Coding with AI For Noobs

This episode of The WP Minute is brought to you by Omnisend. Do you sell stuff online? Then you need to meet Omnisend, the powerful yet refreshingly intuitive email and SMS tool that helps you make $73 for every dollar spent. Omnisend is that good and effective. You probably know the struggles of investing in digital ads with no guarantee it will bring money back into your pocket.

Use Omnisend, and you won't have to find creative ways to make your audience noticeable. You can communicate directly with them. Utilize Omnisend pop-ups to build a contact list and send campaigns directly to people who already want to hear from you. Automate everything with Omnisend. Just set it up once, sit back, and watch those sales roll in.

If this sounds too good to be true, check it out at Omnisend.com. Visit Omnisend.com. Tell them thanks for sponsoring The WP Minute. Mark Samantha. Tell them thanks for sponsoring The WP Minute.

Mark Szymanski, welcome back to the WP Minute. Uncle Matt, it's always a pleasure to be here. Thank you so much. It's been a while. It has been a while.

You know, today's topic is AI. We're going to be discussing my exploration with coding using AI, and hopefully, I'll be getting a little more serious about it. I saw you posted a video, and I almost trolled you on that. I'll save that for our later conversation.

I saw you put out a video using AI to code a plugin, so I'm excited to hear more about that. Of course, I'll link it up in the show notes. But yeah, this time instead of me leading, I'm just as lost as you are.

So, this is going to be an interesting conversation, right? Usually, I'm here coaching you through something, helping you understand something. I'm just as lost as you are with AI, maybe even more so. I'm kind of interested to talk about both of our experiences and what we're up to with it, if that's cool with you.

That's always great. Always great to chat with you, brother. Yeah, I mean, well, I just want to say right off the top, I know we were talking in the green room before this, that I've only known you for, I think—I can't remember when our anniversary is for knowing each other—but it's been around a year. In that time, you've gone from complete disdain for AI to now realizing that it's coming, that it's here, and that it's going to continue to get a little more interesting. You've had a chance to look at it a little deeper, and now, as you're saying off-camera, you basically feel like a senior developer. I feel like a senior developer too, but the code output is probably abysmal.

We definitely need to dig into that. Yes, we definitely need to explore it further. Let's discuss the idea of giving up the reins to AI and my current thoughts on the matter. I still believe that it is primarily a very good assistant at this stage. It is replicating tasks effectively.

So I've been using it. Claude, I've been using Claude largely over the last, I don't know, six months. Generally, it's how I shape a lot of my show notes for the podcast episodes that I put out. Really, that's what I use it for mostly, and it's been fine for that, right?

It's been okay for that. These are things that are, you know, low value. I just need concise information. I need speed and efficiency. And that's where Claude and its recent iteration of its LLM have been really great, whether it's 3.5 or whatever it was as of this recording.

But anything else that I was trying to have it do on the things I was looking for, like, I don't know, even analyzing data, throwing a spreadsheet at it and asking it certain questions, just wasn't working for me. I'd say even like six months ago, when I started getting into it, at least with Claude, I was asking it to code some stuff for me, and it just wasn't really effective. Fast forward six months, and I've made a lot more headway, probably because the tools are getting better. There is one tool that I started to use, which is literally life-changing, and I just discovered it over the weekend.

But all of this is to say that it is still just doing the things I'm asking it to do. Writing words is going well. Writing code seems to be going great because the projects I'm building are working. However, it’s not really helping me or guiding me in making the right decisions. In other words, there still has to be a layer of human interaction with this technology.

Uh, I think anyone taking it seriously will know that. And certainly, if you're doing it for your business—like if you're in marketing and you're trying to create blog posts and pages for customers, or even write code for your WordPress website—I think, at the end of the day, you still need to know what you're doing and understand the concepts. Right now, I'm totally lost with what I'm doing.

I just keep smashing the prompts, and it keeps giving me code. I try it; it works or it doesn't. Then I ask it to do it again, and I move on to the next step. It's the worst way of doing it.

It's totally unoptimized. I'm burning oil, and I am contributing to the world's greenhouse gas problem right now because I'm making NVIDIA CPUs and GPUs go nuts with the code I'm trying to write. There's no optimization on my side. Does that make sense?

I feel like it's still not at a point where it can completely solve the world's problems. That is a really good example of where I would say most of the world is with these tools right now. I want to make some quick disclaimers and continue to set the scene a little bit. The first disclaimer I’m going to make is that I think I can speak for both of us when I say that we don't know how this technology works technically at all. We can only pseudo-describe how it works, right?

It's basically as if they somehow coded something that went across the internet, so to speak, gathering all the language data that we have, and essentially built a sort of language model. We can think of it as almost like a small functioning brain, but it doesn't really expand too much on concepts. I feel like we're getting closer, but it still doesn't delve deeply into topics. It's really just a model of data that it can pull from, giving the impression of interacting with a person or something similar.

It's way better than a Google search in many ways because it has all that data and can express it back to you. It's kind of like talking to a person. For people who are completely removed from technology, it's very new, interesting, and cool. Once you play with it a little bit, you start to understand what's going on. You also recognize some of the limitations and other aspects.

