Rubber Duck Thursdays \\ Let's cowork, code, and learn together!

Na d You're Happy N. Howdy y'all. How is it going? Happy rubber duck Thursday. We're going to let the music's fading out. Perfect timing. Hello everybody. Welcome to Rubber Duck Thursdays. How's it going? Um, happy April. Although it was April last week, too, so I don't know if that counts. I have been off speaking at conferences away from my desk and so this is this is my first rubber duck Thursday back. We're here. We're making it happen. Hey Serly Dev, good seeing you again. Hey Denil. Hey Nean. Hey Brett. There's a bunch of people in the chat. I'm looking at chat from YouTube, from Twitch, and from LinkedIn. And so there's a lot of you. It's good to see all of you. It has been a very busy time, but very fun playing with new things that are here on GitHub. I'm actually going to pull up the change log really quick because stuff have has gone out even today that I haven't actually seen that I'm particularly excited about and I would love to show you. So, one second while I pull up my screen. I have not shared my screen yet on this platform. So, give me a second. It's pulling back. I am back. Hello. Yes, I was I was very specifically um speaking in South Korea this past week which was very exciting but I got back last night so jet lag is real. But it was very very fun being able to talk about um Copilot CLI and SDK talk about all the new things coming to the platform and once again new things. Um, I can't wait to talk to you about what is new in the change log because once again, this one in particular, I think will be really great for open source maintainers. New lowquality option in the hide comment menu. Um, you can now say like, hey, this comment is spammy or this comment is off topic. This comment is um a duplicate or resolved or something. You can now hide a comment with a reason, which is particularly nice. I think that will be very nice. once again with all of the uh AI agents out in the world um posting a lot, creating a lot, it'll it'll be nice to have that. So anyway, this this is something that is new truly as of like an hour ago. It's it's very new, very exciting, and uh this will be a great one. Let's see, there are some other ones. Oh, new sort by control adding notifications. Notifications on GitHub. It's a lot. It's a blessing and a curse. We love it. But now you can sort them. This was a very very big maintainer ask to be able to just sort like newest and oldest notifications that you get um for pull requests, for issues, for all the different things and being able to group them. That that's already been in there. But being able to sort oldest to newest, that one is also new today and very exciting. And then also even more, you can now see roles that um different people have in your poll requests. And so for example, if someone is a regular contributor to your repo or if someone is just a member of your repository or if someone's a bot like dependabot or something. Now you can see those labels. This screenshot is very tiny. Um, and also I'm looking at everything in the change log right here. And so I'm I'm going to post it this is the comment. It's github.blogchange log, but these things are very very new. And um yeah, anyway, being able to see these labels will be very helpful for maintainers who are doing a quick glance at like, okay, where are these poll requests coming from? All all that jazz. The these top three in particular I'm very excited about. But guess what? There's more. literally just today. Um this one in particular, this is the last one of the day. Um the latest release information in the issue sidebar and and default values. So um here there's a good screenshot in here. So on the side here, this screenshot is very very tiny, but that is just the nature of the beast when I am sharing my screen. So let me drop that link in the chat as well. Chat does not work on LinkedIn. And so if you want to see these, go straight to github.blog/changeloglog. Um, or you could jump over to GitHub's YouTube or Twitch to be able to click on these links. But anyway, you can now see the release information. Oh, that's so tiny. Let me see if I can put that image in a new tab. It will not let me. Let me try doing that one more time. Maybe. No. Okay, that's because I am logged in on work things and so it's not letting me actually show it. But anyway, this this image right here, you can see this release tag based on the actual commit in there. You see the you see the release info. That is what's in there. And it's it's just really nice to be able to have like you could say if uh something is a part of pre-release or the latest release. And you can also set these types of labels as default. Um, so anyway, there I think there's more in there, but I didn't read this part. Anyway, these are all the things that are new as of literally today, which is very, very exciting. And there's so much more. You can check it out once again in github.blog/changelog. There's all kinds of really interesting things that are coming out literally all the time. Um, from APIs to things specifically on mobile around security, different keys. I like seeing the Visual Studio Code releases in particular. This one I I there's some cool things that are now automatically built into VS Code. Um so