out on on other things that are happening. So, um, okay, let's jump over here. So, um, yeah, back, you know, on this one, this is where I'm going to pull up the GitHub stuff in just a moment just to show you. But, uh, this is, like I said, a bit of a side quest that's been going on. And if I can Where's my Whoa, that's too many things. I got Again, you can't you can't see what I'm seeing. So, there's too many screens that are pulled up and I want to find the right screen. It's this one. So, I'm going to pull this over. Not my personal email. We'll get rid of that. Um, I want to pull that back up. That's there. I got to pull up the stream again so we don't lose that because I need to see when that's not working. Um, okay. So, let's jump over to this. This is another side quest. So, this is called Maestro. And so I know and if you've watched code commute videos I have like a problem with like constantly apologizing or like disclaimer around this. Um I know there are lots of companies and smarter people than me that are making like agent orchestration software. So disclaimer uh I'm doing the same thing for fun because I'm trying to learn about it. Oh no. Um, it's obviously not working perfectly. So, uh, I'm I'm building this for fun to learn about it and trying to get it to the point where I can basically assign it work to do. And I I had the day off work. So, uh, I see Brad just joined the chat. Brad, I promise I I had the day off work. Uh, so I was getting this to the point again where I could assign an actual uh, work to do. Um, clearly the dashboard's not working for it though. So, uh, the whole sort of meta concept here was that if I can get this to a point where it can run workers to improve itself, like that seems pretty sweet to me. So, uh, that's what it's doing. So, if I jump over to here, uh, there's a bunch of issues that are filed on GitHub. And I designed it so that it's sort of, uh, extensible. You don't have to use it with GitHub. It can be, you don't have to have anything. You can just use it standalone. You could I could build a Jira connector or Trello, anything else. But um so this is using GitHub. And so basically when the work items are marked to be picked up or like ready to process by agents, uh they'll just take them off the queue. And so there's other parts to Maestro which will actually split up work items. So you could file an issue, you could have it go get um sort of disassembled into other work items and then agents will pick them up and uh yeah then the idea is really that we can have like a self-improving kind of orchestration system. Just checking the chat again. Uh Brad says a special sauce is the harness and the model. Yeah, so using Opus and Claude code harness is different experience using Opus GitHub Copilot. Yep, I find GitHub copilot is so uh close now on many aspects that I rarely change harness. Yeah, that's that's been the the thing for me, which is why I I constantly will say like I am not making the statement like co-pilot CLI is the best thing that's ever happened. I am making the statement that it's been the thing to keep me out of Visual Studio. And for me to go switch to using another tool, what I'm what I'm acknowledging is that I I basically have to pause what I'm doing, which is really difficult. I have to pause what I'm doing, maybe take a step back, bit of a learning curve, and there's nothing wrong with that. In fact, that's probably a good thing, but I find like I I'm not carving out the time to do that. And so I kind of just stick to what's working and it keeps getting better as I'm using it. So makes it kind of tricky. Um Richard says your CLI method of working with code is very interesting. I really want to do that workflow now. It's uh it's I don't know. It's just how I've been uh operating I guess. Uh Freedom says if Microsoft want to clone Codex into Copilot, it's not too hard. Just needs a few updates like playright tooling. Um oh like directly into Copilot. Yeah. Uh a bit more pathing entropy into the prompt system like okay takes uh free gentic loop pay some employees and go build skill matrix large vector DB store giant skill matrix. Yeah, I mean the it's interesting because like we have like Microsoft is very big, right? So, um I I feel so far removed from the people that are working on co-pilot just because the area that I'm in is not that and it's huge. And so it it's just kind of crazy to think that like I'm in Microsoft and it's still it still feels like um it almost feels like it's a different company. That's how how big it is, right? But that's also because I I worked at smaller companies for my entire life before Microsoft. So uh I know that if there's someone on the on that team that I wanted to message, it's just, you know, it's a a search away and I could send a message to them. But the dogs. Your dog is not eaten. Uh she was not around to eat outside just getting checked on for the dogs. Um what was I saying? Come on, brain. Oh, talking about organization size. Yeah. So, uh it's ju it's actually super interesting to see though that within my org and and slightly broader org just the amount of like I like ideas and things that like keep coming up because it's like we can do this now. We can try this. Like we should we should build things this way and that's going to help these people. It's going to help. It's just like an explosion almost of like uh of cool things happening. So, I don't know. I think it's super exciting. But, um okay. So, this what I'm showing is Maestro. It needs a lot of work, but um if I pull up not that, we'll go back over to Maestro. So, it said there's a manual. Okay. So the thing that I was encountering today is because it's doing the pull request as my user which I should probably set up something else. Uh it can't use the I approve and then it automatically merges kind of mode. Uh so it it has to it has to be manually merged or I have to go set up some other users. So um it says five issues are eligible. It's running these two waiting for capacity on these other two. So interesting. um like it's not actually working on anything right now and if you think about it without knowing any details but if you think about it um I set up two workers it's running two things so if they did finish if they did then they can't continue because they're waiting for me now maybe I could change the scheduling such that the workers that are waiting on poll requests could they could maybe go dormant and then other ones pick up new work or whatever so we can get through more pipeline but I don't know like I get to make those decisions. I get to build it and see what happens. So, uh it's been a pretty cool experience. So, I'm going to pull up this one and this one and I'll see. Sorry, one sec. That's Why didn't that open? There we go. Okay. So, it has these two issues that it says it's working on. And you can see it says Maestro Ready as a as a label. This one also says Maestro Ready. Um, they have relationships. Like, by the way, I didn't do any of this, which is cool. Um, I literally said to Copilot like, "Go make me uh like go file these bugs. Go file these bugs and then split them up um as needed." But it looks like if we go to pull requests, I would have expected that two were sitting here, right? Because I have two workers. It says nothing's being done. So, where where's the other poll request? I might have to go ask co-pilot and see what's up. But this is entirely built by I mean I didn't build a special harness or anything. This is just running co-pilot. But that's the other thing with Maestro is that I made it so that the actual worker agent is pluggable. So if you wanted you could have a fleet of co-pilot CLI. You could have a cloud agent, you could have claude, you could have codecs, you could do whatever you want. Um so I just built it with Copilot CLI. kind of want to get it going end to end working with something and then I can swap out pieces and see how they perform. So this is a a PR that was done by AI. So this is an example of how I spend part of my time and then the other part is like just coming up with the things I want built. So if I were reading through this um in particular, this is uh code that's on the dashboard, right? So there's some tests. So I'd be curious what's going on here. But on the dashboard part, I don't really care to be honest. Just it's just front end stuff, right? That's why it's not working when I was demoing it. But um yeah, a lot of the stuff on the front end I don't care too much. And that's primarily because as a as someone who's not actually writing a lot of front-end code or historically has not um the tricky part for me reading this is like I don't have a lot of I don't know experience being able to say oh like if you start building things going down this path you're going to regret it whereas if I read some C code I probably could have some opinions. So, um, there's some stuff in in front end where I'm like, I don't know if it works and there's test coverage. I feel okay. And that's just going to take me more and more time to be seeing code like that. Seeing what types of things kind of screw me over over time. Hey, like why do we keep coming back to this file that always seems broken? Like, oh wait, here's some crappy patterns. And the more of those that I learn, the more of those that I will bring back to um that repository I was talking about called Genesis. so that uh I sync those instruction files across all of my projects and the next time I scaffold a new one, they come with all of those instructions. So, kind of like the the knowledge that I gain uh from from doing all this stuff just kind of gets redistributed back across everything else I'm doing. Uh Freedom saying, "I've been trying to convince Microsoft to put a a real human like voice in Windows." Yeah, you know what? Um, I have another little side quest and uh I think I was I was playing around like I hadn't done anything with voice stuff like uh text to speech and speech to text and so I have a little app that I was putting together and I was like oh you know have to get the models and if I want to run this on a phone like what are my options? I want some flexibility and I want it so that you know when you if you don't want to pay for things then like in the worst case we can still have something