You're not ready for the future of coding

So, every time I make a video about agent coding, people seem to think it's just like, you know, not that big of a deal. But what you're looking at right

deal. But what you're looking at right now in the screenshot, this is actually like the future in my opinion of agent coding. This guy has 24 claw code

coding. This guy has 24 claw code instances running all in separate terminals, modifying a codebase, and he actually has it each one working on different GitHub issues at the same

time, concurrently, making pull requests. And then he goes through and

requests. And then he goes through and he reviews the pull requests to figure out which ones he should actually merge in. Then he also has an agent that goes

in. Then he also has an agent that goes and reviews and figures out which pull requests he should merge in first to help, you know, prevent as many merge conflicts as possible. Now, honestly,

something like this probably wouldn't work on a smaller codebase, but as your codebase gets larger and you have a good way to self-contain and isolate tasks or features, you can actually kick off

these agents to work on a uh a feature in isolation so that you don't have a ton of merge conflicts. Now, if you're not familiar with using cloud code or cursor or like codecs, you may think this is a joke, but I have actually had

like five or six different claes running when I'm building out this game and they will all work on various things. So, for

example, I just added a battle royale mode to my game. You can see here I have 74 file changes and it went through all of my codebase and basically added in a different game mode. Okay, so to demo that, we are in the current mode of

waves. And if I just say slash mode

waves. And if I just say slash mode royale, it switches us to a battle royale mode where I can go around and kill all the other players. And when the players die, it basically restarts back

in the royale mode. So this was like a couple of prompts in claw code. It

basically went through and created an interface and a strategy pattern so I can switch out my game modes whenever I want. And then I had to go through and

want. And then I had to go through and like fix a couple things. I had to open up some other claw code terminals as I'm playing through. I noticed like a little

playing through. I noticed like a little bug with the toxic gas clouds that are forming. I noticed an issue where like

forming. I noticed an issue where like certain traps weren't hurting certain players. And if you have enough features

players. And if you have enough features in your game that are like selfisolated, you can literally do a bunch of concurrent agentic coding on your codebase and become extremely efficient.

Now, I will say that I probably want to use this many terminals. That's like a little bit overwhelming for me. I like

to try to find the balance where I can still stay in the zone and review the code as it's being written and make sure that overall the architecture of, you know, the game modes that was added in seems like it's pretty good. And then

also, these models aren't perfect. So

even though it seems like it's working pretty well, you load up the game and then like stuff is just broken. But I

honestly think this is true. I don't

think this is a joke. I think this is real how people are agent coding. And I

do think if you want to learn more about agent coding, I do have an agentic jumpstart.com website you can go to and join the weight list. I have over I think 700 people join the weight list already where I am working on a course

to kind of teach you everything I've learned about claude code and cursor and agent coding to help build out not only the survive the nink game, but I have a bunch of other things that I've been doing agent coding with along the way.

Now, I will say I'm not going to go in depth of like this approach. This is

just too much in my opinion. But I just wanted to point it out because like this is like legit where I think software engineering is going. So, he kind of gives a breakdown what he did here. So,

he basically ran an initial agent to scan his repo looking for general improvements. It flagged 20 things. I

improvements. It flagged 20 things. I

like 12 of them and I told it to create GitHub issues for each. And then he opened up 12 T-Mux PES and told each one to fix the GitHub issue. And this is something I've even done on my own projects where I basically just give it

a GitHub MCP. It's able to create issues. is able to go and pull the issue

issues. is able to go and pull the issue and read it, add it to the context and actually start fixing it. And if you use GitHub work trees, you can actually kick off multiple GitHub work trees to work on these concurrently without having like a huge pull request like I have

here with 74 file changes. Let me go and add this 75 now. And then it looks like he runs like a review agent to basically review it, a respond agent, a summarized PR agent, and then he manually goes through and figures out what PRs to

merge. Now, this is one workflow. It

merge. Now, this is one workflow. It

probably works for him. I don't think I would do it on my projects. I basically

just add in a bunch of features and then I keep on doing a getit add on them and then I commit them once I feel like, you know, it's it's all working. But it's

funny, every time I make a video about agent decoding and how like this is the future. And I've been saying this like

future. And I've been saying this like from the get-go, right? When cursor came out, I've been making like posts on X and I've been making videos about how like this is basically changing how we work. And if you're not like jumping on

work. And if you're not like jumping on the bandwagon and learning how to really utilize these tools, you're just going to kind of fall behind. And this isn't even FOMO. Like the amount of times I'm

even FOMO. Like the amount of times I'm able to like prompt something and having a completely new feature added to my game, it's it's basically crazy. And I

will say that I've been kind of like on the fence about AI. Like I have this video, is vibe coding even sustainable?

Where I talked about the cost of vibe coding, but there's new models coming out and like there's so much competition with these LLMs like Gemini, you have GPT5, Codeex, you have you got Composer

1, you have Claude Opus 4.5. And I think as these models get better, they're becoming faster and they're becoming a little bit cheaper because there's so much competition to get you to use your their models that they're like lowering the prices on all these models. And so

in this video, I talked about how like, okay, I think these models are going to cost more. But I think I'm actually

cost more. But I think I'm actually wrong about that. I think as time goes on, we're going to find more ways to build better GPUs to make more efficient LLM inferences, and the competition is just going to keep on driving the price down because that's kind of how

capitalism kind of works, right? And at

this point, I would say vibe coding is is pretty sustainable. I still have to go through the code and make sure it's like written properly because I do notice that it does add some bad code sometimes. But again, it's just kicking

sometimes. But again, it's just kicking off a couple more prompts to refactor magic numbers out to a constants file or to redesign how something's done. And

it's just a matter of like, you know, prompting clawed code three or four more times. And again, for this game alone, I

times. And again, for this game alone, I would say that probably 95% of all the code was written by an agent. Now, I

will say I try to review the code and I understand a majority of my code base.

