Here are ways AI is actually HELPING me in software development

Transcript is auto-generated and may contain errors.

Hey folks, I am leaving axe throwing which is a lot of fun. We did a little team building event. It's like a million degrees with this car is shooting out hot air. Don't do that. Um yeah, we did axe throwing. It was super fun. I haven't done this in ages. I'm not good at it, but it was cool. Um I was scrolling through experienced devs. I'm going to go um and sorry I'm kind of in this spot where I'm not totally familiar. And uh so through all their experience devs I saw a post that was like on uh sharing things in AI that are actually working well. This person not going at all. Okay. And I figured I could chat through that. Um try to think about a couple more examples from building and brand ghost and otherwise that I think are are kind of cool.

Um talk a little bit about my my current flow for building things. Uh this is all outside of work by the way. So for folks that are new to the channel. I'm an engineering manager at Microsoft. I'm not writing code in my role. And uh I've been programming for like 20 plus years though. So love to code, love to build stuff outside of work. But um this is all going to be the context is this is on the side not for my day job. So and a friendly reminder that this channel is driven by your comments. So if you have questions leave them below um or send them into dev leader on social media. It's my main YouTube channel and I'm happy to try and answer software engineering and career related questions. So uh to start off for some of the things that are working well, I'm going to share one that's totally weird and out there.

Um, I've been meaning to do this for a long time. So, uh, a lot of folks may not know, but, uh, one of the projects I talk about building, literally from when I started programming is a little role playing game. And it's the kind of thing where like over the last 20 plus years, I rebuild it, rewrite it, refactor it. It's never it's never going to be done. And it's just fun, right? It's I enjoy building it. Um, it's always served as like a playground for me to learn things and just kind of like build. So, um, at one point in the past like 5 years or so, I said, "Hey, I should, you know, I should really start trying to define the game world, right? Like I've been I just love building the role playing game systems, like how items work, how you can apply stats and things like that.

I just like love nerding out about it." if you're familiar with games like Diablo or especially Path of Exile now in terms of the complexity. Um like I like building complex systems like that like in the game mechanics. So it's just fun to build and I said okay I probably should start taking some time to make some of the game world, right? Like if it's going to be a game I need to have there's got to be a story. It can't just be a bunch of game mechanics tossed together. So, what I started doing was at the time I was just looking for like I wanted to build a wiki with the information and um there's some stuff that's in the actual game data that I was like, hey, I can I can actually use code to go publish some of this stuff into the wiki.

So, really cool that I can have like data driven parts of all of the content. And basically I needed to find a way that I was going to be able to to merge some of those things. But long story short, haven't worked on this, you know, in a good few years now. And I was like, crap. You know what? The last time or the last time I thought about this really was when Chad GBT first came out. I was like, you know what would be really helpful if I could like and this is even before I understood things like rag or anything else. I was like it would be so cool if I could basically have something like Chappie Chap. I'm like if I could have it build some of the game world for me, like some of the lore, some of the story and stuff like that.

But at the time I was like, "How the heck am I ever going to do that? I'm not going to sit there and just copy paste like pages and pages of the wiki into this chat thing." And the context like seemed like it was super small. Anyway, I'm like maybe this isn't the tool. But if we fast forward a little while now, I'm like, I actually think that we could do a lot here. And one of the things that came to mind was I've been talking about using GitHub copilot with pull requests. So I decided I was going to take my self-hosted wiki. I'm going to transform that into a um like a GitHub repository with wiki pages like basically just a bunch of uh markdown files that are all linked together.

And I said, "Then I can ask Copilot to like go build parts of the game world or format and like link all this stuff." So, um, I was like, I have no idea if this is going to come close to working, right? No idea. And so, I imported all this stuff into into GitHub just like um these wiki pages converted into from a mark up format to mark down. I don't know what the difference is, but they're marked down now. And then all of the links were broken. And all of the hierarchy was broken, too. It was just a It was like 200 files in the same directory. So I said, "Copilot, I need you to go organize these into folders." And I gave it an idea of the types of things that were there. Like there's um you know regions in the world or there's like um religion or there's there's people right like like there's uh historic moments things like that.