different agent permissions, autopilot is built in. Um but I think now GitHub Copilot is automatically built in to VS Code. I know that it was in uh VS Code Insiders which anybody can use. Um I don't remember. Oh, I like this copy final response though. This part in particular was nice. um in VS Code. So anyway, lots of cool things in the change log. Check it out at github.blog/changeelog if you don't mind me putting on my shilling hat to say so. Um lots of very cool releases in April this month. It's very fun. There's a lot of people in the chat today. Hey everybody. Good to see all of you. I see Tamed Craziness is here today. Hey. Hi Jacob. Hi Hari. How's it going? Um, what are you all working on? Let me know. I have been personally working on a few different things. First of all, um building some like demo apps for work in particular where uh figuring out okay what what's some realistic workflows that developers can use and and that we can where we can show some new capabilities in the CLI and SDK specifically but also multimodel stuff where for example if I wanted to use plan mode in the CLI to create a plan for a project say okay make this plan with claude height haiku, let's say, but then use the new rubber duck feature, um, which I think we've posted about and I can talk a little bit more about to ask GPT5 for, hey, is this plan good? Can you verify it? Do you have any feelings about yada yada yada? And then from there, implement the plan with a different model where you can actually use all of these different models back to back to back in the same session and it just works. And it's really nice, especially if you like clawed models for certain things, but like open a models, open AAI models for other things or Gemini models for certain things, being able to really just pick all of the different models and and choose how you want to work with things. I've really liked those workflows and it might be helpful for you all as well. I um I have found that I still I still kind of default to Claude sonnet models, but for creating plans, I really like using Claude Haiku because it's fast. It's really nice. For backend stuff, the new GPT models have been very good. For front end, not so much. That's just my personal experience. It's been interesting to kind of bounce between all of the different ones. And once again, keeping the shill hat on for work. It's been nice to have the option to jump between them just on the one platform. Let's see. There's people working on GPS time tracking and payroll applications. That's cool. There's some people working on some personal websites. Lots of cool things. You can work on anything you want. If you recall, if you were here, I think it was two weeks ago on this stream, we actually built something that I thought was particularly fun. Let me see. I think I think it's actually on my GitHub profile. Let me pull it up. One second while I shift through this screen and find it. Um, this was a theme that we actually made for VS Code recently, literally literally two weeks ago on this stream where my co-worker um, and I we we wanted to make some kind of Taco Bell theme without saying Taco Bell, but it's Taco Bell where we pulled up the menu for Taco Bell. We called it TB Purple instead of Taco Bell Purple. Baja teal instead of Baja blast locos orange. You know things where if you read between the lines, you know what it is. But um we ended up making this Taco Bell theme where if you do VS Code themes and then taco, let me pull it up. Hey, it is in the marketplace and it works. And we made this fun little icon. It's been pretty fun using that where if I were to pull up, let me pull up a repo really quick on another window. Um, I'm just going to pull up the repo I have for my own blog, which my blog, my personal blog is open source. This is what it looks like so far, which honestly out of the box is not that bad. Restart Visual Studio Code. I'll do that later. Out of the box, it's not too bad. It kind of looks like Visual Studio with the purple, but it's just like the brand colors, but you can see that like Baja Blast teal right there. The yellow is actually from uh doing a color picker on a taco. And this is this is real people. Um thanks for thanks for laughing Brett. But anyway, that has been kind of a fun thing. If you have thoughts, I would love to hear them. Or honestly, it would be very helpful if you just made a poll request if you're open to it. Um, once again, the repo is open source. And actually, let me put this marketplace theme in the about section right here. This is this is going to be this is V1, but I'm going to drop this link in the chat. Bam. If you would like to contribute to the Taco Bell vs Code theme, sorry, just the taco theme, but we're thinking outside the bun here. Um, I would love your thoughts. We uh instead of doing live moss, we did Viva more but then code Moss. So once again, it's you know what this is about. You have a bagel shop. Wait, do you have a bagel shop in real life? That would be incredible. I would love to know more about that in general. I think that's fascinating. I love the custom solutions era. Every client can finally get a complete custom software package built specifically for their business. I do think personalized solution is the wave of the future. I do think that's