that works on the phone with local models and then you can kind of scale it right so if you're like I I have resources I want to run a bigger better model great uh if you want to spend money and run something better great we just swap over to that and so one of the options it was using like the built-in uh like something that came with Maui I guess by default and I was like oh my goodness like this is not this is not okay. Um it sounds like I'm back in like I don't know what year it was using Bonsai Buddy or whatever that that thing was that Purple Monkey or whatever. I don't know. Um I don't know if anyone remembers I and Microsoft Sam I guess is the other thing but holy yeah yeah it was it was not good especially if you compare it to um like the co-pilot app or like chat GPT and and talking with AI that way is like that's a pretty crazy experience. So um but I didn't read the rest of the message. Sorry. Uh but Cortana's voice from Halo with problem is cost you 80. Yeah. Um yeah, I mean it's it's super cool that they can actually do that kind of stuff though. Um so the image generator model hooked up with a specttogram style bark AI model. What's uh what's bark? I haven't heard of that. alongside click typing input mapping system feeding the world generator. Microsoft starts becoming OS layer, application layer. Yeah, I mean interesting, right? Just imagine Microsoft world word built like this. Microsoft Word built like this. Yeah. User says to model make Microsoft Word look like Halo 2. Yeah. I mean it's like at a obviously at a very different scale, right? But like kind of the things that I'm going to be continuing to walk through here are just things where I was like I want a whatever make it make it the way I want it. Like it's it's neat that we can do it. Um feature of code is kind of like video diffusion models generating entire apps on command with click map. Yeah. Um future software software list just advanced ML models generating frame by frame. Yeah. What's interesting about this is like I can't help but imagine that I think there are parts that go this way. And the part that I haven't sort of figured out is building stuff this way is like what's a good way to say this? The the it's it's accelerated of course, right? So we can do way more complex things in an accelerated fashion, but it's also like I find that we're we've undone like so many years and layers of optimization. for example, um like there are some things that I do now where I can just tell C-Pilot to go do it and it's blasting tokens, but if I would have like just manually gone to some URL and clicked a button somewhere, like it, you know, would have cost no tokens. It would have taken me personally a couple of extra seconds or something. But like I I find there's so many things like this that there's a lot of opportunity to go optimize and I'm just so curious like at what point do we stop going like just more more and start going okay like how do we dial this how do we dial this thing back and get back to efficiency because I I don't know I don't I don't see efficiency. I know there's token caching and some other stuff, but oh my goodness, like there's I think I think it's pretty uh you know, a lot of opportunities there. Yeah, it's uh it's it's I think it's going to be interesting. I think the future's pretty bright. So, in this case, like what I would do with this PR is um Oh, I was saying I wanted to go check out what it had in here because this whole issue was about some websocket stuff that was uh not working well. Maybe that's why I had that issue with the dashboard disconnecting. I don't know. Um so, okay, it's running some tests. This setup as sync part, is that actually new? If it's not, okay, it's not new. So, I feel okay about that because it's not recreating that. And then creating a chat web socket. That's fine. Um, it just kind of refactored some pieces out and added a new test. So, that's that's cool. Um, on the actual host, what is it doing? Okay. Like, I don't have strong feelings about that. It's kind of getting like a bit of a versioned API or something going on here, but it's fine. Whatever. We'll merge it. Um, and so now because things were stuck, I don't know if Maestro is going to go pick it back up. I'm not convinced. Pardon me. I'm not convinced that it will. So, I might have to go bug C-pilot and say, "Hey, like remember that time I told you we should have the agents actually working on this stuff." Um, they're not. So, what's going on? Oh, something just refreshed here. And one of the work items I I filed for like it made this whole dashboard originally, but like why is there a Q but also issues? I don't understand why these two things are the same. So, um, page object model isn't implemented. That could be a bad sign. What on the PR? We'll find out soon. Um, so where's my session? I got too many of these. Okay, so I noticed only one PR was ready. I just urged it. Um, aren't there supposed to be two workers picking up? So, I would check in with it and then what I would do at this point is go do something else. So, um, Narnia is this other thing that I have built. Uh, and I can show you that while it's going. So, this is a little C-pilot CLI dashboard that I built for myself. I think a lot of these features are actually um like I started this before Copilot CLI had some of these things. So, it's a little dashboard. Um what's the point? Like why don't I just go to Copilot? Well, because that little issue I told you about earlier, um, where if I have a bunch of these terminals like this, right? Um, if I take all these terminals that I'm working in, right, there's more that are hiding behind. If I dock one of these and it crashes all of them because, spoiler alert, that happens to me regularly. Um, I want to be able to restore them, kind of like what happens with your tabs. And so I know that you can do d-respilot CLI. But the problem with that, the problem with that is that when I have 25 copilot CLI sessions, I am not going to d-res a bunch of them. So um I have this where I can go favorite some sessions. Um and then these are across a bunch of different things that I will go back and keep working on. uh I can launch them, right? So I can launch all of these. I can save them as a group. So I can just launch that as a single uh PowerShell window in the future. Add to collections very similar. Just kind of lets me group things, right? So when I build brand ghost across different things, um often when I'm working on brand ghost, I have a few work trees going in parallel. So I have a few dedicated co-pilot sessions just for some work trees. Um, what I've switched over to more recently is, and I'm trying it out. I'm not saying that this is better or not. Um, I'm just kind of sticking to the same session, uh, for most of what I'm doing. If it's like a one-off thing, then fine, new session, toss it. Uh, but I'm sticking to the same once and compacting it back down to to nothing. And so, why? Don't really know. Just been kind of getting in the habit of that that when I finish up work, I just compact. Um, if I'm doing a series of related work, then I compact and obviously keep some type of dedicated focus to continue on that workstream, but I've been just kind of sticking to the same session. So, that's why I have them saved out like this. Um, and it seems to to work pretty well for me. Uh, Steph says, "On system testing, page object model enables uh better testing organization of code each code." Oh, I see what you're saying in terms of how the front end code was set up. Each code used in testing flows and uh segmented into page files from there could be used without Oh, cool. Yeah, I mean I'm going to have to go look that up. The front end stuff just changes magically for me. So, um like I said, it's uh it's a gap for me. Uh what else is interesting here? I'm going to talk about schedules. I'm going to talk about storage. So, um, storage was something I put I I put together, Copilot put together recently, um, because I had this little issue where if I let me do this, I don't know, things that are older than 10 days. There's only I've cleaned up some of them, so there's only bit over a gig used, but um, showing one to 100 of 2500 sessions. Um, so when I was running some agent evals, it was ballooning sessions like crazy. I'll show you a different page that has a graph of this, but um, a combination of that happening, a combination of when I was doing the voice uh, stuff and it was doing work trees and things like that, it was cloning down these these voice models and just like dumping them in different places. And so I cleaned up like 60 gigabytes of just like voice models that weren't being used by anything. But um I built this out so that I could actually go clean up old sessions. And so uh Copilot did the whole thing. The first version of this that it did was terrible. Uh I'm not saying this is amazing, but it basically had uh instead of this workflow being like one, two, three, it was like everything al together, no ordering. The filtering was at the bottom of the screen and then you'd have this list. And so I was like co-pilot like I said I think the the message to co-pilot was something like it looks like someone had a hundred really good ideas like sir sorry sorry there's 100 people each with a good idea and no one talked and someone slammed them all together. Um so you need to make this like a cohesive workflow. So now it actually does the filtering shows you what's a candidate and then I can go through and say like okay I'm going to pick all of these. I'm gonna review the cleanup plan, right? And then it gives me some information. If there's some stuff like protected as if I favored something, then it would it would not go clean that up. Um it would let me know. So, um just a little quality of life thing because again, something I periodically do. And if I wanted to, I could go put that on a schedule. Um, and again, this stuff's probably in Copilot CLI already by default now, but just I can go build it. Why not? So, um, I have some schedules that run for me. Um, these are basically just skills that I, um, that I regularly need to run. So, for example, um, I actually have a meta skill on a schedule. And so, that's this Genesis evolution radar. So, um, can I see it? Wow. I don't know what