I just gave it some ideas, not all of them. And I said, "I need you to organize these things into folders." And it did it very easily, right? And then I said, "Now that you've organized it into folders, I need you to update the links." So then it updated all the links. And then I said, "Hey, those pages that you were touching, they're actually categorized." But I said, "Now the problem is there used to be like these category pages that you could go click like people and then it would list all the people." And I'm like, "That's broken now. There's no landing page with all the people." So told co-pilot that. And I'm I'm telling you almost exactly what I told it to do. Like super highlevel just like I'm not telling you how to do it. Just you know there used to be category pages that listed all the things for easy navigation.

Please go do that. And it does it very effectively. So right now I have all of this game content. Um and I say game content it's just like lore right for world building. I have that in a GitHub repository. Copilot's activated there. And what I still have to do is try to actually get Copilot to like maybe I'll get it to do some like storyline writing or something. And because you can toggle settings in the firewall, I should be able to have Copilot go out to the internet if it needs to. Um cuz maybe there's some I don't know some I try to think of what would be good examples of like websites or things where I could go draw some like inspiration from. Um but the reality is just given LLM like it might be trained on a lot of like like medieval fantasy kind of stuff cuz it's a like that kind of role playing game.

So anyway, I might play around with that more just so that I don't have this project that like literally in the background could be having the game world built up more and more. How like from my perspective, how cool would it be if I just have it do a bunch of writing and I just go read through the poll requests and I'm like, "Hey, I like this, you know, this lore that it added." And I'm reading through it and I'm like, "That sounds cool. Let me add it or tweak it." or just outright like, "No, this thing sucks. Like, I don't like it. Get rid of it." It's I just have to read now. So, I'm kind of excited to try it and we'll see. But so far so good. Anyway, it's totally It seems like a very irrelevant example. I just wanted to tell you how I'm thinking about using Copilot in this case.

Um I still have been using GitHub copilot with pull requests like extremely actively um with brand ghost in the front end and backend code. The the front-end stuff has been mostly small changes and that's because I am less familiar with uh TypeScript in general, less familiar with uh React or with Nex.js. Um, and I'm less familiar with our front-end code. So, I don't want to give it something complicated and then I don't even know how I'm going to go verify it. Right? So, some of the things I've given it are like small quality of life things and it's it's just done a tremendous job. Um, to the point where like I probably like I could go figure it out on my own. I don't want to spend the time because like from a development perspective, I have more important things that I can go do. I I do a lot of the backend work in Brand Ghost.

So I could either be doing backend stuff or I could be trying to do like sales and marketing in the evening and stuff which I don't like doing. I feel like I'm not skilled at doing but like it has to get done. So, it's not really worth my time comparatively trying to sit down and really understand how some of these things are working in the front end, but I'm like, I know there's a quality of life improvement here, and I I don't think it's complicated. I just I just don't know it. Probably a beginner in uh in React like could go figure it out pretty quick. So, anyway, off to to GitHub. it's done like UI level searching, filtering, sorting, like basically working perfectly the first time. Um, the most recent example though, this one's super cool. Um, I I kind of forgot about it because I was like, man, this is going to be pretty complicated, I bet.

And I need one of the like my my business partners, one of my business partners, I need him to go basically review it. So, I'm like, I'm about to go create work for this guy cuz he he needs to go understand it cuz I I definitely won't. And I asked uh co-pilot, we we basically we can upload videos, right? Because it's a social media uh content uh crossosting and scheduling platform that we have. And right now when you want to go upload a video, it's because you're you're making a post. And I said, but my flow as a content creator is like I go export a bunch of short form videos at once. Like I might go sit there and do 10 to 20. And I'm not ready to write the posts yet. I just I'm editing them out or like I use Opus Clip for it.