very true. I feel like we are definitely in a personal software era where everything you build can be very custom to you. Now, there's pros and cons of that. I'm going to I'm going to pull my screen off for a second. There's there's pros and cons to having everything custom because when you build a custom it's great for you but that also means you have to maintain it which is okay. I do think that some things sure maintain it. It's not the end of the world. You can just do that. Other things it could be a pain where uh if you are building something for example that processes payments, do you want to maintain something that processes payments or do you want to use like a stripe or a PayPal or something you know? Um do you want to maintain something that handles information of users? That could be iffy depending on how you set up authentication and your database and everything. You have to balance this kind of stuff. That being said, once again, it's really nice to have custom solutions for all the things. So that's that's that's my very very high level thought on that. Do you want to maintain that? No, you don't. You really don't. Exactly. Exactly. I I've definitely been able to say like, okay, I don't necessarily need a subscription to that app right there. There there are some things that I have been able to fully just build, which is nice. There's some other things. No, one second. I'm going to drink some water. I've been doing so much talking. I I mentioned that I spoke at a conference this past week. What I didn't what I didn't mention is said conference had a lot of breakout sessions. So I actually gave six talks this past week and my voice is feeling it. Um but anyway, we're we're hanging in there. We we're are building building the thing. That being said, what do you all want to build? Do you want to do payments? I leave for PayPal and Stripe. I know, right? Um, it's it's nice to have something else do it for you. Would you like to build something with the CLI today? I was thinking about making something. I I haven't dabbled with the Copilot SDK in a few weeks and I know like new features have been implemented. If you haven't seen it, let me actually pull this up. Um, GitHub Copilot SDK. Don't worry, I will pull it up on this monitor and share my screen for a second. I see a Yeah, let's let's do it. Let's let's build something. You want to build something with the CLI? Cool. Let's let's build something with the CLI. Um, so the GitHub Copilot SDK is built on top of the CLI, and I'm going to go into that a bit more. Um, so with the GitHub Copilot CLI, you can run agent sessions to your heart's content. You can say, "Hey, I want to develop a plan to build XYZ." A water tracking app. Let's just say I'm drinking water right now. I can open up open up the app and say, "Hey, I drank water." And it could track that. You could build some kind of visualizer of data. You could build any feature into any existing app you have. It just you you tell it what you want to build. You can build that. You can create plans for it. You can also say I want to watch the session as it goes closely and you can watch that session from the CLI or from github.com or from GitHub mobile various client applications or you could say I'm delegating it to the cloud go I'll just look at the resulting PR you could say I want to implement all of these different features but I want to implement them in parallel so I can have just multiple sub agents going anyway you can do all of these things but also with the power of the CLI is the SDK and that's what I'm showing on my screen right here. Now, this allows you to build apps or a you a user interface on top of the power of the CLI. And so, you don't have to necessarily build like an agent session thing like where where it just run a runs agents to build software. You can use it to add any AI features to your apps. And so an example is one of my co-workers built a browser extension that lets Copilot control what's happening where for example they opened up a shopping site and they said hey go through the shopping site find a large dog bed add it to my cart and then navigate to the cart and then ping me. It made that chat window. The browser just started doing its thing because the the agent was running in the browser and then they got their notification. They were able to go to their shopping cart and it just worked. Similarly, one of my co-workers, she built a tool where as she was working in the CLI building features, she's able to say, "Hey, based on the features that I just made and this template, create a PowerPoint presentation about all of the features that I just implemented in this session." And then from there, I also want you to schedule a meeting for me with this PowerPoint based on this session. So that way, I can talk about it with our stakeholders. and it all just worked, which is really, really nice. And so anyway, you can use the SDK and the various other MCP servers and tools to build apps to your heart's content. That is my like sales pitch spiel on that subject. Um, how to get GitHub Copilot freely. Just so you know, GitHub Copilot Let me just pull this up really quick. You can get GitHub Copilot for free. So in your on your account, if you go to your account settings, you can just enable GitHub Copilot. I