the special funky characters are, but um so what this one will do is it actually has a list of all my repositories. And uh it will on a schedule go check all of these repositories, look at how the instruction files are used there, and then see if there's patterns that are interesting, and then it will I'll make it more automatic to go to GitHub. But for now, it's emailing me, I think every other day. It does this audit to see if the instructions in these repositories are drifting, if there's any new patterns, because if there are, then I should basically bring that back to Genesis because that is my central spot for having common instruction files. So, if I'm like, hey, look, like, you know, this one of these apps that I was building, what's a good example? This uh this forensic one, right? This one was doing some crazy stuff where um almost every PR it's updating its own uh globased instruction files. So the ones that target specific paths, it's updating them. And so I'm like in the context of this repository, sure that makes sense. But if that's happening enough across these different projects, then this scheduled job is going to run and it's going to go, hey, I see these patterns emerging. like this is maybe a way that you could bring this back to Genesis. And then the idea would be that instead of it just emailing me, it would file GitHub issues files as issues. Uh I'm still going to be the person that's saying whether or not that makes sense because some of the suggestions I'm like, I see what you're trying to do, but no, we're not going to go with it. But um yeah, it um I gotta be careful because I don't know if there's like secrets and stuff like printed into the log files on my computer. But yeah, um that runs and and emails me. Um I have some stuff that will go do log analysis and brand ghost. I have uh one that will do my uh I have one of my newsletters is just like an article and video summary. So that will go run and basically put uh a whole article together that has links to all of the stuff that uh was sent out during the week and that's just run on a schedule. So um that's just something that I did in Narnia because I was scheduling stuff. I know you can do it in the co-pilot CLI, but like I can't tell what's where and so I'm sure they will build a better system, but it's the same thing. Like I know everyone's obsessed with terminals, but man, like when we're doing this kind of thing and now you have to build a little UI into your terminal, like you're you're building a UI into the terminal. Why don't you just have a UI? I just I don't get it. Um, but I know people like to be elitist with their terminals and that's fine. Do what you want to do. Uh, as someone who's used Visual Studio my entire life. So, uh, I don't know. I kind of like having a UI for things. But this is this is Narnia. It's a little tool I built to help me navigate my Copilot sessions. And if you're wondering why it's called Narnia, it's because the database for Copilot that has all the session information was called the Chronicle database. And I thought it would be funny if it was like the Chronicles of Narnia. So, um, another little thing that I've done is that my projects get these documentation pages. So, if you go to GitHub, right, it's on GitHub. So, actually, if you wanted the Narnia dashboard, you can go get it from GitHub. Um, but I I have them basically nested on my website, which is fun. So, they have all the the documentation out like this. And this is all built by Copilot, right? So, when I'm updating open-source software that I'm distributing and I have dedicated documentation, it kind of all just happens together. And then I have it nested under my site. So, if you go to website and I go to projects. Um, that's just caching because I'm an admin. So, the things that I have here like I can go see them like this foundry one, right? And so I just have co-pilot go do that. So, I kind of can build out my little I don't know, portfolio, fun little tools and stuff like these Google MP servers cuz for some reason Google I'm sure they must have by now. I don't know, but um Google didn't have some MCP servers for like search console and keyword planner and page speed insights. So, like I just built them, right? And I had um I had co-pilot do it in different languages. I lost uh all my dark mode when I moved these things over. My goodness. So like if I go to the repo for all of these, right? like this is entirely like vibe coded by by copilot and it's written inn net and there's a go variation too and they're built in parallel and I've never seen the code but um so I have these MCP servers and then I I started getting people that were like using them I guess I know it's like 24 stars like whatever it's not nothing but I I don't know I thought I made it for me I made I know I made it public but like I put this together because I was using it and then some other people started using it. So, kind of cool. Um, so yeah, I I built these just because I needed them and you know, didn't have to do anything because Copilot did it. Um, Stefan, uh, what is advantage here compared to Graphana Stack? What is the Sorry, what's the question though? Uh, what's advantage for where? I think I missed that part and it's probably just because the chat's super super delayed, especially on LinkedIn. So, I apologize, but if you can clarify where um let me know because maybe it was the Narnia one, but I can't tell. Um what else on here? We could talk about Pit Crew now. So, Pit Crew is another um thing that came up for me because I needed some local runners. I have I have an old laptop that's sitting beside me. you can't see it. But um more recently because I've been running so many sessions, what ends up happening is that like co-pilot sessions, I think I mentioned earlier on the stream when it's building code like so not even just like doing AI things when it's compiling literally building code. Um it was just using all the resources on my machine. So everything was grinding to a halt and sometimes to the point like I can't even move my mouse like I can't I can't be here. So uh I was like I should just offload this stuff to my other machine using local runners. And then I kind of went down this rabbit hole. I was like co-pilot hey like what are some you know uh do some research and tell me what some options are. And I was reading through and I'm like I don't really love the things I'm seeing um for whatever reason. I'm just not a huge fan. So I said, "Go make me a local runner and I want to manage a pool and then I want to be able to scale them up and like down the rabbit hole." Right now I want to observe them. Why? I don't know. It's literally running right here. If I look here, I can literally see Docker Desktop running on my Windows machine just because I had this machine. Um, but like I don't know if I want to check on them when I'm not at home. That would be cool. So I I had it go build this. Why not? Um, oh, and Steph and so the Narnia thing is it's less of an observability tool and more of like a just a session management tool. So um the this schedule job things aside like the the fact that I have this session list I go back over to like collections or my favorites. These are just because I lose them. So if I if I end up losing my co-pilot windows and I want to get back to something, say I have to do updates and I want to restore 23 different co-pilot windows, it's just so I can have access to them. That's all that's how it started. Um so even like this, if I I didn't show the stats page, but the stats that it gets are basically nothing. Um it's just stuff that's in this uh this Chronicle database already. So, like I I was like, "Oh, it'd be cool to see how many tokens there are." It doesn't there's no tokens tracked in that. There's probably tokens somewhere else if I tried, but you can see on here um like this huge spike up in sessions. And like why did that happen? This is because I was building some agentic workflows and then I needed to run evals on them. And so those eval were like being run and judged with the co-pilot CLI. So it ran tons of them. It ran way too many. Uh so it kept going up, right? But that's it's just an outlier because of of that. Um and probably any spike you see like this is probably because I was doing evals to be honest. Yeah. These um these aren't super interesting. Um I mean this chart just shows you a different variation of what was above um where the sessions come from. But yeah, I don't know. It'd be cool to see like the token usage and stuff like that. Current streak. These are just like, you know, whatever co-pilot decided to put on here. So I don't really use this, but just kind of interesting, I guess. Yeah. Not not observability. Um, Graphfana is probably I mean Grafana Graphana is amazing. I love Graphfana. So I don't have uh for my co-pilot stuff. I don't have that for like brand ghost and when I'm when I'm running like actual um like production agentic workflows I have that instrumented with graphana. So, um I don't think I can pull up a dashboard easily, but um yeah, like I have I have them for my like my background services running in the cloud. I have them for agent workflows running so I can see like what's going on. But yeah, I graphana's so good. I would not that's one that I'm not going to try to to go remake uh because that works for me out of the box. In fact, um it works for me so well out of the box that if I can find my co-pilot terminal, this one, um one of the things that I was asking Genesis to do, I filed an issue earlier. Um it was it's doing a bunch of Versel templates and components and things like that that I can reuse. And one of the things it said was uh what did it say? Oh, you know what? That wasn't Graphana. I lied. That was amplitude. But it was building something for V vers Versel that was like feature flagging, but it's doing it in a generic way and then you can drop in feature flagging and I was like, "Hey, like go drop in amplitude." Um, I don't know if I have Graphfana might be one of the things I did right at the beginning of Genesis because I use Graphana a lot. Like I really like using it so it's probably in there. Um, how are we doing with Maestro? Right. This is my my ADHD can let me kind of jump around. Yes, both workers operated correctly. It says they started concurrently. 