So, um, there should in all my videos, there should be a an affiliate link to Opus Clip in the bottom. I don't even know if it works anymore, but click it and do that and give me affiliate revenue. But I I export these videos from Opus Clip and then uh I'm my mind is not in the space for writing the post, right? I'll do that in bulk another time. I just want to edit the videos, get them out, but then I have to sit there and upload them one by one into Brand Ghost. And I'm like, like it shouldn't be like this because as a I am a content creator and I'm having this pain point with my own tool. No, but I'm like the other guys are busy like and arguably they have more important things to do. So I said, "Okay, co-pilot, this one's on you.

I need a multi- video upload like option in this workflow. I needed to show progress. I needed to like upload the videos one by one showing the progress. I needed to basically create these placeholder posts that are disabled. So basically, you can go import this stuff and it's not going to go post right away. And I was like, there's if it can even do this, good luck. Like I need someone, one of my co-founders to go to read through it cuz like I'm not going to have any idea and I think it was yesterday or the day before um must have been two days ago. I was uh talking with one of my co-founders and I was like, "By the way, I'm like I haven't got you to look at this thing yet, but like I'm going to start looking. I'm going to try it out because I haven't yet." But I pulled the the branch down and Oh, I got to switch lanes here.

Oh. Oh, man. This is going to be brutal. We did it. Um I'm like, I got to try it out, you know, no idea if it's going to go. And I I just picked two videos to upload and this thing worked so flawlessly. How does this road work here? Oh, this is the right way. Sorry, there's a big roundabout, but there's got there's an extra lane around a boot because I'm Canadian. And uh so this multi- video upload thing just worked perfectly. It showed me progress as I was uploading multiple videos. It made the posts that like are disabled by default with all the correct values. Um just like per like it was flawless. I didn't touch a line of code. It everything just ran perfectly the first time. And um that's a ridiculous win. I still need my co-founder to go review the code because I don't trust myself to make sure that it's like done well.

It worked. So it's got to be done well enough, but I don't know from like the patterns and practices we use if it's uh up to par. So I wanted to share that example because these things are like in my opinion not examples where we replace software developers. I still need someone to go double check that that's all done properly, make adjustments as necessary. Um, and this is like an additive thing. I didn't go, hm, like seems like my my co-pilot can go write this code. Hey, co-founder, don't need you anymore, right? Just me and co-pilot now. I'm not only are we, you know, cutting jobs for software engineers, we're cutting jobs for, you know, founders and CEOs. Get them out of here. We got co-pilot. Like, that's not the case. It's like there's just so much other stuff to do that's a high priority and now we can get additional wins on top of that.

Like in the background, man, this thing was done in the background and my involvement for this particular feature was try it and it just worked. So super awesome. But um I just thought that was a really cool experience. So, my my current flow, let's talk about that, is I'm still offloading tons of like I guess like two two major categories and one that I want to get more automated, I think, somehow. The two major categories I have are um like almost like prototype work where I'm like, "Hey, it would be really interesting if there was a way to do something. I don't have time to start the first prototype, but I don't want to lose track of this idea. I already have the idea. Co-pilot can literally do a first attempt of the whole thing, and it might be absolute trash. might be the worst code that's ever been written.

But the point is, oh, I don't want to turn there. I want to turn after there. Um, the point is that I can have co-pilot try it. I don't have to wait until I'm freed up and potentially that idea never sees a light of day. Instead, I can have a first pass at some code and the worst case is I go, "That sucks. Delete the branch." That's the worst case. Okay. It's there's there's nothing to lose. So why not? The best case is that it works perfectly, but that's going to be pretty rare. So, the most likely case is probably that it writes a bunch of code that doesn't do exactly what I want, but then I can get some inspiration and see some patterns and go, hey, like those are some things that I think that I really want to do. Maybe I didn't think about that, or maybe there's examples where I'm like, okay, now that I see that, like definitely not going to build it that way.

