Getting Back Into AI Development After Time Away? Here's Where to Start

Transcript is auto-generated and may contain errors.

Hey folks, we are going to a submitted question. This one is around getting started with AI in particular withnet development. And so this person submitted a question to codecomute.com and the challenge that they faced is that they've been on maternity leave sort of a while they feel like a lot of the the AI boom has been kicking off and so at work the current project that they're on they're not allowed to to use AI for it which uh you know different workplaces and stuff there's going to be different restrictions for everyone. Uh that might even be that you're allowed to use AI, but like different tools and stuff you can't. So um you know this this person's particular case, they're not able to use AI for the project they're working on, but they're asking, okay, I got to get started. I know I have to get started using AI.

Kind of been missing out on it. How do I start? And because it has to be outside of work, like what are some ideas for some for some projects? Right? So this person is a .NET developer. I said, which is convenient cuz I do a lot of .NET development. And um I figured we'll talk through this and then I'll try to to open it up to be a little bit more generic, too. So, if you're not a .NET developer, it's not like this is going to be completely useless for you, but we'll try to give it that kind of kind of spin as we go through. So, um I think the first thing to to start off with is that like there are if you've been kind of on the sidelines of this, you're probably um aware there there's tons of tools. There's tons of different things.

And to be honest, it's probably going to feel a little bit overwhelming. And uh sort of my suggestion to you is like first thing is like acknowledge there's different tools. Uh the second quick part as I go through this is like different tools are going to serve different purposes. And then the third part that I want to say is that um I I do recommend that you try different things out, but you don't need to try everything because if you try everything, you're it's kind of like you'll you'll never get used to anything either. So, um the few things that I would recommend kind of dabbling with are of course like uh Chad GPT just in general. It's not going to be the best thing for like you know getting code created but uh everyone's familiar with chat GPT um yes it can write code it can do a good job uh but it's not going to be like integrated into your workflow per se.

So chat GPT cursor uh cursor is basically like Visual Studio Code if you're familiar with that. It's a fork of it, but you have um you know the agentic tools built into the IDE. Uh VS Code of course also has this. Uh at this point they're they're similar in terms of like capabilities. It's just that you might have a preference for one over the other. So I would recommend trying both. Um if you're a .NET developer then uh Visual Studio like the full-on thing. Um, personally as a .NET developer, I don't like using VS Code. Um, it's probably just because I've been using Visual Studio forever. So, VS Code feels a little weird. So, I have to throw in Visual Studio, the full-on thing there. And then, of course, Claude. So, Claude Code. So, you know, again, just to quickly recap, cursor and visual studio code, Visual Studio itself, claude code, and then chat GBT.

All of these have sort of different strengths. Um, and I'm going to go through those super quick and then we'll talk about some project ideas and stuff. Um, okay. So I do think that at a high level some of the some of the major workflows I guess are are going to be like conversation. Um there's going to be what I would call like co-piloting uh and I would say like supervised co-piloting and then uh the the third would be really just like hands off. Okay. So what does that even mean? Well, okay. So conversationally, I think that there's a lot of benefit to be had for building with AI. And the nice thing is like literally treat it like you are uh you know in this case like if you were talking to me to to answer this question, right?

So, not that I'm trying to get myself replaced, but if you said, "Okay, I want to I want to go, you know, build a new .NET project and I want to practice uh leveraging AI more." Literally have this similar conversation with chat GPT. Okay. One of the things that you'll notice is that uh picking different models and uh they have different uh mechanisms now like there's a deep research kind of thing. deep research will literally go take many minutes before it responds to you. It's searching the web and stuff. Uh there's a thinking mode which I would say for even a question like I just asked thinking mode is probably uh I mean that's probably great. It's probably even overkill to be honest cuz that kind of question at a high level doesn't involve like tons and tons of thinking and reasoning.