103 succeed at creating merge PR, but 101 105 107 109 each hit exactly 30 minutes retain dirty output. Oh, so it was screwing up. There are zero active workers because retained output claims prevent duplicate redispatch. Oh, interesting. Only one PR appeared because 103 completed. So this is where I would say do you um how do I ask this? Because I find with this soul model if I say um if I say did you do an analysis on on why it will say no I didn't. So now I have to change how I think about things and it's not did you it's like go do it. Um but I'll try. Did you do an to understand why um 101 105 107 and 109 hit their 30 minute cap and so I would ask it and then I would go do something else. Right. So um pit crew is the the thing I was talking about for my um what's it called? for the local runners. So, I don't have a link to it on my dashboard. What's a good way to get there? Dev leader. Did I put it on here? It's not even It's not even on Oh, it is there. Right. So, this is the repo for Pit Crew. By the way, Chad GPT is so good at making pictures now. I remember like we all know like the the Will Smith eating spaghetti pictures. Like the the old times were not okay. Nightmare fuel. And then it was like, "Oh, it's good." But it was still I found it was still bad in chat GBT. Like the the models to make things were good, but the experience of like I remember typing to it like no like take the words off of the sign or whatever and it would like give me the exact same picture and I'm like what is going on here? Um it is so good now. I can't believe it. So uh if you haven't used Chat GBT for making pictures recently I would say definitely go back and try it. Um let me see. Uh this is what world model concept deployed. You can try it already live. Done. One sec. Let me see what the link is. Making realtime model from happy oyster. Interesting. Oh, is there there is a link? One sec. Create a world. What? What are we What are we going to make? What? What are we What are we going to make? I don't even know. Describe your world in real time. I'm like, there's too many options like Can we do something with what's this directing venture? This isn't you're It doesn't make an app though. This makes scenes, right? Freedom. I think you're saying you're talking about what you would want, but this one makes scenes because I can see all these. Can I just click one of these? What? Oh, yeah. That I mean, I want to prompt it. I just don't know what to do. What does it Okay, so you can't see me pressing things, but it's either super laggy or Okay, it's super laggy. Like it's not doing the thing. What's Megapunnel do? I'm nervous. I'm not pressing anything, so I don't know like what's happening. Just press press space to jump. Okay, I'm going to say it looked really good, but control-wise, like absolutely not. Um, oh, whatever this is. Yep, giant mech. Let's go. It's just super laggy, but like it does work. Oh, yeah. It's just it's just laggy. Oh, it's jumping finally. Is it like is the LLM doing the controls? Like is that why it's so laggy? Smash tower go. What does it do? Okay. I don't know. But this is still super sweet. Um, yeah. I I think that's pretty fascinating. That's cool. It's a a bit of a like a janky demo, but like the I'm sure like if I knew what to type, I'm sure we'd see it do build something cool. Um, hey, infected FPS. I haven't seen you in a while. Totally legally distinct Gundam. Yeah, definitely. Definitely not that. Um, can we do like a fluffy cat snowboarding? Uh, very fat and very fluffy cat snowboarding down a mountain. Let's go. Yeah. Driving a jet ski on Mars. After you wait for me to press enter. Come on. We'll give it a shot. Let's see. Come into life. I think the other thing to consider here too is like the speed of this is uh it's pretty insane, right? Like Oh yeah. This Okay. Puff up. Is this real life? Oh my goodness. Okay, turn around, buddy. Come on. We're stuck. Puff up. This is amazing. Can we get some air? Oh, we're done. We ju we just did that. Oh, sorry to hear about that infected FPS. Well, welcome back. Uh, sorry to hear about the situation. I Yeah, that that's cool. I don't I don't know what to say. Like that is just the the part the part part that's taking away from the experience for me is that it's like the controls are so laggy. I don't know why, but that's cool. Yeah, that's that's super cool. Um I got to get rid of that because the music's going to drive me nuts. Um where do we get to with this thing? What did it did it finish an investigation? still going. Right. So, this is again why if I'm just leaving stuff like this and I have a bunch of other things going on, I'm like whatever. Like I'll go do other things. This is one of those things where like the token usage not worth it. Um but I'm going to do it anyway. So, this is an app I'm doing that is uh yet another voice assistant. Uh, but this one's going to be so that I can uh control co uh my goodness control GitHub. And so that way I I was telling people this on code commute except I don't think I've published the video yet. Maybe I have. Um what I'm going to do is when I have this app working, I'm going to demo it. And basically when I have all of these projects that I have going on on my commute or doing whatever uh instead of me like I'm obviously not going to code when I'm driving because it's insanity, but so that I I can keep tabs on things or if I have an idea because this is the most common one for me is I'll have an idea when I'm driving and I'm like, "Oh, that would be really neat." but I'm also driving a car and I'm going to be 40 minutes before I'm where I got to be and like I just whatever idea is gone forever but it was fun to think about. So, uh, with this Laya app, because it's a voice assistant, what I'm going to do is basically have it so that I can, you know, push to talk, uh, or leave it in a mode that's just running and that way I can I can say like, hey, go make a GitHub issue about this on this repository or, um, hey, like check if there's any open pull requests. um you know were any critical you know reviews or any critical comments left open that kind of thing and if not merge them blah blah blah and some of these pieces for me will come together a little bit more when things like maestro are working um but just again it's like it's kind of for fun right I like I was saying earlier I haven't done much of anything with models for speech to text and text to speech so this is kind of cool because I know nothing about them literally nothing about them. So, it was really neat to see, you know, just some exposure to to a little bit of it where like the difference between the models and how they perform, like not knowing anything about them. You were to ask me like, you know, between Opus 48 and GPT56 Soul versus Sonnet, like I'd have opinions about those things for coding, but zero experience with these other models. So, just been a good learning experience. Um, but you can see why this is such a terrible use of tokens, right? And time. I said launch the app and we're still waiting here. Not okay. But also, I think if I check my computer, like I'm at 90% CPU and 70% memory consumed. So, there's there's some things going on. Um, probably due for a little reboot some point here, too. But anyway, that's that. So, what else we got going on, though? Can we Can we build anything? Um, how did we do on this one? Branches rebased to-do completed. Okay, it was just lagging. So, it opened this PR rebased. Let's check it out. 49 files changed. So, this is going to be a big sweeping change. But if I do a quick check through here, there's not a lot of real estate when I have have it like this. Okay, it's kind of across a bunch of things. Primarily this SQLite one as I figured, which is fine. Um, so what had to get done for this is that We got a a factory provider instance singleton. So it it made some progress on these, but it didn't take it all the way. Uh what it was doing was if I look at one of these, um it was basically avoiding all of my dependency injection patterns. And so it was making factories that look like this. And if there's one thing I don't want in my code, it's this kind of nonsense. I cannot stand this. Um, yes, great. It did used composition. It made smaller classes. Very nice. Um, probably a bit of a smell if this many dependencies have to be jammed into one thing. Different story. We can figure that out. But what's even worse is that like it just kept perpetuating this pattern and then it was also making like a a singleton version of it. And like if I click over here, here's another one where it did the same thing. If I click over here, here's a another one where it did the same thing. And I was like, "Look, man." I think almost quite literally said, "Look, man. Um, stop. Stop doing this because we have patterns for this that are better. And so I don't know if it actually did it. Can we find one of these? Can we find one? Start at the top. No, it's not going to be in the test. It's not. It's It'll be in here somewhere. Oh, provider. I don't even remember building this part in needler. Okay, that's not the one I want, though. It doesn't need So, this is wrong. It doesn't need to do this for what it's worth. Like, that's just the wrong thing to do. uh my my dependency injection stuff by default registers things as singletons because I often build classes without state. So there's literally no point to make more instances of them. They don't have state. They're just holders for methods. Like I just organize things into classes. Um so like this is redundant basically. It already does this by default. Uh what I'm looking for is this one. this generator. So what's cool about this I I I think this is super cool is that this is a source generator and how it will work is that um it makes a factory class and interface. So this this nonsense that you see is basically generated for you. And the other cool part to it I shouldn't have clicked that because now I lost where it was. Um, the other cool part to it, like it missed this one. This is terrible. Um, the other cool part to it when it does it is that if you have this big stupid list of all these things that have to get uh dependency injected. Say you had all of these services which should just come from your