Like, that's far too messy. Don't want to do that. So that's one type of uh thing that I'm giving to co-pilot just on the side. Um, and then I would say the other is kind of like I already described, which is these uh these smaller things, right? Like go do this UI thing, these quality of life things. Um, for the most part, right? It's like it's it feels like a distraction to get someone on on the team to do it, but if co-pilot can do it and it's small and the likelihood of it being good or good enough is high, just fire it off and then when you get a chance, pull the branch, try it, and maybe it's perfect. If not, we don't even have to look at it for whenever. Like, we can just wait, right? It's it just feels like there's this really loweffort way to get stuff done, which is super cool.

But the third thing is that, and this is the thing I'd like to find a better way to automate somehow or automate more of is like I want to get more like maintenance done on the codebase. This is something that I'm very interested in. I've said this before, but I feel like uh and maybe co-pilot with pull request isn't the right format for it, but I would love if there was a way that I could have agents running that identify drift in architecture, right? I I'm just going to give I'm going to make up an example. Perhaps I want my entire codebase to follow better uh vertical slicing. I want to make sure that I'm using uh more event driven architecture patterns. I want to make sure that there's consistency in like how I'm caching things, how I'm doing authenticate like all these things. And it would be really cool if I could have agents identify those opportunities.

So if I could have agents that are in the background basically like migrating the code to follow that in a more automated way because and I don't I don't actually know what that looks like in practice. I'm just kind of speaking about this like a a concept that I'd love to see and right now I don't have that in an automated way. What I can do is I can ask co-pilot like I can make an issue and say like basically make me a readme file or just a a markdown file that has your analysis the codebase. I've mentioned this on a video. I've already done that and I have it committed into the codebase as like a living document and um then I can take pieces of that and make new issues and say go address this go address this architectural drift. So that's kind of how it's getting done right now.

But given that it's like refactoring and like some architectural changes like that can be really disruptive. Like that's the the drawback is like sure let's let's pretend it could do it perfectly. Okay, let's pretend it's like doing these sweeping architectural changes and refactoring and I'm also trying to build in the codebase at the same time. Like that might be a bit chaotic even if it can do it perfectly. So, I don't know the right balance there, but I'd love to see something like that because I unfortunately like I am a strong believer that basically most code bases just like deteriorate significantly over time. Uh, I'm not the only person with this belief. There is a very popular YouTuber named The Primagen. You probably have heard of him. He has a great video on this and I brought it up before and um I strongly believe in it.

As much as you try to keep your code base from going to it's going to go to at some point. So, it would be really cool if there was a way to have more automation around agents to basically keep it from going to Now, with that said, maybe having agents in their current state trying to improve things is only going to make it worse perhaps, right? Maybe they try. Maybe it's not actually better. But I like the concept. I just don't know if we're there yet. And I certainly don't have my setup that way. But that's what I would like to have. I'd like to have that just running in the background. Um, I want to say non-stop, but I don't know what the right interval is. And just to make up, right? Maybe maybe on the weekends. It like refactors your code base and cleans up patterns that are in use.

And then you start work on your project. This only, this doesn't work for me because I do a lot more project work on weekends, but then you start work on it after it's, you know, cleaned things up. You have a fresh, happy code base. I don't know. I think the idea is cool. Try and help your code base from rotting but my in practice that's not happening. So the other two things I mentioned that is in place and um what I'm doing more of is I have this balance between like light use of AI agents in the IDE and chat mostly and autocomplete obviously. So I work in Visual Studio for like 99.9% of what I do. I have C-pilot obviously with the autocomplete and stuff. So I write my code that way, but I'm generally consulting either in chat in the IDE or usually with chat GBT to be honest.