Um, and then there's sort of like these quicker modes that like um are probably the more traditional uh you know chat GPT style conversations where it's just replying to you uh pretty quick. But I would use chat GPT for things like this. You can do the same thing in uh you know with Gemini Claude uh with Copilot even like you can still use these tools to have conversations but the point is in sort of a conversational approach you're not uh the in my opinion the focus isn't like how do I crank out a ton of code it's like how do I explore this problem space all right or I just want to understand how something works um you know pretty pretty typically You know, if we were getting started on a project and we said, "Okay, you know, that you want to go build something in ASP.NET Core, you're going to go build a uh, you know, a calendar app, a to-do list app, a a habit tracking app, whatever it happens to be.

Um, you know, you're going to go build this and when you start, you're like, "Oo, like, how do I get how do I even do authentication or how do I um do, you know, hook up to a database?" Usually what we do is we go onto Google and we search this and we might get to a blog article or we might get to a YouTube video. Uh obviously like Google has their their Gemini like AI search in the Google search results now. Um but you could go to your conversation with chat GPT or whatever your favorite LLM is and you can ask it these questions too and just like keep it in the conversation, right? So you can you can and I recommend that you do use it to go explore getting questions answered like this. The conversational part is so powerful because it's like having a person.

The the thing that I would recommend and this is what you should do with people too is like you don't just take anything that people say blindly and go run with it. And I would say that's the same thing about the stuff I'm telling you right now. Right? I I never want people to think that, you know, just because Nick said it, like, go do it. It's law. Uh, not the case. Like, this is my opinion, my experience. Go validate it. And you should do the same thing with Chat GPT or any other LLM. Right? If it's telling you things, go double check it. Go do some of your own homework. Okay? But conversational modes, I think, are super helpful for exploring problems, for getting questions answered. And the one of the best parts is that you can ask it to explain things to you in different ways.

So if you were to watch a tutorial or read a blog post and you're like, "Man, I still don't get it." You're kind of kind of screwed because you have to go find another one. But when you're using AI tools, you can say, "Hey, I still don't get it. Explain it in a different way." Okay. Um, supervised agent mode is kind of like the pretty typical flow that I would say is you're getting some prompt structured. Uh, you want to give it a lot of detail uh, upfront, right? You don't want to leave things ambiguous because it's just like a person. If you were to, you know, expect something of someone and you left a lot of details out. If it comes up with its own choices and they're not the ones you wanted, it's kind of hard to be disappointed in it. you didn't give it enough information.

So, when you're doing sort of like a a supervised co-pilot approach, you're watching it, you're trying to give it clear instructions. Try not to give it too much all at once because the more you give it all at once at this point in time, uh the more opportunity there is for it to go off the rails. So, trying to keep uh you know, smaller features, smaller refactorings, things like that. But you're basically in Visual Studio or VS Code. You're in claude code. You're in cursor. Whatever it is, you are giving the instructions to the agent and then it's going to carry it out. Usually, especially now, uh, a lot of these tools will create like a small plan in like a markdown file or you'll see it kind of write it out and these are the steps it's going to go follow. And having these uh checklists or these like small plans is uh extremely beneficial for keeping the agent on track.

So you'll see it do that. Um and then the the nice thing is if you're watching and you're like, "Oh crap, like that's it's doing something I totally didn't ask or didn't expect," you can interrupt it and you can say like, "Hey, like that's not not what I want. Here's what I want instead." And that's where the supervised part comes in because where a lot of the you know the push back on AI development has come from is like you know it's making slop or it's doing stupid stuff and it's like sure but like if you're supervising it you can you can stop it from doing that. Uh there's more advanced techniques for like how do you you know over time have a feedback loop to prevent it from doing things like that but we'll save that for later. And then the final thing on this is just like you can you can go build out these features and stuff like almost fully autonomously like GitHub copilot.

Uh or you know if you're getting a lot more detailed in your prompts up front uh you can say like go follow this markdown file or this uh this entire prompt and like here's the entire feature delivery plan. Uh, honestly, I would say if you're still getting into this stuff, having AI go completely make things for you without being involved is uh I would kind of save that for later. It's just a little bit more advanced. Um, you just might find it frustrating because it's like sometimes it'll make things and you're like, I don't even know what this is or how it works. And it's it's not ideal if you're trying to learn and explore. But I would try to progress to that. Okay, so that's bunch of different ideas for how to use AI. I briefly talked about uh you know just some some buzzwords, some keywords for you to go look at some of the different tools, right?