But that's a big disclaimer. The other big disclaimer that I feel I can set the scene with is that, as of November 2024, Claude, ChatGPT, Gemini, and a ton of other options are out there. However, none of them are, and I think this is the key point you're describing at the beginning, magic. They are not mind readers.

They're not independent thinkers in any way. They're not mind readers. They are literally just assistants for the most part. I feel like, in the last month or two, I've really stepped up my prompt engineering and my utilization of these tools.

The mindset change I experienced was like, this isn't magic. It's similar to being a senior marketing director or a senior developer, with a junior team member sitting next to me. I can take on more of a managerial or director role while they handle some of the work. They won’t be able to do everything, but it’s actually quite amusing. I’ve thought about this many times, and although I don't have an extensive background in a corporate environment, it feels almost like that.

It can take the place of some less senior positions. I'm not saying that we should get rid of junior developers or anything like that. I'm just pointing out that this technology has a certain level of capability. However, as you mentioned, it needs to be directed by someone because, at the moment, it's not quite there yet, at least in the way that most people are currently using it.

And again, the biggest disclaimer of all: this all started only a few years ago, so we are extremely early to this. That's why I'm creating some of this content—I know this is where it's headed, and things are going to get crazy. Recently, I had a conversation with my brother, who has been using ChatGPT for a while. He primarily uses it for coding tasks. A couple of weeks ago, we discussed how this technology is changing the world. He said, "Hey, man, this is really going to make things weird over the next couple of years." The way people interact with the web—whether it's searching for information, creating content, writing code, or engaging with other sites—is definitely evolving. I agree with him, but I feel like we're in a bubble right now, similar to the housing bubble a couple of years ago or the period during COVID when money was essentially free and everyone was getting mortgages at around 2%. It seemed like money was just out there for anyone to grab. I feel like we're in that mode right now with AI, where the costs are still too low—like the $20 a month to use what I'm doing. I can't see that being sustainable in the long term. I think we can expect that in the near future, a lot of these platforms will likely start ramping up their prices significantly.

Like we see with streaming services everywhere, right? They are burning cash and power, literally. Inevitably, there will be a major corporation that needs to win, as they always do. We're going to start to see the tightening of what’s available and at what cost.

So the way I see it is that all these guys, Claude and ChatGPT, are the only two I really use. I don't use Gemini or anything else, really. But those two are just going to be like, "All right, we're done." It's either going to be at least a hundred bucks a month or two hundred bucks a month.

What is the pressure point where people will pay, or else they will not have access to all the new cool features they are currently getting for $20 a month? I believe that's when things will change dramatically regarding who has access to what. The question becomes: what is your best AI, and can you afford to use it? I think that's where we are headed. It won't be the same commodity that we are accessing today.

I mean, I could be wrong, but that's the way I see it. It could be. I wish I knew who it was.

And if I find the video again, we can link it below. There was a video—it was just a short one, like a one to three-minute video—where this guy was saying that around the time when OpenAI went private or public, I can't remember which, they had a weird non-profit structure at one point, similar to what Matt Mullenweg and Automattic have. It was just like, "What?"

Yeah, what happened there? Around that time, I saw a video where someone mentioned that they weren't worried about what you just described. It was interesting how this phenomenon occurred. We heard the term "AI," saw ChatGPT, and then, seemingly overnight, everyone started implementing these technologies. It's strange how that works; sometimes, it only takes one person to say, "Oh, I figured it out," and then, despite not being able to figure it out for years, everyone suddenly starts releasing their own versions of the same idea.

I guess just reverse engineering or whatever. But because we're seeing that and all these things are cheap, people think that everything will continue to be inexpensive. Now, certainly at certain points, there will be more developments, and maybe some will be more expensive, but then they'll come down to a more commoditized level. I just don't believe that this version of AI will ever truly be expensive because it's already affordable, and it doesn't do anything completely groundbreaking. It's definitely a significant shift, but it's not like you can simply tell it to do something.

We think about it, and it’s not like you can just tell it to do something. I mean, you know, we consider this. I don’t know if you’ve seen it, but obviously you’ve heard of Tesla and the cars and everything like that. Those are still kind of a premium product. I don’t know if you saw the Tesla bots that they showcased; that level of technology feels like AI 2.0 or whatever they’re going to dub it. That’s going to be the next big thing once it starts moving into hardware and physical items that will actually enhance your life in a more tangible way, rather than just the digital stuff we can do now. I work in this field, and I believe that’s when the world is really going to change.

Because everybody's going to see that. When you see robots walking around and things, it's literally like "I, Robot" from that Will Smith movie. That's going to be wild. Yeah, I agree. And I think we're missing this layer right now.

I want to talk about our experiences with coding in a moment. However, we’re missing this connected layer. So, what would AI be really great for right now? I want it to look at my to-do list and prioritize it based on what I’ve prioritized in the past.

I want it to recommend what my next steps are to improve my productivity. Getting things done can be challenging, especially when you have a to-do list that includes tasks like writing blog posts. Sometimes, it feels overwhelming, and I wonder, "Where do I start?" I would like smart intelligence to analyze my list, break it down, and then prompt me with suggestions on what I should do next.

