AI Coding Is here to stay

The last couple of weeks, I've been thinking a lot about the future of software development, but also how to position ourselves as developers to kind of take full advantage of this AI boom, whether that's for shipping faster or building faster or learning faster. And the reality of the situation is AI is here to stay regardless of your personal opinion or my personal opinion. This is the future. And you know unfortunately when it comes to this topic people get overly emotional about this and they end up falling into two two groups. One of the groups is people saying AI generated code is completely trash. All the code ads I need to rewrite. It's harder to maintain. You know it just does damage to my code base. And then you have the other group that says lol. I vip coded this and made $1,000 off of it. Lol. Uh so the reality is somewhere in the middle. It's more of a gray area. But if there's one person to be feel like personally attacked by AI, it would be me. Because guess what? AI completely killed the niche I'm on here on YouTube, which is the uh web development tutorial videos, right? I do React tutorials, NO.JS tutorials. And AI completely wiped this out. Nobody nobody in 2026 watches these types of tutorials anymore. They barely get any views. Why? Because it's it's slow. Nobody's going to sit through a six-hour Python uh tutorial anymore because you can just do way faster with AI. And you know, there's literally two parts to this when you're a developer. You're going to spend time watching tutorials and learning the basics and then get to building. Right? When I got started in programming, I probably spent two years here just trying to absorb as much information as I could. And guess what? By the time I actually started building my first application, I I I totally forgot what I learned up here. So I had to pretty much restart and start from scratch. So I just ended up wasting time. But with AI now that you can start stray here now and building stuff out and kind of learn as you go through the whole process. So even like a week ago, I released a new project which is called Asky Jen. And this whole idea started with me just being really fascinated about ASKI animations. And I ended up making a post on on X about it, just kind of documenting kind of my what I was building out here. And people loved it. And guess what? People were asking for me to release it. So I made it and I managed to take the idea and put it out live in under two days thanks to AI. And honestly, I don't think I could have pulled this off without the help of Opus 4.6 six or codeex 5.3. Uh if I didn't have those tools, I think it would have taken me at least two to three weeks to get this live. But thankfully, I managed to ship it quick and it did really well. So, I really appreciate it. Uh also, for those who don't know, I'm really excited about this, so I'm going to plug it really quick. Ask Genen, if you want to add cool animations to your React applications like Asky Art like this, check out this tool. It gives you a live editor where you can upload a video and kind of make your own animation, adjust it. Uh, it's super cool. And we have a library here where you can just oneclick add these animations to your React project. And I also had a big update that was released recently, version 0.3. Uh, my development server is dead there. Uh, but we revamped the whole UI. It's really cool. Check it out at askjen.art. Okay, now just to go back to this tweet because I thought it was a little bit funny. H, this guy said, "Believe me, shipping code written by AI without a reviewing is catastrophic. Not because I hate AI, just because they can lead you to vulnerabilities, impossible to maintain, impossible to do this and that and that. Now, there is truth to this post. Absolutely. You should review the code the AI gives you, right? It's it's probably a good practice to do. But at the same time, you have Peter here, right, that shipped OpenClaw, right? That had like 200k stars on on GitHub now. and uh you know probably made a couple million dollars off of it. So who's really in the right here? Uh doesn't mean that this cannot be true where you can actually ship stuff that you don't even read, right? It's it's totally possible now especially with these models. U Opus 4.6 is actually shocking how many things just like one shots. It's just really really impressive. I think a lot of people are put off as well because they might have an experience where they tried Oh, I al also what I thought was funny is uh I just quickly checked this guy's profile out and not to hate or anything uh but I I opened it and I had a little chuckle because I thought it was like generated by GPT4. Uh but anyway, I think another thing is that your experience can vary so much depending when you use these tools. So if you're in the you're in the group of oh I tried this out three months ago and it was just not good enough for me and it just caused more problems. The thing is the evolution of these tools are so so quick that if you'd probably try it out now you'd have a totally different experience with it. So as a developer, I think it's a really important thing to kind of adapt to to new tech that's coming out on the horizon, right? Like for me example, right, with the YouTube, I could have stayed and made tutorial YouTube uh videos and that would have just not done great at all and I could have fought against it and be like, "No, I'll keep making these tutorials and it would have ended up nowhere." Well, same for this. It's new models coming out every time and things are improving all the time. So, uh, being in touch and in tune with this is really