Cursor, claude, co-pilot, um between Visual Studio Code, full-on Visual Studio, especially as a C developer. Um but what do we build? And this is the cool thing because I think the the reality is this part doesn't change. Uh in my mind when people talk about building side projects and stuff like that, I always say like, well, what's the goal? Okay, so sometimes people will say, well, I want side projects so I can uh put them on my resume. And then I say, great. Like, what is it that you're trying to show on your resume? you know, and and there's no like there's no wrong answer for this stuff, right? It's like, well, just what is your goal? Um, and in this particular case, it sounds like this person's trying to practice building with AI.

So like my suggestion here is like either try creating uh you know go shallow on a few things or pick one thing and think about the different features that you want to build into it as sort of like these uh individual experiments that you're going to go do. Right? So maybe you we're going to go talk about building um a habit tracker app. Okay. Um maybe you pick different parts of this and you build it out different ways with the AI tools and approaches we talked about. Okay. So to give you one example um maybe to go make the first part of the project you just go through the Visual Studio like new project wizard. Okay. So now you have a solution with a project file and it's an ASP.NET Core app. Uh, which means it's going to be the the weather forecast um like sample project.

Sorry, there's tons of water on the road, so I'm hydroplaning an awful lot. Um, and just trying to keep my brain doing a couple things at once here. It's been really bad on this highway. So, you have this app. It's a weather forecast app out of the box. It's not a habit tracker app. So, um I guess actually if you're doing uh ASP.NET Core by default, it's just going to be an API. So, like maybe do you want to do Blazer actually for for a front end as well? Uh do you need to go build out with a uh a different kind of front end? Like do you want to use React as a front end? Oo. Looks like someone crashed up ahead, too. Oh, man. One sec. Sorry, folks. Probably because they were hydroplaning on this shitty highway. You'd think for as much rain as we get that people would have this problem uh more figured out over here, but some of the areas on the road are just ridiculous.

Um, so you can already see that it's like I can't tell you how to go build or what you need to build like do you need a blazer application so you get you know front end as well as your your back end do you need to go build asp.core and then have like a react front end or whatever you want to use for your front end. So what would be a good a good way to go figure this out right? We could go to Gemini, Chad GPT, any of your favorite LLM tools and ask it, right? And you can balance that with like what is your goal? If you're like, dude, I don't I don't want to spend any time with like React or JavaScript. I just don't want to do that. Okay, don't like that's fine. Tell that to the LLM. It's probably going to recommend if you want to stick with C.

Blazer is why Microsoft created that so that you can stick with C. But if you're like, "Hey, no, that would be cool. Actually, I can figure out how to integrate, you know, two different things. One is in JavaScript, the other's in C." Then like, yeah, it might tell you React or Angular or what, whatever you want to use in the front end and an ASP.NET core server. But that's the like one of the perfect things that you can use chat GPT or LLMs in a conversational mode to go figure out and then you can ask it why. If it's like you should use Blazer and you're like, "Dude, I have no idea what Blazer is. Why would I use that?" Ask it, right? And then it can guide you through and explain things. So, point being that you might have your own goals and you can go back and forth and try and and figure this out.

So, you have this conversational mode to kick things off. I would say once you have like a little bit of code, right, you have a a basis for your project. So maybe it's a Blazer uh sample project you're going to start. And I think that has a counter. Does it also have the weather app? I can't remember. It has a couple of pages so you can nav and stuff as well. Uh you can even pick to have off uh included in the sample. But now you have this. Okay. And so well what's next? Do we want to build a habit tracker? That's not a habit tracker. So what what do we do? This is where I would say using something like chat GBT sucks a little bit more because when we're talking about working with AI, we want to be able to provide a context and you have a let's assume like a blazer application at this point.