your service provider by default automatically resolved, but there's like one parameter that comes in for some reason. You have it like that, right? Um you have all these things, but you want to pass in a version as a string or a number. I don't know. You can make a factory that says, "Hey, look like all of these things get dependency injected. Sorry, get dependency injected." And then for this one, you know, parameter at the end, when you say, you know, factory.create, the only thing you have to pass in is that one parameter. So, um, that allows me to write really like slim down factories automatically just by by doing this. So, this one's kind of weird cuz like it shouldn't have kept this and I don't know why it did. Weird. Um or or what it you know what this one's a parser. It's not the factory. So um let me clarify. This actually says to not register for the dependency container. So you can't resolve one of these things. And so if you want one, you have to go ask for the factory that makes it. That's why. So it left this to be really messy because no one ever has to go do it. Um, I personally like using primary constructors for this. The C world is divided on this, but I like primary constructors because if you see this massive list of disgustingness, uh, we can collapse it down into into one-third of it. The difference is that in this repository, co-pilot wanted to start doing this for everything. Um, let me put guard clauses on everything. Nah, like not for me. But it started doing it. But if I got rid of this, this entire class would be reduced down to almost nothing. But I can see there's already there's still some code smells in here for me cuz I I don't like that it's like it's making new instances of things inside of here. Like something something's wrong. But uh I'm not going to on this repository. It's all incremental. So if this is net better, sure, like I'll I'll merge this in if it's not busted. And um then what I'll do is once I see the pattern continue to emerge and I have enough examples, I will say get rid of this. We need to go make an instruction to to go clean it up. Um and then go refactor. And then again, that way the rest of my repositories learn to like not do that pattern. And if I do find like there's some limitation. So, just to give you an example, uh I think I think there's like a an attribute we can say that's like not null. Um and I actually I think there's even a project setting where we can basically get some of this effect. Like this isn't adding a lot of value in my opinion. It's pretty much useless. Uh especially if we're using dependency injection for it. It's not going to matter anyway. If we wanted this kind of thing and for some reason the language uh didn't support in a nice way with dependency injection, I could go add that into Needler. I could say, "Hey, look, there's this gap and I see it in this repository. Let me go file a feature into Needler. That way, all of the spots across all of my code that do this kind of terrible thing, they get cleaned up over time." Uh posh, yes, I do work at Microsoft. Um, Steph says, "I'm wondering if we can't control opening closing sessions because AI workers are processing on their own, why is there a dashboard for making more efficient prompting?" Um, Stephan, is this back to the Narnia thing? I'm wondering if we can't control opening and closing sessions because AI workers are processing on their own, why is there a dashboard? I'm not sure I understand the question. Are they going Okay. Hey, uh, Posh, I'll give you one chance here. So, either you don't do that or I ban you. And I am assuming you probably don't care, but like, fair warning, I guess. I think everyone deserves a warning unless you're a spam bot, but up to you. Um, so yeah, I think in this case, this one is net better. I would take this as long as this passed. And if we go back and check. Nope, still busted. Let's see. One more. man. So, what I'm going to do in this case is I'll take that. I'm just going to go back to here. I'm going to go. It's not this one, is it? We were in this. Um, uh, you need to fix this issue as well on the PR. Yeah. So, Stefan, uh, so your question, we can't control opening closing sessions because AI workers are processing on their own. Why is there a dashboard? Um, the for me, the dashboard is just for when I lose them almost exclusively. So, I don't I don't want to show you because I will lose all of my terminals and I'll have to go restore all of them. Um, but if I take one of these windows, actually, let me take this one. If I take this tab and I drag and drop it into here while it's running, it will crash. Um, oh, there was a graph for number of sessions used by an agent. What is the purpose of the graph if we know we can't control AI agents? Uh, I don't know this uh probably on this page. Oh, I'm I'm already on there. Um, yeah, I mean this is just like a a dashboard co-pilot put together from the data that's available. for number of sessions used by an AI agent. So these aren't sessions by an agent I don't think like this is just the number of raw sessions. So I I don't know maybe it was a misrepresentation of what the data is but this is just count of sessions. Um, again, total raw sessions, but like I don't these are just kind of like vanity things that are taken right out of the database. Like I I literally I go to uh either Windows. So you can see these are the ones that are open. I have 20 sessions open right now. And so I should double check here. So some of this is still kind of busted. It's funny because it automatically gets this assignment and so it took it like it's not in this repository. This is a repository that was mentioned in the chat and when I asked C-Pilot about this I said hey you have to fix this bug. It was suggesting to me that this is actually inside the database that it's like reading from co-pilot. So this uh repository when I was in my foundry session, if I pull it over, um I was saying like I want you to go uh review this repository because it has a lot of like workflow patterns for agents, right? So I was like, go review this because there's probably some really cool ideas in here. We should see like what the surface area is like, what we're missing. And so um I pointed to this repository, but that's not that's not what this repository is. So what I can do actually is change it and I can say like that's not what it is. It's actually Nazentino Foundry, right? And I can save that. And then if I go back when I'm looking at my Well, I'll go back to Windows actually. Um now I don't have a weird I don't know. It's same thing with this one, right? Um, can I pull this up? Where's the session? Uh, here. I was having this conversation. Just want to show you because if you look at what it's doing, this one says Nexus Lab Framework, which is this co-pilot tab I have, right? It was saying that the the repo was this Andrew Lock one strongly typed ID, but that's not. But I was having a conversation about strongly typed ID and I said like I want to see if I can uh do something where instead of just a normal GID, we have a uh a GID like a V7 UYU ID so that we can have like an incrementing GID instead of a totally random one. Um, and so I was just having this conversation with it, but like that is not the right repository. So I would just go back in here, save that, and I could give it like a different name or whatever, but when I launch it again, so if I had to go back into this Windows to relaunch it, I could go do it. You can even see like on this homebase one which I haven't talked about homebase. Um on this homebase one uh I I wanted to update the the repo image, right? So homebase I basically built my own link tree because I could and link tree costs money and I was like I can build a better one. So I have my own open source link tree. This is uh what you're watching right now. And it's cool because like I didn't make the update here to do that. Like it just gets pulled automatically, right? And so then my open source projects and stuff are here as well. It's got um my links, a shop tab. So anyway, that's that is homebase. So if I go back to the links, there should be a link to homebase itself and now it has an icon. So that's what this session was when it started, right? Was about taking an icon from chat GPT and dropping it in there. So that's why if we look at Narnia, for whatever reason, Copilot named the session that. So I would go name it something different. I would say this is home base. Uh terminal title is just homebase. The repo is already right. I would just put main here. Boop. There we go. And so I can just keep a little organized set of sessions like that. And yeah, Stephan, I think uh to your to your question though, right? Like raw sessions. Yeah, so it is raw sessions. Um there probably is um there probably is some other data in there. And I think the the last time I touched the stats page was just to correct something because it was like so wildly useless, but most of it's pretty useless currently. So um I would be I really wanted to see like uh model usage right so now that this sole model is doing a lot of parallel code reviews and stuff do we see like is it doing a lot of different model usage right I see it's spinning up a code reviewer that's a sonnet or opus or or something else I don't know I don't know why it's picking those but I'd be really curious to see like the model usage and stuff um even if it was just number of times if they can't tell me tokens I think it would be super cool. Uh, yeah. So, that's that's Narnia. It's not it's not like I don't know. Nothing I'm showing you is like I'm not trying to make it sound like it's groundbreaking. It's not. And that's kind of the point. The point is actually just like these are things I wanted when I was building things and I can just go make them. Uh, Maestro was still running, right? So, yes. Now, I had not done this analysis before. I'm glad it went and did it. Um that was a mistake only. This one was genuinely session sized. So uh issue 101 expanded to 19 paths replaced full start another full gate 26. So basically um these workers are capped at 30 minutes and like that's configurable but they're capped at 30 minutes. Uh it probably was just going too slow and it's probably because I don't have resources on my