say just because there's a button that you click to enable GitHub Copilot and it will just work everywhere you use it. And there's chat in Visual Studio Code as well as as well as a few other um IDEs. You could use it in Jet Brains and Xcode and Eclipse and a bunch of other places. Um but outside of the IDE, once again, you can use the CLI, you can use the SDK, you can use it in a bunch of different workflows, but you get it for free automatically. Bring back cloud models to Copilot. They are already in Copilot, so don't worry about that. they already exist. There is a free plan. It's not a free trial. It's just a free plan. You can do a lot on the free plan. I have a friend where she's been learning how to code and has been building a bunch of stuff. She built like three or four pretty large projects just on the free plan. And so, you can do a lot, but there's also a pro and a pro plus plan that you can use in there. Also, GitHub student pack. If you have aedu email, you should be using the GitHub student developer pack. It's at education.github.com/pack. And I'm going to put that in here in the chat. education.github.com/pack. This gets you so much free stuff. So, first of all, there's a student plan for co-pilot, which gets you access to more models. There's also a ton of educational resources. There's courses the so many courses I really like the front-end masters course. There's you can get certifications for GitHub copilot as well as a bunch of Microsoft things. So many courses. All of these are just courses. There's also uh interview prep tools that you can use. This virtual event kit is great because it allows you to also get free domain names and like an Azure plan. So there's all of these different courses, but then under the offerings once again, free domain names, free hosting, free notion, project planning tools. I'm going to say courses again because I think it's very, very important. Database credits, terminal tools, so many different things. Observability tools, you can use that for free on the GitHub student pack. Oh yeah, one password is a really good one. You get free one password. So anyway, all of these things you just get for free. So to answer your question, sigh there's a GitHub student pack and there's also the GitHub co-pilot free plan. You can use all these things for free and that does include cloud models that includes GPT models that includes all the models and depending on your account depending on the things that you have enabled you will have different models where my models are even different from some of my co-workers just because depending on plans that you're on tiers that you're on whatever your company might have enabled for you that is what you have access to and it's pretty cool. Let's see. Pro Plus isn't enough for me anymore. I'm needing to add tokens halfway through the month. Could you make a new tier with more tokens? I'm being very careful with how I say this. You will be very happy with some new things that are coming soon. And I'm leaving it at that. I hope you can read between the lines. Okay, cool. So, let's see. Copilot SDK. Can I integrate it with I don't know what Y III2 is, but with a Laravel project? Yes, you can. Can I make a C-pilot customer service agent chat just for my company with specific answers? Yes and no. That's this is an interesting question. So first of all, if you want to integrate it with your Laravel project, there are specific languages that are supported officially. The official languages that are supported are the Node.js, TypeScript one, Python, Go.NET, and Java. Yes. Also, really quick, because there's some questions about what I was just talking about. Education is only for students. Yes. Otherwise, you there's the free tier and then there's the pro and pro plus plans. There's also business and enterprise plans. How soon? I cannot tell you. Anyway, back to this question. These are the official language supports. There are other languages that are unofficially supported. Um, I don't know if there's actually a link in here to those unofficial ones, but I think there actually is a PHP unofficial SDK. That being said, if I were you, I would use one of the official ones and integrate some TypeScript into your PHP apps. Personally, that's what that's what I would do. Um, just because I like official things and that. Um, now to make a co-pilot customer service agent chat dedicated to company specific answers. What I would do for that if I were you thinking about it, I would create the chat app create create the actual app itself that's integrated in and you would need need to put in your companies or your organizations your personal token to be the one used that that Copilot uses to generate responses where you would need to put in your documentation. You would need to put in like all of the things that how you want it to answer based on how people are responding to you and then it would use co-pilot with your tokens to respond. Does that make sense? You can do that. You can that is how I would do it. I hope that makes sense. Let me know if you uh let me know if you need a little bit more clarification on that front, but that that should answer your question, I think. Anyway, I also I have built some things with the SDK. Typically, I just do like agent runners with different