How are you going to give Chad GBT the Blazer context like your actual code that's sitting on your desktop? Well, you can I think uh and I still haven't used it and I really should play around with it. I think uh OpenAI has uh uh codecs I think is what it's called. So I haven't used it yet. I can't comment on it, but I think that it's kind of like some of these other things where you can provide, you know, your uh your code directly to it. Uh that's just a good reminder to me that I probably should go try that out and make a YouTube video walking through it. Um but this is where something like VS Code or Cursor or even Claude would be really helpful. Um, Claude, by the way, you can use Claude as a plugin to Visual Studio Code, or you can use Claude Code, like the command line application.

Uh, Claude, you know, at the time that I'm recording this, everyone is freaking out in a positive way about Claude Code. Uh, I think it's great, but I don't like using stuff in a terminal. I'm I'm I'm the guy who wants a guey, so that's my preference. But this is where I would recommend using these tools because they can have context to your codebase and when you go to chat with them, it's a this is really a skill that you're going to want to refine is like how do you give it the relevant context? If you point it to your entire code base, uh you know, when it's small, it's really not that bad. when you have a large codebase and you're like go make whatever. Um, it's probably going to go do an okay job figuring things out, but you're you're also leaving room for it to go pick the wrong thing to focus on, the wrong pattern, the wrong spot to integrate it.

Um, you know, go refactor the logic that adds the numbers together. Okay. Well, are there multiple spots in your codebase that add numbers together? Did you want only one of those? Did you intend for all of them? Um, is it possible that it, you know, found three of the five spots where you're doing that? Like, you give it the context and it will steer it significantly better. Um, so context management is a huge part of this uh to really, you know, practice and skill up on. And so this is where I would, you know, go back and forth with AI in smaller increments to go do little features or uh refactor code in a supervised way. So you you give it the prompt, you say, "Hey, like I want to go build out a new page where people can uh list their habits." And you know, if you're already thinking through this, like how is that going to work?

Like did we even talk about a database yet? Um, you might say, "Hey, don't worry about the database for now." Like, "I just want to get the visual part." You can stub out um, you know, the calls to go query data. Like, you can can and you should kind of give it this bit of guidance. Oh, there's no parking spots, man. Oh, no. Okay. Uh, we'll do a lap around the gym. So, at the same time, you might you might not have realized that and it might go figure that out and go, "Oh, we don't have a database. I'm just going to start like you you'll have to play around with this and see because sometimes it does way too much and other times uh it does not enough." So, the more you play with it, the more you'll see kind of like the different behaviors that these tools have.

I don't know. Maybe here is a good spot. Sure. But this is why like the supervised way is such a a helpful way versus just like, you know, co-pilot, go build me the the list uh view of this and then you come back and it's like, you know, has a bunch of methods stubbed out with to-do and you're like, oh, this is so frustrating. Like, why does this thing suck so much? And it's like, you didn't give it clear enough instructions. Did you expect that it went and integrated with the database? It will. It will go write the code to do it. Did you tell it what database to use? Did you tell it MySQL, Postgress, SQLite? Did you say that you want a repository pattern where you can, you know, swap out the database? Like if you don't know, um, you can't be disappointed if you're not giving it uh those instructions.

So that's why I do recommend kind of breaking it up in smaller pieces. If you're like, I don't even know how to break up the features into small pieces. Where do we go? We go back into a chat mode, into conversation mode, and we have this discussion and ask it, I want to build out this list view feature. What do you recommend are the pieces or the steps to go do this incrementally, right? We can keep using AI. We can keep using different tools to go tackle these different parts, but um I realize I didn't spend a lot of time talking about the different types of projects. And honestly, I would just stick to like really simple things like habit trackers, calendars, uh you could go build a a web calculator, like simple things, right? And you can keep adding more features to these and build them out with AI.

So, you're practicing or you can say, "Cool. I've experimented with a, you know, a console app that's a calculator that got built. I want to try Blazer. I want to try an ASP.NET core server. I want to go try uh Maui development." you can go, you know, more broad. The point is that you have lots of different tools and mechanisms to go try that out. So, I hope that helps. Uh, and of course, if you have more questions on this, just leave them below in the comments. You can go to codeame.com and submit questions and otherwise just message me on social media as dev leader. And I will see you in the next video. Take care.