Here’s a reformatted version with correct capitalization and spelling: Like these are, but it's not going to happen. There's no AI to do that. Let me stop you there for a second. First question, and then there’s a follow-up: Do you think it can't do that right now? Do you believe that it can't, based on your experience, or do you think you just don't have the tooling? I just don't believe that the necessary tooling exists. I don't want to give up my Todoist app; I want it to work with it. You know, we've seen some of these things. I don't know if folks are just listening. I'll try to find it and link it up, but I know Claude launched some new experimental thing. I don't think it's generally available yet, but it can control your computer to do things. That's getting to the next level where, okay, I can tell Claude, "Hey, prune my to-do list, open up my email, and start responding to that stuff," but I still think that's so far away. And then, do you really trust that human interaction layer?

Do you trust the code that you're deploying to your server? Are you going to spot-check it before you do that? And are you going to trust it to manage your to-do list? You know, I don't know. But yeah, I mean, once it gets incorporated into that layer, I think we will see a few years of immense productivity.

But then what do the next few years look like where, you know, it's starting to do things for you, and you're like, holy shit, do I get paid the same because of these efficiencies? I don't know. That's why, again, shout out to my brother. I think he's looking at it and thinking things are going to get weird. Anyway, let's just talk about our experiences with coding.

And I'll start. So I have this idea for a web app. It's not even something I'm making generally available; it's more of a tool for myself. But I want it to operate as a web app.

One, because I want the challenge. But two, there are other ways that I can extend it and do some other things. But anyway, the point is, I was just like, all right, I'm going to get more serious about it. Let me see if I can figure this stuff out, because Cloud 3.5 came out, and I heard it was great for coding. So my first experience was just dumping all these thoughts into the cloud.

Here's the idea: Here's what I want you to build. I had some direction. I asked it a few things because I do have a sort of network operating background. So I was like, let's compare this to running on an AWS stack versus a Cloudflare stack.

I've heard some great things about Cloudflare—efficiencies, in particular. I just don't know if it's as mature as some other options. So, I asked it to compare and contrast some of these features. Ultimately, I chose the Cloudflare path because I didn't want to use AWS for this.

I don't even know if that's the right thing. So that's problem number one. I'm asking AI to give me the advantages of both of these technical stacks. It said, you know, Cloudflare is the most efficient. Okay.

You know, it’s so great! I decided to go with that. Then I just started pouring out my ideas and hammering it with the questions and what I needed it to do. The first major lesson I learned was about these interfaces. You know, whether it's claude.ai or chat.openai.com, you just dump stuff into them.

And it is just like this huge pool that you're throwing stuff into. What I found with Claude was that I just started burning all of the tokens. I'm a pro user at 20 bucks a month, but I would reach my limit in about an hour of asking it questions and having it code for me. So I found myself talking to it, and the answers it was giving me felt like it was guiding me.

It's like saying, "Here's what you should do next." Then it would prompt me and say, "Well, do you want to extend these features? Do you want more clarity?" I would be talking to it, saying, "Yeah, man, give me more clarity because I don't know what the hell I'm doing."

So, it would provide me with these answers and elaborate on them. However, it was all coming so fast that I was burning through the tokens, and then it would just say, "Hey, you can't use 3.5 until whatever, 2 p.m." It would be like a three to four-hour window, and I’d have to shut my laptop and think, "Ugh, I can't interact with this anymore." But I could switch to the other version or whatever it's called.

The answers were vastly different. The experience of asking it to code was surprising as it didn't know where it left off with my previous conversation, which was super frustrating. The first lesson I learned was that you still have to be dialed in. Otherwise, at least with Claude anyway, you're going to burn through those tokens by asking a million questions and just having it talk to you.

If that makes sense. No, yeah. And if you've tried Claude, I would recommend you take a look at it. But what Matt's describing there and what you're describing is entirely like that. So I made a couple of it, but what Matt's describing and what you're describing is indeed entirely like that.

So I made a couple of videos maybe two months ago at this point where I was just diving into this concept of literally maybe my first time using Claude and trying to code. When I did that, I had that exact same experience that you had. I was like, "What the fuck? I have all these questions still, and I can't do this until tomorrow or whatever." I'm not sure if there's a way around that.

I don't know if you can buy more tokens. One thought there, though, is that whatever you're spending to get that information, you absolutely would have spent way more money trying to get somebody to do it for you. Oh yeah, sure. That puts things in perspective. But again, your point is completely taken.

That's kind of annoying where we are right now. I do think it's probably a bit of, again, where we are currently, so to speak, with being extremely early. But the fact that we can do it is something we can focus on, and that’s the main point. What I've learned from my second go-around with the video I just posted today, as of recording this, is that it’s like an hour and a half long.

And I did it specifically in two ways. The first part was just me explaining what I was doing and the finished product, and then the latter half hour or so is me literally in Cloud, playing around and building it. There are a bunch of little gold nuggets in there, but a couple of the ones I found were that if you're going to do this, you must realize that you have to continuously get better at being a prompt engineer. You must understand that you need to provide all the necessary information and ask it to prompt you with questions—particularly one at a time—so you can have a conversation. There's also some information in there about saving tokens and all that. But the first time I did it, which is a really big point, was when it gave you a guide for making a simple WordPress plugin.