important. Now, a couple of mistakes I see a lot of people use. For example, here, let's say, um, I want to make changes to this, right? So, you do something like adjust the navbar to do whatever. Okay? And then you prompt, okay, I don't like that this gamma there is a bit too too close to the corner, right? And you keep doing these prompts, doing these prompts, but your context keeps growing. And then I see a lot of people after they do small adjustments and small bits like that, they do a like big plan. It's like, okay, let's rewrite this whole section now and maybe change out the core logic of how the ASKI animations get rendered out here. You shouldn't do that. After you fill up your context to like 100k or 200k tokens, you should just start a new session and just delete this. Start a new session and create a new plan. Okay? Because again the way these LLM models work right it's all like probability but if you have half of your context that's filled with irrelevant stuff um you know it's just better to start a new one off because the output can actually be um you know be be much worse than that. Another tip is when it comes to MCPs and skills just avoid that. Honestly, I never really had a good experience with those tools and I just let the raw LLM dictate the output for me. There's rare cases where I'll use an MCP maybe for a specific instance. Maybe we're like better off. For example, I'm doing stuff with off. Okay, I'll create a new session, make a plan if something's not working or I want to add a couple of features, load in all of that context, finish that job, and then kill off the session and make a new one. A lot of other annoyances that I hear people have is where the LLM does something that you don't want to do. Always just create a cloud MD or an agents MD and start it from scratch. Don't add anything here. And for each project that you do, if it does something you don't like, tell it to add it. So this is one of the like really important rules that I usually do is never run mpm run build or or dev or any of these tools. Let me run these. So as the project evolves, you can keep adding the rules like that. Don't get like a preset off the internet. I see a lot of people do this. It's just not going to work for your use case. So always start off from scratch and then kind of add on top of it as you go on and you'll find the the best results are like that. I think another important thing that you should do is actually do follow-up questions after the model implements something for you. Ask it how it did it and have it break it down step by step. This is such a good way to learn new stuff and I've done this you know without AI I probably would have not dabbled into so many aspects of programming. I've done some swift UI. I've done some electron development as well and learned so much about those technologies just by asking the AI model questions even like I have another project here that's been delayed now because it was originally intended to be just like long format courses but again I think that those will just not do well anymore. So, I delayed this project. It's not out yet. But we ended up taking the AI route of actually getting you to write code and and becoming a better programmer, right? We want to skip this step of watching stuff up here. I deleted it, but so when you go to learn JavaScript now, you're straight up into okay, let's see how you can do a URL query builder. And when you click on here, you're given the problem. You can start writing the code. And guess what? If you have any problems, Opus is going to do a fantastic job here answering your question. is probably going to do a better job than 80% of the YouTube tutorials on here. Um, you know, even my tutori I'm not saying that 80% like like just a bunch of tutorials here on YouTube just because there's information as well. You got to realize this because a lot of people say, "Okay, well AI sometimes doesn't get everything right." Well, guess what? There's a lot of bad tutorials out there online as well, including mine. But generally speaking, uh, for example, for this project that I'm building here, the AI model here, OPUS 4.6 already has context of how to build this problem out because I gave it all the information it needs and all the system prompts. So, it's it's probably going to do a better job explaining it than me personally. So, I think it's just a more intuitive way to learn than just watching long boring uh tutorials. So there's a lot of other things uh you know you can do to kind of make your experience a bit better. The way I like to also work is is having T-Mox open and then having two open code instances. Now I think if you have multiple agents it can be quite hard to keep track of everything. So I tend to keep it at two as like the maximum. I don't really go any further than that, but having two uh you can create a plan here for like a huge project and then uh for the second one here, I usually tend to do small little bits whilst the other one's loading. Yeah, that's kind of why I wanted to share with you here. And you know, I'm not the smartest developer, but I know a lot of people, you know, that work at Google and other like huge companies that are senior developers there and they're full in on using Agentic coding. They barely even touch the code anymore. So, if it works out for them, who am I to say? But I I just use it as I think it's such a powerful learning tool and even to bounce ideas around. I end up like going on walks and just putting up voice mode and just bouncing ideas. It's it's really fantastic. So, let me know what you think. Leave it down in the comments below. And yeah, I'll catch you in the next one. Peace.