UIs and stuff, but one something that I built on this stream for example, I think like a month or two ago, probably a couple months ago now, was a color picking app where like it's not only picking colors, but it's storing those colors wherever you want. And then I added it so you could authenticate with GitHub and then with the SDK, it could like automatically name color palettes. And that that's a very silly one, but it works. You can do that with again any agentic capabilities that you want in your apps. You can do that with the SDK, which is particularly neat. I'm currently coming up blank on app ideas. I'm I'm trying to come up with one in my head right now, but I've been doing so much talking that I haven't actually done the brainstorming part. If you have an app idea that you'd like to build, let me know. Otherwise, I will keep talking until I come up with something. Um, there's also, actually, because I'm going to just keep talking, there's a repository called Awesome Copilot. And Awesome Copilot, I'm going to put in here. It's literally like the GitHub or github.com/githubesomecopilot. I'm going to drink some water. One second. in here. It has just a bunch of community contributed instructions things. Let's see. Is my stream still working? My stream my streaming platform just started crashing on me. One second. Let me I'm going to watch my own stream for a second to make sure that it's showing up. Okay, cool. It's showing up. I'm just going to ignore that then and hope that it works and continue on. You can still hear me. Great. You see me? Amazing. Awesome co-pilot. This It's nice. Um, and so in here, uh, we have custom instructions, custom skills, custom agents, all these different things that a lot of people contributed to. And if you want to do like a free poll request, contribute to some open source and and have that uh on your resume, there is a full collection on the website for awesome GitHub copilot. And actually, there's an extra space in here. Do you see that? I might fix that user interface because that annoys me. Anyway, um in here again you can see all the custom instructions, all the custom agents, custom hooks, custom plugins. People have made so many interesting cool ones where for example I really like the accessibility expert one. A custom agent is a glorified markdown file where in here there's front matter and that has like the name, the description, the default model that you want to use, the various tools that it wants to use and then instructions where you're a worldclass expert in web accessibility. Sure it is. you put in like okay these are the very specific things that it wants like the conformance to these particular um standards being able to say like okay non-ext comment it has like alternative text media and motion here's the approach that it takes it's really just super specific prompting and then you can add it to your co-pilot sessions and so you could say for example now that you have these custom agents and you can put those into your GitHub repositories. When you're running a session, you could say, "Hey, I want this custom agent to run the accessibility check. I want this other custom agent to do this." You can say that you want these specific custom agents to run various tasks for you and then it will conform to the styles that you want. This one is a very practical one. This um this very specific accessibility one that I showed, but it doesn't have to be just specific ones or or ju just um practical ones. One that one of my co-workers made at one point was just a talk like a pirate one where it was, "Hey, for all of your responses, sound like a pirate. Make sure you go y and and ohoy matey and stuff in all of your responses. It's very silly, but it works. You can use custom agents, you can use custom skills, you can use custom prompts. A lot of these are really just glorified markdown documents at different ends of the stack. And so I showed you the accessibility custom agent, this is accessibility custom instructions where this instead of it being like you are a custom agent that is a specialist in these tasks, this is more guidance and instructions overall to your system. that is running things where okay I want you to conform to these specific things once again it's all markdown under the hood it's just where is it being read in a particular session at a particular point I think awesome copilot in particular is a great place to start for all of your custom instructions needs uh and customization needs and once again uh this is the repository for it and so check it out there there's actually it's only 16 issues now and only 35 poll requests. There have been so much more in the past. Contribute where you see fit or just try using it and and give feedback because the more people use it, the better it gets because you can see again how things are being done. You can contribute your own improvements and once again it's a nice like free contribution there. Something that I've been working on in particular personally is um let me pull up my to-do app. So I have this task tracking app called to-do meter. This is an app that I have been working on. I'm not kidding 10 years maybe nine years but nine or 10 years close to a decade. It it has been a very long time. This app I hadn't really made changes to