So, it codes a bunch of stuff for you. It outputs the files and the nice little artifacts feature, so you can kind of see everything. You can download the files and save them on your computer directories and all of that. If you ask it to change something or if it has to rewrite it, I think what it does on purpose is only give you part of the file back because that uses fewer tokens. It doesn't want you to rewrite everything. I was telling it, "No, I want you to rewrite the whole thing for me so I can just copy and paste it." That's not really a good idea because you're going to burn a lot more tokens. Yeah, that's what I was doing too. This is what I mean: you have to be the senior developer, and it's the junior developer. You're still working; you're just guiding it, and it's doing a lot of this stuff.

So that's one thought on my mind. We could discuss whether the code is good or not. I believe that if you're working on something relatively ubiquitous, it’s likely to be quite good unless it’s hallucinating, which hopefully you can sometimes identify. However, I don't know. You definitely want to get it vetted, let me just say that.

But the one other piece of it is, you mentioned where you're just kind of dumping everything into a pool. The cool thing that I've found with Claude is that it has a projects feature. I don't know what plan it's on or anything like that, but it kind of reflects where we're not fully there yet and somewhat on our way. Think about these LLMs, which are large language models that have scraped the internet and analyzed all the data. Okay, that's the whole LLM. What I want, though, is something different. I've tried this, but I haven't made a video or fully documented it yet.

What I want to do is have the LLM understand my needs for this project, as well as information about me and my business. The secondary aspect of this is similar to what some people have done with personal AI chatbots, where they essentially recreate themselves so that people can ask questions and receive answers as if they were speaking directly to that individual. What we should all be doing is building our own personal LLM. This is actually a great piece of content that I should write about. While we have been focused on building our Facebook pages, LinkedIn profiles, websites, or blogs, we need to shift our attention to developing our own personal LLMs that encompass everything about us. This way, at any time, either we or someone else could access that LLM and get the answers they need.

Because that is like, I feel like maybe 1.5 or 1.75 in terms of where we need to go with some of this stuff. Because if you can do that and create your own little segmented area, then you've added more information to it—kind of like how you're doing, right? You included all that information, and if you could silo that, you could always come back to it. I know they do it in the chat formats, but the projects thing is a separate issue. You could always revisit that and continue working on that part of it.

There’s a lot to unpack here, but that’s been my experience. A couple of tips I’ve found so far in my short journey: when I got more serious about building this app, I purposely started in a Claude project. However, I still find it lacking, especially when you hit a certain point. I need to get better at understanding how this stuff works. But if you start a chat, let’s say with 3.5, I believe it’s called Sonnet.

Sure! Here’s a reformatted version with correct capitalization and spelling: "Yeah. Yeah, a sonnet, which I believe is what it's called. Yeah. A sonnet, which is, at this point, their most advanced tool that is accessible to average users. You start that chat, and you begin to go down the road of, 'Okay, build me this script.'"

And, um, I'm building something again that runs on Cloudflare. Cloudflare has Workers, which, as I understand it, is a serverless code architecture. Basically, I can run tiny, tiny, tiny little operating systems that execute these commands to build this large script that handles all these API endpoints, routes, connections, and so on. While I'm chatting with it, using version 3.5, I'm improving it—adding API endpoints, testing, building, and deploying it. Then I run out of tokens, and it's okay; I can't talk to it until four o'clock. I can't back out to, air quotes, the projects folder, and then pick up with, like, version 3.0 and say, "Okay, let's reference this file that we've been building over here with 3.5 and continue working." It just has no idea how to do that—unless I'm just not prompting it well enough. I mean, it creates the artifact; you see it. This is getting really inside baseball here, but if you use Claude, folks who are listening, you know what I'm talking about. You have the previous files that it's created in the project.

You can click on it at any time and see it, even with a name, because it assigns a name. You can actually tell it to give something a specific name, but if you try to reference it, it still has no idea what you're talking about. This is one of the most aggravating things for me, unless their projects don't work the same way I thought they did, or it simply has no idea because the 3.5 version was much more advanced. Let me backtrack. If you're asking version 3.5 to code something for you, it is literally like having someone sitting alongside you.

And it talks to you. It explains everything. It asks you what you want to do next and suggests things that you should be considering for future features. You're like, "Oh, I never even thought about that feature!"

And you're like, "Wow, this is amazing." Then, as soon as you go to, like, three, it's like the baseline one. It just says, "Here's the code." It doesn't say anything else—it's just, "Here's the code."

It doesn't say anything else. And it's just like, "Shit, man, what the hell happened? You were so nice to me in this other chat, and now we have nothing. It's vastly different."

And so, what I started doing too is, yes, I began to, as I understood the limitations I was running up against, to make the requests a little bit smaller. And then, in the—I forget what it's called—let's see if I can pull it up really quick. There are instructions, I think you can set them. I believe it's called "instructions."

Sure! Here's the reformatted text with correct capitalization and spelling: "Yeah, I set custom instructions for the project. In the upper right-hand corner of the project, it says, 'How should Claude respond?' I specified my preferences, and I mentioned that before writing a lot of code, I would like a quick explanation of what will be done so that I can confirm it. Essentially, I wanted to avoid wasting those credits or tokens by having a lot of code written without prior approval. I asked it to tell me what it was going to do first, and then I would approve it. Even with that, it didn't work all the time. It was effective when I requested larger tasks, and it would list some features it might implement in bullet points."