I find something about using chat GBT. I get really good depends I guess what I'm asking. Sometimes I do ask stuff about the codebase and I can't be bothered to like go try and copy a lot of uh context out. So then I would use co-pilot in Visual Studio to ask it. But other times if it's just like general like, hey, I'm thinking about building a system that looks like X, I I ask chat GBT and I get pretty good results. So, I use that to to go back and forth on designing systems like, hey, I'm thinking about using these patterns, this architecture, this tech. What else do you recommend? Why do you recommend that? What are the pros and cons? Right? Like just getting it to break things down. Now, even if you don't if you think like um well, it's not accurate. It's like it's hallucinating, at least it's getting me to think through things and I can go follow up and get more data as necessary.

So, my my point is not to say I blindly trust the LLM. It's just that it's getting me thinking because otherwise, you know, I'm not on the phone with my my my partners all the time. and uh when I am, maybe that isn't the best use of our time together. So, you're good, buddy. So, having Yeah. having chat GBT or co-pilot to have those conversations with is like super helpful for me to to brainstorm. So, that's a huge part. And then I'm actually doing the coding. Sometimes I'll get co-pilot or chatbt to give me a you know class or block of code or like um you know I need help. I want to think through this algorithm like you make it. I'm going to go like refine it and tune it. Uh or the other way, right? I'll write something naive and I'm like I think this is naive.

How do we optimize this? But it's a back and forth kind of conversation like pair programming experience. Then I'm finding from an agent perspective, there's a couple of things I'm doing. So sometimes I have like a pattern where I'm like say I'm writing some tests and I'm like okay I've established the pattern like the setup for this test and how I want it to look but I'm like I know I probably have like five other test scenarios. They're going to follow a really similar setup and stuff like that. So I will tell it tell the agent, hey follow this pattern and go test these other scenarios. And then while it does that, and it's usually pretty good, um, when I give it really really narrow context, I I context switch to something else. Um, so while it's doing that, then I go check on the other pull request and stuff that it did and see if I can like start reviewing things.

Uh, maybe there's some really obvious stuff I catch. Maybe it's in the front end. So while it's writing a couple of extra tests in my IDE in agent mode, I switch over to VS Code, run my front end with a, you know, a feature branch that it made and I'm testing that out. So I get to be like the like tester or QA or product owner like evaluating the feature in the front end and like I'm just getting two things done at once, which is super cool. There's a lot of overhead to context switching though because I've had days where I go through like that. I'm like, I feel like I worked on like 20 things, but I'm like, I don't know if I finished more than like three of them. So, that's a thing I got to be careful about is like it kind of feels like in some cases there's too much in parallel because of how you can context switch.

Um, but like I'm I'm really like interested in exploring this flow more because I feel like right now it's not I'm not the most efficient at it because all these things are new, right? It's a very different way than I've ever programmed before. Let's rock and roll. Sorry, there's a light to get on the highway and we got to get up the highway speeds, right? So, what better way to do that than to get up the highway speeds? Um, how do we get to the fast lane? So, yeah, I feel like it's a a powerful flow. Like I feel like it has all the pieces I want because when I've talked about being disappointed by agent mode in the IDE, it's um I'm working around that now because I have co-pilot doing my pull requests. Like that is working really really well as an agent. And then that just leaves me with like really simple stuff for agent mode to do.

And like that's not so bad cuz I can go multitask, right? It's like it's only a sliver of having Visual Studio like Copilot doing some stuff in the background versus me just getting frustrated that it sucks at everything I'm asking it to do. So, uh, it feels like a good balance. Right now, I'm finding that the big drawback with agent mode in Visual Studio, even when I give it some of those constrained things, it's really just the user experience is like pretty sometimes. So, I'm not sure if you've tried it in Visual Studio. I think it's very similar in VS Code, but it will run the agent changes and then it shows you a diff. And I've had some times where like I'm trying to scroll through the diff and like the scroll bar is stuck and it like uh like flips up and down really fast and I'm like I can't I can't even scroll through the changes.