for about five or six years because it it just worked. It got it got what was done. I'm going to show this screenshot right here. It it's a glorified to-do list with a progress bar. I like this app because it's plain and simple. When I check something off, it fills the progress bar. When I pause things, it fills the progress bar. That's it. I've liked that this is all it does, but there were some feature requests that people have added. For example, being able to edit tasks. That's something that you historically have not been able to do in To-Do meter because I personally didn't want to implement it. Now with agents, I can. And that has been very exciting where um for example, I'm actually going to pull it up. pull up my actual to-do list. You know what? Let me make some of these things a little bit more generic. One Okay, cool. This is my actual to-do list, but this is the newest version that I'm working with. I need to I need to cut a release. You know what you can do here? You can click on the text here. Actually, I'm going to zoom in a little bit more so it's easier to see. Okay. Go through all of the later things on Slack. This is actually my like reminder section to just go through things. When you click on the text, you can edit it. So, I want my inbox. I want I want to specifically say email inbox less than 30 less than 30 emails or something, whatever. It's so simple, but you can edit things now. That was a task I've been putting off for so long, and now with a bunch of AI tools, I can just do it. And also updating dependencies is a chore task that I dislike so much. I honestly really like maintaining this app. Once again, it's an older app, but it just gets the job done. It's Electron based. Electron has gone through a lot of changes in the past 5 years, 10 years. It's it's it's gone through so many changes. I did not want to update dependencies. Now, I can just tell the agents to do it and it's done. And granted, I still go through all the code and stuff. I could still like pause this task or later I could say I want to resume this task right here. Let me resume that one. Um, but that's that's been the main thing that I've been working on where it just it gets the job done. And I've been really really happy with the latest changes in the CLI with the latest changes where I can say, "Okay, this is my plan to implement yada yada yada and then have it vet the plan, the latest plan that I've been working on." And actually, let me start over a session and show you what that looks like. One second. Let me pull my screen off while I do this. Um, thanks. I also think it's nice. Um, let me pull this up. Let me zoom in my terminal and then I will share my screen again. One thing that I've wanted to think through with my app to-do meter in general is making it so that way it also hosts some kind of MCP server. So when I run the copilot CLI, it's going to load. Gorgeous. There it is. So, what I can do in the CLI when I'm thinking through something, I could say like, I just want to think about it. I just want to go. Or I could say, I want to shift into plan mode. So, when you do shift tab, that goes into plan mode. Everything turns blue here or cyan. And then what you can say is uh I want to come up with a plan for XYZ feature. And then it creates a plan in markdown. Everything in AI is marked down. So what I could say is like this app uses blah. And as I as I explain what I want to do, I can basically write out I want to have a plan for these features. Let's walk through it and then it'll create again said plan for us. And so for example with to-do meter what I want to do is I want to say um wait actually it says I can run update. I don't know if you see that run update right here. Wow. I'm interrupting myself for a second. Just like two days ago, it was version 1.0.18 and now it's already 1.0.22. The team is cooking. Okay, so I'm going to run update really quick because that goes quickly and then we'll go through the plan part because I like getting all the latest and greatest. Okay, nice and fast. Shift tab into plan mode and then I could say what I want to implement. And so I could say this app uses Electron as the base and React as the front end. In the middleware is where the two ends of the app talk to each other. Are we getting any premium access to GitHub Copilot? Yes. Um, I act I actually already talked about it, but luckily I still have the tab open. Um, there's a free plan for GitHub Copilot right here. And so if you want to use GitHub Copilot for free, you can do a whole lot on the free plan honestly. Um, I do honestly most things with free plan level stuff because you first of all I want to be like this is what a real developer uses. um and not like just an employee. But um yeah, like right now I actually have Let me let me delete what I have here for a second. Um why isn't it letting me select all? I think that's my computer's problem. Hang on. Run model. Um you have different model lists than I do cuz everybody has a different model list. But Claude Haiku is what I typically use for plans for example. Um you can use various plans for various things and currently I am looking at my to-do meter app. So I could say like um my app uses electron to host uh very client heavy react shell. It uses local storage for all saving of data. I'd