You know, it would ask me, and it was really hit or miss with that. I have another instruction that says, "When writing code, please use good inline documentation." I'm a beginner-level developer, and that helps me find the areas you need me to update. What was happening was that it would suggest changes to the code. Instead of writing out 800 lines of code every single time, which would be a complete list of the file, it would just say to update these sections.

And I'm like, "Fuck, I don't even know." I find this section. "Yeah, where the hell is that at?" So I'm trying to say, "Look, really document this and help me understand where it needs to go later on." So again, it's hit or miss with that.

I wish it was a little bit better. Yeah. I mean, again, I've had similar experiences, and it sounds like what you're trying to achieve is probably more robust and advanced than what I was attempting in that video specifically. I think that project had maybe seven or eight files or something like that. So I'm not sure if there's more to it.

But I mean, if you have more functionality, then files are just getting bigger. It's more tokens, more decisions, and more features. It's all that sort of stuff.

But I mean, I don't know. I'm bullish on it to this point because I think it's pretty incredible what we can do so far. However, all of the feedback that you're giving, I'm sure a lot of other people have had as well, and I don't know. The thing that confuses me the most is how you actually make something like this better. If we want to go off topic from coding, I don't know if you have more to discuss regarding the coding stuff specifically, but when it comes to other use cases, I've actually found a ton of ways to use Claude, ChatGPT, and Gemini in specific situations. I have a laundry list of things, just from the past two months, that I've used these tools for which have made a difference in all of my workflows, my thought processes, and the way I tackle tasks.

Let me just wrap up some of this coding stuff. So, what happened next? Here are the things I don't know. I'm way out of my league with what I'm coding. It's pure JavaScript.

I mean, I'm actually, I can understand some of it now, right? Because I've literally been neck-deep in it. And it's actually a great educational process. It sort of reminds me of learning WordPress all over again, which is cool, but it's way out of my league in terms of what it's developing. So I found an app called Cursor. Have you run into this yet?

I've heard of it. I haven't used it yet because when I was coding and initially started this project, I thought I needed to set up a whole local development environment. I haven't done that in years, and when I did it for WordPress, it was so easy.

It's both the good and bad of WordPress because it's quite archaic. It involves MySQL, PHP, HTML, and CSS, and of course now JavaScript. The essence of it is that you can set up a local environment pretty easily. In contrast, with newer technologies and package installers, such as npm install, you're building all this stuff, and then you have to deploy it. There's a lot more involved in getting things up and running.

And then, what are you coding in? So I've been asking, "Hey, Claude, what is everyone using for coding?" And they're like, "Oh, VS Code." Okay, so I started using VS Code.

And then my mind is blowing up because I'm like, I don't even know how to set this up. I just feel way out of my league because I'm thinking, this thing isn't set up right. I don't even know if I'm optimized correctly. Am I doing this properly? Then I started hitting that wall of doubt. However, over the weekend, I began making a lot of headway.

I built my Cloudflare Worker, and it was working. I could send data to these APIs, and it would communicate with other APIs to send me the information I needed. I thought, "Okay, I'm making headway." Then I realized I needed to create a front end because I couldn't manage everything through the terminal.

I need to be able to manage this in a web browser. And that's when everything fell apart because I was just like, "Hey Claude, build me a React app that connects to all these APIs that can do this stuff." It was just a nightmare. I mean, 48 hours of trying to figure out how to deploy this to Cloudflare. Is the problem with Cloudflare?

I don't know. Is the problem with my code? Probably, right? There's all this stuff happening.

And I found Cursor. I don't even remember how I discovered it. I think I was feeling like I needed to find an alternative to VS Code because I just wondered if I was using the wrong tool. Cool. I just can't deal with this stuff anymore.

And I found Cursor. Cursor integrates directly with Claude. The magic is that it's not your tokens; it's their tokens. So I could just pound away at this thing and say, "Here are the 900 lines. Look at it."

Tell me what's wrong. Alright, give me the update. Here's another 900 lines. Please provide me with another update. That has dramatically solved the issue.

And basically, what it has allowed me to do is in Cursor—anyone who's a proper developer is freaking out and flipping tables right now. When I ask it to do something, it shows me the code, just like I was doing in my Claude web browser. Instead of copying and pasting it into the file, I just say, "Apply it," and it writes it directly in the file. Right?

So I'm just like, "Oh yeah, all day, man." Like I just said, "Yes, yes, yes, yes." This is all the code I want. And I finally got to a point where I'm not using React anymore. I'm using another framework called Svelte, which is like a pure JavaScript play on it.

It's supposed to be more efficient, yada yada. I don't know. I'm watching YouTube videos while I'm coding. You are not coding.

You're not coding. I'm coding, man. You're prompting. You're prompting. I'm coding.

The way that I see it is that I'm the guy with the idea. We're working together on this. It's a partnership: me, Claude, and Cursor. I have finally reached the point where I’ve launched the app.