machine to do anything. So, um, it sounds like it tried to do stuff, failed, and then like apparently the way that this orchestration's done and is configurable is that when it ran out, when it ran out of available work, because it tried this, it like ran out of work and couldn't schedule more. But when each item failed, it basically kept it out of rotation, so nothing else could pick it back up. And maybe we want to change that. Maybe when a worker fails, right? Maybe I want the default behavior to be some other worker tries. And maybe I can have a back off where it's like, you know, one worker tries fails and then another worker picks it up and if two fail, then like nope, like leave it. Maybe that's good. But I want these types of things to be configurable. I'll have my own opinionated configuration of it, but just leave it. Um, Stefan, could anyone join the open source projects? I would assume you're one of the admins. Yeah. So, everything that I'm showing you is mine. So, yeah. I mean, I I think if there's stuff that you want to contribute, you can. Uh, Posh, so you were saying doesn't want to talk about Microsoft. I will talk about Microsoft, but I think that the way you're approaching it's maybe not uh not conducive to a like a productive conversation. So, you are here's what's happening. You're coming in asking a question, but you're also insinuating like kind of what you want me to say, right? So, you came in and you had some derogatory thing to say, which I I don't appreciate, but then you're shifting it to we're going to be replaced. So, um if you want to if you want me to talk about layoffs, what I think about that, I'm happy to do that. But this is why I was kind of saying like I think it needs to be approached in a different way. So, I'm not going to talk about uh what you said at the beginning because I think that that's not appropriate. I don't think that should be something people are saying. Uh not only is it like offensive, it's just like wildly inaccurate, not really grounded in anything, but like primarily because it's offensive. And so if you think about this, um, just for what it's worth, right, because I'm happy to have conversations. When you do that, you are number one, uh, you're offending me because I don't appreciate that. You're offending other people that might happen to watch this. They probably don't appreciate it. But the other thing that's challenging about that is that like now it's in my chat and you are trying to get me to engage with it. I don't want to engage with that because I think it's offensive. So that's why I said the options are like don't do it or I block you. Right? So if you want to talk about layoffs, if you want to talk about the impact of AI, why layoffs are happening, you want my perspective on that, I'm happy to talk about that. But what I'm not happy to talk about is you setting something up that is offensive and then trying to get me to engage with it. Does that make sense? And I'm not I'm not saying this in like a condescending way. I'm being genuine because if you want me to talk about that, I am happy to have a conversation about it. But just that's that's where I'm coming from on it, right? I'm I don't engage with the earlier comments because I think they're offensive and I think that that's inappropriate. Doesn't Microsoft have more people in India, etc. than America and UK? I actually don't know how many people Microsoft has in different parts of the world to be honest. Um and it's tricky too because um when we talk like when we talk about Microsoft in general right Microsoft is enormous. So my my like world view of Microsoft is probably not fair in terms of a representation. So the part of Microsoft I work in is called Microsoft 365. And so that would be like um and and in particular like on the platforms. So I don't build um like uh apps and things like that. So like for example uh you know Excel online or uh Outlook web client I don't build those. And so there are teams that do, right? And some of those teams, they're they're in the US, they're probably distributed across the world, but um if you were to go through the different teams that build things in Microsoft 365, like there are hundreds of teams and that's that's just Microsoft 365. And so I'm not even that like I am I'm the platform. And so that's called substrate at Microsoft. The office the Microsoft 365 platform for people to build on at Microsoft is called substrate. So I when I started at Microsoft I used to work on the deployment team. So again on the platform. So we built deployment infrastructure so that all of the services that build on the platform they have deployment that is done for them that not every one of the hundreds of teams has to go cool time for us to go figure out how we get whatever we made sent out like there's a central team for that it's a platform now we work on the routing plane so similar concept but not for deployment if we need to get traffic to services um on like a planetary scale Ideally, the service owners don't have to think about that from square one, right? But so I work on the platform for Microsoft 365. My worldview of that um if I had to guess, I think most people in Microsoft 365 I think are in the US, but I don't know. I have half of my team half of my team roughly I think is in Mexico actually. So um I got uh you know Mexico and a person in Costa Rica. So I actually have a handful of people in Latin America and same when I was on the deployment team prior. Um but yeah I I actually don't know for Microsoft 365. I I think I think primarily US. But what's also interesting too is like so I've been at Microsoft for it'll be six years in in a month or so and I actually have like limited this is going to probably sound funny given your original comment but um I have like limited experience working with people in India. I work like ex like heavily with people in China, right? So like for the past six years I've been set up with teams where like we are um on different like opposite sides of the world um but like rarely overlap with India. I do work with many people who are from India but not um like like I mean they're in the office with me. So uh what do I think about layoffs? I think layoffs are dumb. I don't like them. Um, I I think there's so many I don't know like there's so many things that come from layoffs that are like chaotic, confusing, don't make a lot of sense. Um, I'm just like purely like I just don't agree with them. So, um, I think there's a lot of confu like rightfully so, I think there's a lot of confusion that comes from from layoffs and especially at different points in time, right? So depending on the company, depending on the amount of layoffs, depending on whatever um there is so much left for interpretation as to why, right? So for a while, and I'm I'm going to talk about them in general, you know, the the Microsoft specific stuff. I might narrow it in on that, but in general, um you would see companies, especially over the past couple years, like we're just going to lay people off because we can replace them with AI. Right. So there's there's been this fear of like hey look the models are getting good all the like I do use AI to write all of my code but like if I were to think about that at work like absolutely not. I would still want people on my teams if they can use AI to write all of their code. Hell yeah. But um you know it's not I wouldn't be like okay everyone time to go cuz I'm just going to sit here with my 23 co-pilot sessions and code things. Absolutely not. If they want to go spin up 23 co-pilot sessions and they want to go manage their own co-pilot workforce. Sure. I don't I'm fine with that. But I need people uh and I need people that are on a range of uh experience levels. And so I talk about this on code commute a lot, but like it to me makes absolutely no sense to, you know, you lay people off so you don't need humans, right? Because we're going to replace them with AI and then and or like we just get rid of the juniors because like we don't need junior developers. We can just have like the seniors. And it's like that also doesn't make sense because like you literally can't just have a team of people who are senior just just do senior things. Like it's just not going to scale. What's going like what's going to happen is that those people will gain seniority and like there won't be humans to come back fill them. They won't be people to ramp up and learn and everything you need for continuity. And you might say, well, yeah, because there's AI for that. And I would say, great. Like, go look at all of the companies who've had so much success doing that. Um, because it doesn't work. And so maybe, maybe someday, maybe someday we don't need people doing anything. I don't think we're there. I think we're pretty far off. And so I think a lot of the the rationale that gets put behind it is probably incorrect. And I'm not saying that companies have not done that because they absolutely have and I think it backfires and doesn't work. So either you have companies doing it for a really ridiculous reason and it backfires or you have companies doing it for another reason and then people assume that's what it's for. Right? So, I can't sit here um I can't sit here and say like this is why Microsoft or some other company did layoffs for the same reason that like anyone who is telling you that they probably also don't know and they're just making assumptions. So, I think a couple of things are a reality when it comes to layoffs. Generally speaking, um like there is some type of uh effect on stock, right? investors love layoffs for some reason. Um, but I think primarily this this framing I think I was told this in a general sense from someone at Microsoft uh and it it stuck with me because I think it makes a lot of sense. So, um, ever since I've been at Microsoft, it's been kind of like weird times because like I started during the middle or not the middle, near the beginning of the pandemic. And so there was like a pretty crazy boom at the beginning, right? Like is everything in tech? It was like, oh my god, like everyone's remote, there's all this cloud stuff. So pretty crazy. And then it went to like, wait, wait, wait, wait, wait. Like everything's getting weird now. So in terms of like people like slow down