Like I cannot review the diff to say if it's good. So I just have to take it because otherwise nothing got done. So I have to take it and then like use get extensions to go through it visually which is that's dumb. or um it like gets stuck. So I'm like sure take the changes and then I'm like it's not take like the UI is not changing. So I'm like I don't know if it took or not. Um there's just a lot of like dumb stuff like that because I feel like the tooling is pretty rough still. So I'm not writing it off. It's just currently it's a little bit frustrating. I know it's going to get better. I'm not not concerned about that. It's just like in the moment I'm like, man, this is kind of shitty. Um, so still a lot to go there.

But overall, I've been finding this uh this kind of pattern of working is like it's exciting. I think like I'm trying to I want to zoom out a little bit, right? like as a software engineering manager, I'm not doing any of this at work. I think that the future of my role if we have tooling like this um I could imagine using like we use Azure DevOps at work so I haven't seen the the equivalent yet but I could imagine myself if there's like smaller um quality of life smaller bugs and stuff instead of me being like oh to the bottom of the backlog no one will ever look at that uh or I have to give it like okay who's like a junior person on the team I can give this smaller thing to and like instead of being in a situation like that

I can just be like you know even if I have juniors on the team like if people are busy they're busy let them keep working on what they're working on I could try feeding this off to co-pilot and seeing if it can whip something up. Um, the the other reason that's handy is like it's extremely trivial for me at home to clone my GitHub repo to make changes in Visual Studio and push them up. It is extremely not trivial for me as an engineering manager who doesn't have any of the tools set up regularly, right? I haven't cloned code in ages. We have many repositories. I'm not well equipped at work on a regular basis to be able to ship code. I can review code like crazy. So that might enable me to go tackle some of the development flow and free up the engineers. So I'm excited about something like that.

Um, I think like if I weren't at Microsoft and I was 100% in brand ghost, I if I could refine this pattern, I feel like I would be an absolute machine getting through stuff. Um, because the it feels like I have all of the right pieces to keep myself consistently like working on important That's how it feels. And to me, that's like super exciting. Like, I've made videos before on this channel talking about like I have a a thing where I'm always trying to see like how can I be more productive, probably to a fault. Um, you know, my my wife will joke when there's if there's a scenario that comes up and it's like there's clearly an inefficiency, she'll look at me and she's like, "That probably drives you nuts, doesn't it?" Like, she knows that I'm always looking at things being like, "How would I change that to try and make it better?" Always.

Um, but I do this for my own productivity, right? when it comes to to making YouTube videos. The fact that I started doing code commute in the first place because I wanted to fill my commute time with creating content is a good example of me trying to like optimize my time for stuff. I'm not saying that I'm good at it, by the way. Not saying that. I'm just saying that it's something I keep focusing on to try and do. Um, I don't know. Maybe a better use of my time instead of code commute is to go learn foreign languages while I'm driving or something. I don't know. But so I don't know if it's the right thing to do. I just feel like I'm always trying to to be productive. So having this tooling setup feels like really good. I just feel that like the tools have some room to go.

Obviously, um I keep saying it, but like GitHub co-pilot with the pull request has been like, "Oh, buddy, come on. You got to get out of this lane." Stopped. Like, I'm not exaggerating when I say this. This person stopped five car lengths back and tried to to switch lanes, but they weren't getting out of the lane, so I couldn't keep going. Anyway, the GitHub co-pilot with poll request setup is like that is an absolute game changer for me. I can't say enough positive things about it. There's a couple couple usability things like I think we saw that um we can't yet and maybe we're using it wrong but we couldn't get it to like check out a branch. We couldn't say, "Hey, like here's an issue. Co-pilot start on this branch that we pushed." I think there's ways around that. Um, but like I that didn't feel like a first class thing that was there.

So, that's one thing. Um, there's this weird situation where like if you're rate limited because you you've assigned it too many issues, copilot like it's running like a session is what they call it. And I'm just making up numbers. Say you're running five sessions in parallel. It will rate limit you and fail, but it will fail all of the sessions. So to give you, just to exaggerate an example, say overnight you're like, "Okay, I made 20 issues. I'm going to go sleep for eight hours. Over the next eight hours, I want GitHub Copilot to go crush these 20 issues and see what it comes up with." So I wake up in the morning and I'm just reviewing code, right? What will happen is that you assign the 20 issues, it starts working on them, and then all of them hit a rate limit and all of them fail.