The very basic iterations of it are working, but the cursor is really the thing that helped me turn the corner. This has allowed me to use it in a more effective way. Instead of working with someone on the inside when a big issue arises, I can now ask it to help me style the front end.

So instead of doing that even in Cursor, because you have this tiny little window, I’ll start asking ChatGPT and Claude for their recommendations. Now it feels like I have two different people in front of me. You know, a big question, right? Like, "Should I use Tailwind or should I use some other CSS framework?" I’ll ask them both, see what their answers are, and then I’ll aggregate that information and bring it into Cursor.

Be like, "Okay, here's where I think I should be going with this. What do you think?" It'll give me the answer. And man, I’m a one-person enterprise, that’s what I am! Okay, I'm going to be launching so many apps, dude, it's not even going to be funny. All I have to say is WordPress is in trouble. WordPress is in trouble, yeah. Oh my God. Man, people are going to be so mad about this. No, it does help me really appreciate what WordPress has done, like really.

Like, even now, I'm looking at it and thinking, "Holy shit, there's a lot of legacy stuff here." I can understand why people on the cutting edge look at WordPress and go, "That's a joke. Look at that dinosaur over there." However, as an end user, imagine trying to do … well, imagine that AI is going to change all this stuff, and you can just create your own WordPress. That isn't going to happen, at least not anytime soon, and certainly not to the extent of what you can do with WordPress. Still, it does make me appreciate what we have in this little bundled application that is WordPress.

I have a lot of thoughts. The first one I have to share, though, is that I didn't realize it took less than a year. It took you only a week to become completely bullish on AI and essentially become a senior developer. That's amazing! I don’t know if anyone has ever accomplished that in just a week.

I'm going to quickly become, like, literally a senior developer. That's amazing! I don't know if anybody has ever done that in a week. I'm going to go back to the CEO level in a moment, but I'm just saying, you know, once I figure it out, it's smart.

Um, yeah. The one random question that I have, which I don't know if we can really answer, but it would be good to get other people from the community to respond to, is that we are not developers. I wouldn't classify myself as one, and I don't think you would classify yourself as a developer either, right? We know some code here and there just from being around; we understand what certain things do, but we've never spent eight hours multiple times per week, day, or year actually developing software or anything like that.

We know enough to be dangerous. And now, with Claw, we know enough to be really dangerous. So, I'm wondering if we're that camp, so to speak; like we're that archetype, and we see this and think, "Oh, shit, we could build our own apps and stuff like that. Or our own plugins."

And it's like, we know the dangers, kind of, if we're being honest with ourselves, but we're trying it because it's interesting and we're trying to learn. I wonder if software developers, as the other archetype, look at these tools and leverage them. I wonder if, like what you just described with Claude and Cursor, I don't know enough about this, and I'm sure there are videos and people exploring this topic, but I don't know if software developers are fully utilizing these advancements. You would think that if you put these tools in a software developer's hands, like GitHub Copilot or any of the many others available now, they would just be flying at Mach 10. Yeah, I think that's where the biggest wins are, for sure. You know what I mean?

So the thing is, it's not just us who have these tools; they have them too, so they're not in trouble or anything like that. I don't think that by any means. They still know what they're doing, so they can move faster and are confident in their abilities.

It's like if they encounter the problems that you and I are having, then it's like, "Oh, I know exactly what to do." Boom, boom, boom. I know exactly what to tell it to do, and they're off to the races. I think that's where Cursor is supposed to really excel.

I'm using it for its chat features, which I feel are supposed to be just an assist. However, I'm using it to code the entire thing. From everything I've been watching and listening to, people are talking about how amazing Cursor is because it has this autocomplete feature. If you know what you're doing, it speeds things up by about ten times. You can start writing the code, and it will autocomplete the code it thinks should come next. Oh, wow! Since I don't know what I'm doing, I'm asking it to write stuff for me.

And then that's what I'm going to focus on next. Oh wow, I don't even know what I'm doing! I'm asking it to write things for me, and that's how I'm using it.

Whereas, like, a person who knows what they're doing is, you know, operating at Mach 10. And then if you really have a question, you can ask AI. So, yeah, it really helps that kind of person for sure. I mean, just in general with the coding stuff, I love that these tools are available.

It's awesome that they're available to anybody. I'm sure some people will say, "Well, you shouldn't be coding your own stuff." But I mean, if you want to play around and you understand the risks, that's fine. In that video, I mentioned that I'll provide you guys with this code, but do not—do not—hold me liable. This is for your own use, so do so at your own risk.

Like, uh, it's there, but I don't know. It's got an encryption class for the API keys, but other than that, I have no idea what vulnerabilities you could find in this system or anything like that. So I would always advise exercising caution there. However, it is pretty incredible what we're able to do, and seemingly in such a short time, we have transitioned from everything being hand-coded to now being able to achieve so much more at a much faster pace—both for us and for people doing this professionally.

It's pretty insane! Yeah, so what other areas are you using it in, aside from content or coding? Well, the other area would be that we're not coders, right? Like we mentioned, but we are content creators.

And I think that you and I are probably similar; we're alike in that regard. We understand the nuances, the hang-ups, and the limitations of creating content. However, we don't really know the software side. So, if we switch to this content perspective, how can we possibly speed this up? Let me think of several different ways.