like to talk through a plan for two things. One, coming up with a file based way of storing and exporting data that's as lightweight as possible. and then opens up the possibility of syncing tasks across devices. And number two, being able to host an MCP server for tools to interact with and read to-do tasks. This is actually a very hefty plan concept, but once again, when you run plan mode, you can just talk through things and it'll ask you clarifying questions and not actually implement anything. I honestly use plan mode more than I actually write code with the CLI just because using it as a thinking tool is a very helpful exercise for me specifically for my own personal open source projects. Um, let me see. Uh, hey from Stockholm. Welcome, Jack. Let's see. In my repo, I use uh I use cloud code calling Gemini via vertex. Cloud code CLI uses Haiku for planning. Anyway, yeah, with the CLI once again, you can choose your models. I like to use cloud haiku for this, but then there's actually a new rubber duck feature as well. I'm I just keep interrupting myself, but this stuff is very cool. Let me go to github.blog. There's this one. It combines model families for a second opinion by default, which is really cool. And so uh and this is in experimental mode. So I say by default if you're using experimental mode, but um what it'll do is that it it says like our evaluations show that Claude Sonnet and the rubber duck makes up 74.7% of the performance gap between Sonnet and Opus alone, which is huge. Um and confident mistakes can compound blah blah blah. This blog post is very good and I'm going to paste it in the chat here really quick. Um, once again, chat does not work across to LinkedIn, but on YouTube and Twitch, you can click this link. But what it does is there's this rubber duck agent that it uses um GPT 5.4 under the hood. And so, as you're making plans, it'll consult GPT 5.4 for on the plan to point out holes and give it another perspective on your plan which is really really cool. So anyway that that is that is a very fun thing to use and I recommend um with GPT 5.4 I've stopped using claude altogether. That's interesting. I have found that 5.4 is very good for like architectural stuff. I don't like it for user interfaces personally just just way the way I've messed with it for front end it's not been the best but it's worked it's works really well it's a it's a nice it's a nice model um is GitHub getting that new anthropic LLM probably I'm not sure which one to which one you are referring but it has access to all the things which is particularly nice anyway so what I'm going to do is I have these two points that I've made in plan mode and then I'm using cloud haiku and I'm just going to hit enter there. Now what it's going to do is it's going to think through what I've said and it's going to actually analyze my codebase. So it's going to read through everything, look at my package json, see all the different parts of it. So I have like the uh I have the runner on the front end and on the back end where there's like the renderer process and then the main process. Um and then it uh it also has like a middleware that talks between them. I think it's called IPC that that where I could say like okay this number of tasks and stuff I want the main process to send notifications and everything. For filebased storage which format would you prefer? JSON is simple and human readable which is best for portability and syncing. SQLite is for more robust for complex queries which is good for an MCP server or a custom binary format for minimal size. this. I'm not sure. I' I've literally been going back and forth between JSON and SQLite. And so what I'm going to say is not sure which one would be best. I'd like for you to make recommendations on SQ SQLite versus JSON based on other similar local first applications like Obsidian for example. I'm going to do that and let it make some recommendations based on that. How do you envision syncing across devices? Do you have a preference for the approach like pushing files to a cloud service like Dropbox using a custom sync server or would you want flexibility to support multiple sync backends? This is a very good question and thing things that I want to think through. Um, also Melody, I see you in the chat. I'm curious, do you think that you're going to lean towards CLIs from now on? I feel like CLIs for me work really well once again for planning, but when I'm looking, especially at older code bases, I really like using agents in the IDE instead because I like being able to see the code. For new code bases, CLI all the way. For old ones, I kind of like going through the code myself. JSON formats are best because it's easier for AIs to play with. I do I I do prefer JSON stuff in general. I feel like it works best with a lot of different features, but at the same time, I want it to be as flexible as possible. And so, I'm going to say I would like to prioritize flexibility and portability. Ideally something that is very low touch where as a user I could save it once and never touch it again if I don't want to. So a file in a Dropbox folder for example. Awesome. Or if in the future I provide some kind of syncing service, it would take said file and just store that somewhere in the cloud. I'm making this up, but I