on the hiring then there's been layoffs and there's been these weird like dynamics going on. So there's been people internally at Microsoft, rightfully so, going, "Hey, like what the heck is going on with this, right?" And like turning to leadership and being like, "Hey, so um we should probably talk about this, right? Because you realize a lot of people are scared." And so the I'm not going to do this justice, but how this was kind of shaped for me was when it comes to um sort of these circumstances around layoffs in a like to oversimplify it. If you were to look at uh call it like for just simplicity like profit margin what happens is that the decision is not like it's not like SA sits there and goes hm like let me just get the Excel list of all the people here and just draw this line and okay bye. Um doesn't work that way. So the way that I heard this is that it's primarily more around like business units and specifically at what level and what granularity I'm not sure but like I said Microsoft is enormous. So there are there are still parts of Microsoft that I learn about like almost on a daily basis. So these different business units, the people that are leading those business units are basically in a situation where it's like, okay, it's either find a way to increase the revenue or find a way to reduce the cost because again to oversimplify it, those are the things that affect the profit. And so if you had an opportunity as one of these leaders to say, "Hey, I'm pretty sure if we do X, we could just generate a lot more revenue for like not more cost." I think they would do that. I think they would probably already be doing that. And so what ends up happening is like well what's forget this is going to sound terrible and I'm not saying that I side with this but if if we think about that situation specifically what is the quickest way to save money and I'm not saying I agree with it as a full disclaimer. The quickest way to save money is to not pay people. Right? So I believe that people are put into this situation where like they're unable to for whether it's bad business planning, whether it's whatever, like you could come up with a list of a million things, but in this situation, the the most sort of effective way to get out of that situation is like if we don't have as many people to pay, we save money. Now, I think personally that if you find yourself in that situation, I can't first of all, I can't imagine having to make a decision like that cuz that seems terrible. But if you were to find yourself in that situation, I would hope I would hope that leading up to that, you have already been trying to do everything you possibly can so that the business is in good shape, that it's healthy in terms of like how you're hiring, in terms of turnover rate, in terms of everything else, so that you don't have to go, hm, I guess guess we're kind of screwed right now. I I would personally if I needed to find myself in a situation like that, I can't can't imagine having to lay people off, if that was like the only thing that was going to keep a business from imploding and I had to make such a decision. Uh I would want to make sure that like if I hadn't done everything I possibly could have leading up to that point, I don't think I could sleep like ever again. So, I genuinely don't know what it would be like to make such a decision, but I I don't think it's uh made easily. I don't think that it comes from the top down as a mandate like just get rid of people. Um I think that decision is left up to different parts of the business for how things get done. The reason why I know at least at least part of this is true is because I do know that there are business leaders uh that I have worked under that have different mechanisms for trying to buffer this. And so without getting into details like there are ways that if you understand that this is an ongoing type of dynamic that uh by the way like something like layoffs layoffs happen they're not new. The volume of them and the scale is pretty staggering currently but like for a for layoffs to occur in some capacity is not a new thing. uh and even if they were done on a very very small scale that is mostly like a sort of natural thing in business unfortunately. So if you understand that that is an ongoing thing and then you understand other parts of the business and can anticipate and buffer some of these things better when it comes to a decision like hey how are we going to balance these things? If you've already been doing some buffering, then like the amount of balancing you have to do is hopefully um significantly less. Uh sometimes you can't increase revenue. Agreed. Uh there's always a ceiling when your market share is already Yeah, I agree. Right. And that's why I'm saying like I think that's kind of what's happening, right? And I'm not I'm not saying that I agree with it. I will keep saying that out loud because I don't want my words to be twisted. But I think in a situation like that, it's like, hey, look, if I could increase the revenue and like for for minimal extra overhead, like I would have been doing that. It's a business, right? Like I would be actively doing that. Uh unless I came up with some brilliant plan um would be actively doing it. So I think that people find themselves in this situation where they have not created enough of a buffer for the business to be able to absorb it in some other way. So, um there are lots of things going on and I don't think personally um I say this as someone in a management position. I'm not like I'm I've been an engineering manager for like 13 years now as a frontline manager. So, I'm not like overseeing enormous organizations or something like that. that in the conversations I have with people who have significantly more direct reports than me, there's no um there's no like uh what's the right way to say this? Like uh like how do we how do we start getting people out so we can replace them with agents kind of thing. Like it's just that's not it's not the conversation. It is significantly more around cool like we got some really awesome tools. How do we start making sure that people are using them effectively? What are the effective ways that we can use these? How do we help people become more and more effective with them? It's I have never heard in remotely, you know, close to an example of like great now that people are slightly more effective, we can start getting rid of them. Like just it's not a thing. Um again, I'm not saying other companies or maybe even other parts of Microsoft I don't know. I'm not saying that this doesn't happen in different places. I just have not experienced anything like that. So, um yeah, I don't know. I hope that kind of helps paint a bit of a picture. I try to be as transparent as I can about things because like I don't have anything to hide about it. Um everything I'm saying is not like not like a a secret like someone at Microsoft told me and like don't say this and it's top secret. But I think unfortunately what happens is that it's so easy for for media, for social media, for things to get out there that it's like it's you can see something that looks like it makes sense. You have people sharing it. It's like, well, that's got to that's got to be it, right? Like no one's given us another reason. So like I think there's lots of other reasons that aren't just replaced by AI. Um, how many managers are above you before getting to the CEO level? Um, good question. We had a bit of a reorg and I don't know if someone was kind of bumped out like five or so, I think. I think five and then SA call it six. I don't know. um something like that. But again, like in my in our part of the business, it's like I don't I actually don't know how I can't rationalize there must be some other interesting reporting structure at the very top. Um because there's there's so many parts of the business that I'm like this even though I'm telling you like the tree like looks very tall. Um the the width of that tree at the top I'm like I I can't understand this. It's it's on our org chart though. I could go look it up I guess. But um yeah, I think there's five. So So yeah, that's my take on the layoffs. I don't like them. I think it's dumb. Um, I you know I as I say all this like try to imagine if I was running a business like if I what would I do if I ever found myself in a situation like that? I I don't know. Um, you know, part of me wants to say if I was running a company and it was something like, you know, uh, I'm certainly not like Microsoft obviously, but if I was running a company and I had employees, like if there was some way for me to take like, you know, less salary or something to keep people employed, like I think I would want to do that. But, um, I just I just don't know. Um I don't I I can't speak to the scale of some of these things and the example I was just giving for myself like those business leaders even if they were to like you know give up their salary and stuff it's I think for some of the size of the cuts that are happening it just wouldn't it wouldn't make a difference. So yeah, uh why do you need that many? Uh the I can't the organization's huge. It's just enormous. Um yeah, I I don't I don't know how to explain it. Like I have I think 11 direct reports. Um and I have three um sort of headcount that I'm kind of waiting to to work out. So I would have like almost 14 people reporting to me and I can tell you as a manager uh it's basically I have three sub teams and if that was 14 people across three sub teams I can't do that simply can't uh not effectively right so if I had that many people trying to be effective across that many places um it just wouldn't be I don't think it would work properly and so there's a natural point where it scales and I would say that could I ride that line for a while like absolutely but at some point because what doesn't happen is like less and less responsibility or less and less things to do there's always more and so it would reach a tipping point where it's like hey look if you're trying to make more happen I can't participate because you're going to see it kind of come apart for me I won't be able to do more it makes it for me it makes it harder to be innovative and forwardlooking if I'm trying to uh manage wider and wider. That makes