The first way—some of these are like small wins, as they're not groundbreaking, but they're actually pretty awesome when stacked together. So, the first thing is, we record YouTube videos. I’m not going to list them in order because I can't remember everything off the top of my head, but I’ll share what comes to mind. The first one is that we record YouTube videos, and I like to include chapters in all my YouTube videos.

I was literally doing this. I would watch the whole video back and put the chapters in. I don't have to do that anymore. I upload the video and wait—sometimes it's right away, depending on the length of the video. I took that whole hour-and-a-half script.

And ChatGPT, I tried to do this with ChatGPT a while ago, but it's not as good. Claude is way better at pacing things, and maybe it’s just the new model now—it can just understand more. So, I upload the video, go to the bottom, show the transcript, copy the whole thing with the timestamps in it, paste that in Claude, and say, "Hey, give me YouTube chapters and one-liners for it," just like formatted perfectly. I just ask for chapters for these videos. I can even say, "Hey, get a little more detailed," if it doesn't provide enough or if it’s not what I want.

That saves a significant amount of time if you were going to sit there and watch all of that. And that's just a small win, as I mentioned. So there's that. You already mentioned things like explaining. You could even say, I literally just tried this with the last couple of videos.

Let's stick with that. Just from the transcript, it knows everything your video is about, right? And it's an expert at YouTube if you tell it that. So you can get the subtitles from that. I'm sorry, you can get the chapters from that as well. You can ask it to give you 10 really good titles for this video based on its knowledge of YouTube. I’m not suggesting you copy and paste that part; I just need the ideas because I'm not creative enough to think of them myself. I've even experimented with this. I mentioned the description and other elements you could obtain, but I also tried asking Claude, "I know you can't generate images, okay?"

You can talk about images because that's kind of another mode of this piece that's good and bad—very hit or miss compared to text. I know you can't generate images, Claude, but can you prompt another tool to create a really good thumbnail for this? Or at least give me some ideas on what a good thumbnail would look like? It doesn't do it great because I copied that over to Grok on X, since I think Grok uses Flux, which is probably one of the best image-generating tools available now.

One of them. I put it over there. It didn't provide me with great content, but it offered something that might spark some inspiration, you know what I mean? Even with text, you can generate ideas for imagery and other things. Those are just a few ways I've used YouTube to speed up the process and come up with many more ideas, because, as you know, those ideas can be hard to come up with on your own.

Sure! Here’s the text with correct capitalization and spelling: "Yeah, I mean, it's... I don't know. Yeah, I mean, as an assistant, it definitely helps. It still hasn't really knocked it out of the park for me with titles and stuff. I think it's still very..."

Baseline: it does bring in a different perspective. What I'll do is ask, "Here’s the type of viewer I’m hoping to attract to this video—whether they are freelancers, WordPress beginners, or whatever the relevant keywords might be." Then, I consider what would appeal to them, using that as a basis for ideation. I also think about other keywords that may resonate with this audience. All of that information works great for the chapters, because I use Descript. Descript has a chapter feature built right in, so you can just click it. It's still all AI and ChatGPT-driven, but I've always found it to be a little more accurate because it processes the entire video file. It generally does a better job. So, yes, I often use that.

I use Whisper AI, as I think it only runs on Mac, though. Whisper AI, or Whisper Transcription, is what it's called on Mac. It's a Mac app, and you can just drop in your video or MP3 file, or any media file, and ask it to transcribe it. Then you can run the same commands on it. It's just like a local app, and I use that from time to time.

Yes. The other thing I try to do, although I haven't done it as much recently, is take the transcription because it's not great. I mean, maybe Descript already does this, but the transcript from YouTube isn't very well formatted. So the idea would be to take that text and put it on your website so you can index that text.

I'm not even sure if that's going to be completely necessary as we move forward, because it's probably going to get, you know, SEO juice and ranking, along with any sort of indexability from the actual words in the YouTube video at this point. But that's just another random thought. Yeah, there's a lot of things. I'm trying to think of anything else that I've tried to do. I just really hate making thumbnails, so I'm waiting for a really good thumbnail tool. I've explored some custom GPTs for that.

But yeah, I don't know. This is all kind of very interesting and slightly concerning at the same time because, at some point, you know… I mean, I don't know if you’ve experimented with Eleven Labs and any sort of voice technology. Very soon, we’re going to be able to just generate our own content using all the voice and video data we have of ourselves, along with any LLM stuff we’ve put together. It is going to get weird. I’m just hoping it doesn’t get weird too soon, or at least we can capitalize on it. There’s a lot to consider. So, are you still planning to build out that plugin and continue making plugins for the sites you’re developing? What’s the future of plugin development for you? I mean, that specific one was aimed at creating automation via the YouTube API—to automatically check when I’m live on my website. I just created a custom tool that checks the YouTube API to see if I’m live or not.

It provides a couple of pieces of data that it pulls in, such as the YouTube URL and other related information. Then, it integrates specifically with Bricks to utilize all those elements with dynamic tags and conditions. It works great—perfect! To achieve all of that, I would have either needed to hire a developer or find a plugin that could do it.