also am genuinely curious about how it would approach things. Let's see. I'm looking at the comments. There's a bunch of things. Again, I'm I'm looking at YouTube, Twitch, and LinkedIn. And there's just enough comments that it's kind of hard for me to catch up. I use JSON mixed with SQite. JSON for front end and then JSON to SQLite. So, you get the best of both words. Oh, you know, that's a good idea being a Yeah, having both. Do you David, have you found a discrepancy between the two of them? because that's I I want to like keep things well synced while also keeping things again like very light touch, you know, and so having both. Have you had that kind of issue before? Let me know. I'm curious. Okay, let's see. For the MCP server, what's the intended use case? Should it allow other applications to retask, create, modify them, or both? Um, let's see. I want this to be something that's primarily read locally. Although, if I host syncing externally, that's a whole other thing. This has been such like a local only simple app that these kinds of questions are good to help me think through a little bit which once again this is why I like plan mode. The fact that it gives me a lot of Socratic questions where I have it doesn't actually give me the answer. It just kind of gives me questions to think through that helps me so much in my own app planning for sure. And so anyway, I want this to be something that's primarily read locally where if the app is running and for example, I'm prioritizing what I want to build from VS Code. I could say, what tasks do I have today? and it can pull from to-do meter easily. That being said, if I do want to eventually provide cloud support and sync, enabling some kind of API service in the future is something I should consider. Do it that way. Should the MCP server be able to detect changes made by the Electron app in real time hot updates, or is it okay if it only reads the current file state when queried? Current file state when queried, I think. Yeah. Yeah. Let's be certain. I almost started to say I think would be best, but we'll just be specific. I love that it's asking these questions and honestly the questions that it's asking are better than I normally get and maybe it's because this is a more complex app or maybe it's just the latest updates in the CLI but it's it's really nice to have these sorts of things tracked so that once again I can kind of work work through them in a particular plan. So anyway this is this is how this works. This is how plan modes works in the CLI where you can create this plan, figure out how you want it integrated and then run it. Oh, J uh David's talking about JSON. It's actually been running very smooth. JSON is just faster and gives faster user interaction. Then periodically or on exit, I'll write to SQLite. So I have the secure permanent DB that I can append to with the live JSON data. Okay, that that is helpful to hear. I can Yeah, I was considering something like that. And actually, notice how it has the SQLite schema and then the data abstraction layer. I wonder if it's going to do something very similar to what you're saying right now. I feel like that makes a lot of sense and what I was kind of expecting I might have to do. And once again, you know, me thinking as like a developer who hasn't used AI for a majority of her career, I keep thinking like, okay, but maintaining this, building this, some of this sounds just complex enough where I'm like, okay, I have to figure out how I'm going to do it. But then now I'm also like, but I have the power of agents. A lot of them could do some of this maintenance for me and I don't have to worry about it. And that's a really freeing thing to think through, you know, it's pretty nice. I also just noticed the time. So, really quick, before I do anything here, it has created this plan. So, this is a markdown plan that I can then either take and give to my team, give to myself, I'm a soul maintainer on this project, or various various other humans, or I can hand off to different agents. And I could say, okay, one bullet point I want to defer to this agent. Another bullet point I want to defer to another one. I can run that with a fleet command. I could have just one agent build everything in parallel. I could have an accessibility agent build something, a backend specialist agent build another thing. I have I have this plan and I can run it and use it. And that is very nice to have where it it lets me kind of be in control of how things are implemented without having to um worry about all sorts of maintenance tasks. And if I don't like this plan for whatever reason, I could say actually don't implement it. Now I know that that's how you were thinking. I'm going to go back to the drawing board and figure things out. or I want to update the plan with more of my thoughts. It's nice to have that all working under the hood. All of that working for me in a glorified markdown document that I can then manipulate myself. So, that being said, this has been rubberduck Thursdays, everybody. I really uh have had a good time talking about all the things with the GitHub change log, all of the things with the CLI, all the things with the SDK, all of these tools with you today and we will see you next week. It's been fun and I'll see you next time. Bye.