sense. It's just like too many too many things to be keeping on top of for that to work effectively. Um why not work for another company or start your own? Yeah, it's a great question. Uh I do I do like Microsoft. I really like what I do. I work in a security space. So I work for I work on the routing plane. So all of the traffic that flows through Microsoft 365, that's my team. So that's really cool. Um, interesting technology. Uh, there's a huge overlap with security, which is really interesting for me because I worked in digital forensics. It's not the same thing, but I like I like the space. It's pretty neat. Um, so working for another company, I don't know. I don't like I would be comfortable saying this to like my boss, right? Like my my skip level manager was on this stream earlier. Am I am I going to be at Microsoft my entire life? Like I I would say probably not. But also, can I say like hm like I'm trying to go here or trying to go there? Like also no. Um like I don't know. Is Google interesting? Yep. Is Meta interesting? Yep. Is Anthropic doing cool stuff? Yep. Um, so it's like I think there are other places or other opportunities, but uh I'm I'm not in a position in my role with the things I'm doing where I I'm looking at those other things going like let me say it a different way. Um I'm not like I need to get out kind of thing. And I think for some people because everyone's different this way. I think some people if they saw an opportunity, so let's pick like anthropic or open AI, right? Because everything's AI, AI, AI. If I was like, hey, the stuff I'm doing at work is routing traffic. I really want to be closer to AI. Like maybe that's an internal transfer at Microsoft. Maybe that's something I do, right? Maybe it's like, hey, you know, uh, Meta is doing some cool stuff or Google or Anthropic or OpenAI. maybe I try to go move over there. So, some people I find are much more motivated because they they see an opportunity and they're like, I just I want to go try for that. Things things are good here, things could be great right here, but like the the thought of that opportunity is more appealing. I'm going to go try for that. And I think I'm different that way. Um, and I'm not saying this is better or worse. I'm just different. I think for myself I'm very much like um I will say like let's imagine right hypothetically say things are good today and then six months from now things are less good another six months are less good um I probably wait for things to not be good before I'm like okay I need out I don't think that's great I think that's probably uh side effect of like it's it it is more comfortable So, I would be more comfortable and I would wait until it's uncomfortable before making decisions. I don't think that's a good move, but I think that's kind of how my brain is wired. So, um to answer your question, are there lots of other cool companies? Yes, absolutely. But I like what I'm doing and it's not nothing to kind of force me out of doing that. Um and then or start my own. I I think that at some point that is like the you know that that is the future that I want when what I don't know so I build brand ghost on the side this is my you know social media scheduling platform that I have um I do this like on the side I've been doing this for a while and it's all like I started doing it because of creating content right and then just sell it as a platform so um I do have that, but like is that that's not a full-time thing. That's just on the side. Um, how does my day look when I get up in the morning? Oh, good question. So, um, it's not really that exciting, but uh, I wake up at 5:30. Uh, used to be 5, but it's 5:30 now. I get to sleep in a little bit. Go to CrossFit for 6:30 in the morning. Um, which pains me to say because I was into bodybuilding for like 20 years and my wife does CrossFit and so I do CrossFit now very early in the morning. Um, more recently I will go to CrossFit, I come back, I have a little bit of time in the morning depending on the day I might drive to the office which is it's hit and miss. Could be 25 minutes, could be 60 minutes. So, um, sometimes I'll I'll I'll risk it and kind of go in like right after rush hour, but I'm usually starting working somewhere around like 8 in the morning uh to 9 depends on on the start of the day. Uh, and then either I drive in like 9 to 10 kind of thing or um more recently I've just been working in the morning and then I drive into to work at lunch. But I only have to go in a few days of the week. Um, so I kind of use my lunchtime to do that more recently. And then uh like what am I doing at work depends on the day. So uh some days are really heavy with one-on-one meetings with my employees. Uh other days are uh a little bit more open so that if I have to schedule things with like partner teams and other stuff, then I can have those conversations. Uh, I jump on to code reviews and stuff in between things. Uh, design review sessions, things like that. Uh, my my Thursdays are crazy. Like there are one ones and then I have a bunch of team sync meetings because like I was saying, I manage like three sub teams. So, it's kind of all over the place. Like Thursdays for me start at 8 in the morning and they end at 9 at night. So, uh, that's a really busy day. But yeah, um us it's like the day the day is roughly like 8 to 6 for me depending on what's going on. But that includes like you know I will commute in the middle of the day. So saying 8 to 6 but maybe there's you know 25 to 60 minutes in the middle there that's me not working. I'm just driving. So depends roughly like that. Um yeah and then I try to make sure like depending on the morning if there's any like agent work I can kick off like early on then I will do that and then when I come back in the evening or if I'm working for home and I stop work at the end of the day then I can jump more into coding and stuff because uh as you can see if I have like 20 23 tabs kind of go between stuff I just I like to build things. It's also the reason why I haven't been doing like other YouTube videos though because I don't make time for it anymore. But yeah, I think I'm over time now for the stream. This has been 2 hours and 15 minutes. I'm going to I'm going to drop off here and say hello to my wife. So, I didn't feed her dog. Her dog's very picky when it comes to food. She's a very big dog and she's picky with her food. So, um, if I feed the other two dogs, she kind of looks at me and she's like, "Nope." And she just leaves, but as soon as my wife comes home, she'll eat the food. So, so yeah, there's that. But anyway, I think that's it, folks. Um, so yeah, if there's any of the projects you want to check them out, feel free to. It's just on my GitHub. There is nothing that is, you know, like groundbreaking or whatever. like it's just these are things that I build on the side and just wanted to give you a little bit of insight into how I use co-pilot. So primarily um across a bunch of different things and then I try to take the learnings from those things and pull them back into shared libraries, shared instructions and then templates to reuse and that way the next time I spin up a new project it's ideally starting from the most recent sort of like standard that I have. Uh, and then Posh, uh, do I use Linux as a daily driver? No, I don't use Linux for anything. Um, I guess Linux gets used when I'm running Docker containers, but I don't use Linux for anything. Um, just not not against it. Just am content with Windows and it's doing great. Um, if I set up like my my server right now is my old laptop, which is off the the screen here. Um, and that's like an old Windows machine. Is it a good idea to use a a laptop as a server? No. Um, but I also just have the machine, so I use it. But yeah, if I were to stand up like a a rack, um, I'd probably do a Linux install and have it all containerized and stuff, but don't uh don't have such a thing. Maybe when I move houses or something, get a little server rack. I'd have to ask my wife about that. I don't know how that would go, but cool. I think that's it though, folks. So, thanks for being here next week. Um, I won't be on because my in-laws are visiting, so I'll be uh hosting them. But, uh, for what it's worth, these streams are generally 7:00 p.m. to 8:00 p.m. Pacific on Mondays. Just like I said, next week won't be in. If, um, if you're interested, a couple of things just to quickly show. So, I mentioned a few times. Uh I have this YouTube channel called Code Commute. Um so if you like the career stuff like software engineering in general, uh code commute is where I would recommend checking out. You can leave comments on videos if you have questions or you can go to codecomute.com and you could check this little button here and submit stuff and then I get an email and I can make a video response for your question. And like you can see like I just use my commute time to go do that. Um and then my main channel is called Dev Leader. This is the one that I'm behind on tutorials for. So, um you can go ahead and check that one out for uh some earlier AI tutorials. I'm way overdue. And programming stuff on this channel is primarily in C. And then if you want to watch these uh live streams on YouTube, it's just Dev Leader Podcast. It's another YouTube channel. So, um you can check that one out if you're interested. Uh but otherwise, yeah, like I post all the social media platforms. So, if you are on social media and you want to see the things I'm doing or talking about, just look for dev leader. Um, and that's the sort of branding I go by. So, thanks for being here. Um, one last one. Posh, uh, any feelings about Microsoft telemetry? I don't know particularly what stack you're talking about. Uh, like application insights maybe. Um, I like application insights, but I also just like open telemetry and graphana. So I think application insights works really well if you're building a .NET stack and you kind of just want a paved path for getting uh analytics and stuff. But yeah, I don't know if that was your question, but that's my answer. I got to go though. So thank you so much and I do appreciate the questions. Thanks for turning that around to something that was more positive. I appreciate that. Take care. See you next not next week, two weeks.