So I think for very niche things like that, you're definitely well off doing that. I'm considering if you conduct an analysis on the ROI of doing something like that—just personal ROI—nothing is going to come along and fix that right away. I would love to have a larger conversation about where this is all going. I love WordPress and I love websites, but I don't think we will necessarily be on websites for the rest of our lives. I believe there will be more to it, especially in terms of the different ways we interact with things.

Because if you think about the Google search era versus now, the AI prompting era, I don't know about you, but there have been times very recently when I have not gone to Google. I've gone to something like ChatGPT or Claude because I want the answer to come to me. I don't mean that I just want to take the answer; it has all that information, so it can explain it in a way that addresses what I'm specifically asking for. I'm sure you've experienced this: you have a very specific problem, so you search Google and have to go through ten links to find the answer. But now, AI does that step for you in a lot of ways. Another big thing I do—this is one of the things I don't like about Claude just yet, but I'm sure they'll improve it—I love ChatGPT and Gemini, which is Google's, for conversational use because I hate reading a lot of text. So I actually have a conversation with it and ask, "Hey, what do you think about this topic?"

I'll do this in preparation for any sort of podcasting debate or live topics. For example, what do you think about open source? Do you think it's a good thing, a bad thing, or something else? You can gather a lot of ideas and perspectives that you would normally have to read from various people's opinions, but now it's all aggregated and presented to you. I'm not saying to take it at face value; rather, you can engage in a mock conversation or debate on these topics from both sides. This way, you can explore all the perspectives, form your own opinion, and go into the content or the show with that understanding.

I mean, I've literally done that before just for the ease of it. I'm just talking into my phone; it's so easy. Yeah, there are certainly many different ways to use it. I was also thinking, while I was using the cursor, that if I could just talk to this thing, which it probably even has, I don't know.

If I could just talk about this and explain what I want. But I still have a lot of uncertainty about what the answers are going to be. Even from, you know, I don't know. I like to see a ton of options. I don't mind sifting through it as long as the options I have are close to, you know, giving me the answer.

You know, argue if Google even still does that anymore. I don't know. But I don't mind sifting through that stuff. But I totally get it. Like, just give me the answer.

I signed up for it primarily because of Kevin Rose from Diggnation. I think an advisor or the VC firm he works for is invested in Perplexity, and you could get a free year of Perplexity Pro if you were on his email list. I've heard a lot of people talk about it, referring to it as something akin to search results, and it’s supposed to be the tool that really challenges Google. I used it for a project I’m working on, but honestly, it didn't provide me with anything that Claude and ChatGPT weren't already offering. It's a cool interface, and I’m sure if I used it for more complex tasks, it might yield some better results.

But once again, I'm just like, I haven't tapped into the full potential of this stuff. Effectively, I feel like I'm just getting started, but that's where that is. We're all just getting started, Matt. It's kind of early. We're all just getting started, Matt.

You know, it's kind of early. We're all just getting started, and we're all on the way out at the same time. That is a larger topic. I'm sure this has already happened in history, in waves—obviously not AI-related. But again, like with so many other things, I think we're just going to have to continue to adapt. I believe that if we're talking about this now, we are potentially in the best position. I'm trying to engage more people through this type of content. If we can encourage more people to think about this and understand that it's real, we can foster a greater awareness.

The hope is that we can continue to leverage it for as much as we need. We need to evolve and adapt to fill in the gaps that these types of things create. There will always be a reason to innovate and explore new possibilities. It’s going to be interesting, regardless. We'll see what the future holds for all of this. This is the WP Minute. Visit thewpminute.com/subscribe. Still not powered by AI yet. Yet.

Maybe someday. Hit that newsletter. I think I mentioned this to you a couple of weeks ago; I switched over to ConvertKit.

Is that going to be the kit? Has it already arrived? Excuse me. Sorry, it’s just the kit now.

That's right. It's just Kit. Kit.com. One, because this is a semi-tangent to what we're talking about today, but everything that happened with WordPress—I’ve always had this feeling anyway. However, everything, especially with WordPress, has highlighted the importance of understanding who owns the business you’re doing business with. And yes, I was using Mailchimp for a while.

It was fine. I never really loved making the newsletters there. But I switched to Kit because I had Nathan on the show many, many years ago. He'll DM me every once in a while to ask a couple of questions, and I was like, "Yeah, you know, I'll use it."

We got a new add-on for it for Gravity Forms, and so far, so good. I consolidated it to Mailchimp. The accounts setup was a little bit of a pain because it flagged me with questions about where I got these users from. I explained that I merged two different lists: my MAT report list and my WP Minute list, which I was always mailing separately. So, I just consolidated them into one, but it flagged a bunch of stuff. However, crafting the templates there is much easier, and I'm happier about that. If you do receive the email and have any feedback, please hit reply on the next email I send and let me know what you think. Mark, what about you?

What are you up to? Where can people find you? I'm just trying to double, triple, and quadruple down on content on YouTube. You can visit mjs.bio to find out more and sign up for my newsletter.

All my social links and everything like that are there. I'm just trying to provide as much value as possible to you all. I really appreciate it. Awesome stuff! Thanks for watching, everybody.

See you, or you'll hear us in the next episode.