That's it. No progress. You'll wake up in the morning with zero done. And to me, that's like a pretty feels like a pretty easy one that they can fix. like you hit the rate limit, like pause the session that you're in or like only allow so many sessions in parallel. A combination of those two things I feel like probably addresses this, but I'm sure we'll see improvements like that cuz that's where I'd like to get to is have like just a ton of stuff that I can fire off overnight. But it's a really enjoyable experience. So, I'm going to spend this weekend. I got to make some YouTube videos, but um we're going to try to crank through some more Brand Go stuff. We uh we're just working on some quality of life stuff in the front end over the past few days. And then I want to get I need to get to a like a media processing like job framework.

I'm not excited about this. Makes me a little nervous. And it's because um like pictures are pretty simple. I think it's videos. I have to yawn really bad. Sorry, I warned you. Um pictures seem pretty simple because I already do picture processing and brand ghost right now. And um the thing I'm nervous about is like videos are big. Videos are big. I have to use ffmpeg to like do video stuff and like I don't know, man. It makes me nervous. So, we'll see. But what I want to be able to do is that when like this might not be obvious to some people depending on how much you use social media, but um there are restrictions to the types of content that you can post on different platforms like the formats. And for video, one of the big things is the um is the file size.

So, for example, if you wanted to post a blue sky, that's a 75 megabyte limit on the the upload. Other sites have different limits. Can't remember Instagrams. It's pretty low. And I think they have there's also like a time limit, too. I forgot that part. Sorry. So, time limit, like a duration and um file size. So, I'm not going to trim the duration for people. Like, if you if your videos too long for a platform, I'm not going to guess that you're okay with just the first minute. I don't want to do that. But if you had a 100 megabyte video and the max size is 85 megabytes, I suspect within some threshold that you would probably be okay if I could compress it a little bit, lower the resolution, do something to get it posted for you. Probably. I say that because as a content creator, I have I don't know like 400 shorts that are clipped and some of them are too long to go to uh to Instagram.

And I say too long, so I mean the file size is too big. And I'm like, man, like or sorry, to to Blue Sky, the file size is too big. Instagram, the duration's too long. But for Blue Sky, I'm like, I don't care. Just compress it. Like just post it. Uh because then it's no content. So I'd like to do that a little bit more automatic, but it just makes me nervous. So I might get to that over this weekend, but we might have too many other things we're trying to to crank out as well. So we'll see. But I got to wrap it up there. Hope that was interesting. Um, I realize it's kind of me rambling just a lot about different AI setup. This is going, man, come on. Get it together. But yeah, I will uh I'll keep you folks updated as I'm kind of changing or evolving my AI tool usage.

I I saw that Cursor has some Slack integration thing. It sounds like GitHub Copilot but from Slack. But like I don't understand where it's running. I realize that might sound stupid. I don't know enough about cursor. Like is it going to kick it off to my IDE or is it going to run an agent somewhere in the code like GitHub copilot? If so, maybe I'll start using Slack and and Cursor, what however that's set up and then I have more throughput for agents working in my codebase. Um, I I think I paid for cursor and I I haven't touched it. I was like, agent mode is I hate it. It's completely useless to me. I have to handhold it so much. So, paid for cursor, whatever. I understand too. It's going to get better. I don't question that. Can my prompting get better? Yes, it can.

But right now, if I have people are like, "Oh, it's just it's just your prompts are bad." I'm like, "Dude, if I have to spend an hour crafting a perfect prompt for this thing, I might I should have coded it myself. It's not a productivity boost if I'm spending more time trying to perfect a prompt than coding it. So, no. But you can't win against the AI bros because they're AI bros." But anyway, I'll wrap it up there. Thanks for watching. I will see you next time. Take care.