So, if you ask me like, "Okay, Cody, can you explain how the networking works?"

Well, I can go to the uh the server socket manager and that's where we're basically, you know, accepting the sockets using u web soockets. And then

if you want to figure out how do the entities serialize over to the back end, I could go find like you know the server entity over here and we have a serialize method uh serialize to buffer. Okay, so

this basically runs to the entities and serializes all the data to a buffer and sends it to the client. And then there's an entity on the client that's going to deserialize this. So let me find the des

deserialize this. So let me find the des serialized method. D serialize.

serialized method. D serialize.

Okay, here's a des serialized method where we basically take in that bite buffer and we dynamically figure out how to like take the values out and put them back into the entities that might be using it. Now, you probably don't care

using it. Now, you probably don't care about this code, but the point I'm trying to tell you is that like the way I do agent decoding is I try to review all the code and make sure I understand what's going on so that later on I'm more efficient at prompting and I can

just tag context specifically when I need something changed. because if you don't, you're just basically leaving up to the LLM, to the GP tools, to the find commands that it's going to run to try to find the things that are relevant.

But sometimes, you know, exactly where you want to refactor code. Like let's

say I want to change how the map is working. I can just go ahead and tag the

working. I can just go ahead and tag the map manager down here and I can prompt it and say, "Hey, I need to refactor how you're starting up the map. Where are

you spawning the zombies? How many

zombies are you spawning? Maybe I need to go to the game master file so I can change where uh, you know, the the distribution of zombies, how those are all spawned up." Okay. Now, the reason I'm showing you this is because a lot of

people leave dumb comments on my channel saying that like, you know, you're just a vibe coder. You don't know what's going on. I literally know most of this

going on. I literally know most of this code because I review it. I understand

it. And a lot of the code is not really important to review, but there are core concepts in your codebase and core patterns and core, you know, modules that you do need to become familiar with and how they all kind of interact with

other parts of your codebase so that you can have a good holistic idea of how the code is evolving. Okay. And so that's kind of what I teach in my agentic jump start is review the code, understand how it's evolving, not just becoming a vibe

coder, but actually becoming a engineer who can leverage these tools and become as effic as efficient as possible. So I

guess when I was scrolling through this and I saw this post, like it just made me want to make this video because I just want to make sure that people who watch my content, who have been following me, I'm sharing this stuff because I do think this is the future.

Even though I get like angry comments and down votes on my videos anytime I do like AI related content, I just don't really care because I do think this is the future. And I do think that if you

the future. And I do think that if you are following me and I have helped you out along the way, this is the next progression where you should probably have some type of subscription to one of these LLM models and you should be using

them and really trying to understand how to utilize them in your day-to-day works or your side project to see how fast you can actually start shipping code. I've

been watching another YouTuber, go check him out, Bridgemine, he's basically been streaming every day vibe coding on a project. He's on day like 85 right now

project. He's on day like 85 right now and he also does something similar like I've seen him basically have like I don't know about 24 terminals but he's had like eight or nine terminals opened up. I'm sure I could find it in one of

up. I'm sure I could find it in one of his streams. Yeah. So I mean this stream alone he has 1 2 3 4 five different terminals and he has cursor open. And so

I'm trying to let you guys know that like this is not a joke. This is legit how people are coding and he's a vibe coder so he's not really reviewing the code either. And he's able to ship a

code either. And he's able to ship a bunch of features. I've been kind of watching his journey along the way and he's able to create quite a lot for someone who's not even like reviewing the code mainly. And this guy's also one of the reasons why I went back and

subscribed to Claude code because I basically watched him in a live stream one prompt a solution to add auto captions to a video which if you guys remember I was working on the video crafter a while back. Let me see if I

can find that the video crafter. I don't know if you people even

crafter. I don't know if you people even Yeah. Yeah. So, if you guys remember, I

Yeah. Yeah. So, if you guys remember, I was working on this uh side project where basically you can create AI videos and you can generate different images for every segment of the video and you can create the prompting and you can pass that to 11 Labs to create the text,

the voice, and then I have a process that basically strings it all together into a video and adds auto captions.

Now, the auto captioning part took me quite a while. This is back like in September of last year, so the models were not as powerful as they are right now. And it took me like multiple days

now. And it took me like multiple days just to build out an auto captioning system with FFmpeg. So, like here's an example of what my SAS product basically built out.

>> So, it basically creates a video of it scaling in the images and then adding the auto captions and stuff like that.

But the point is that the auto captioning thing took a decent amount of time to build out. And now me watching uh Bridgemine, a vibe coder who basically one prompted the same thing I

had to spend multiple days working on using Opus 4.5. Like people watch my content and they're like disappointed I'm doing AI stuff, but I feel like people are just a little dense that

they're not seeing how powerful this stuff is. And I need you guys to wake

stuff is. And I need you guys to wake up. Okay, that's all I'm trying to say.

up. Okay, that's all I'm trying to say.

The people have tried this stuff, they're blown away. The people who are kind of resistant to it, I don't know why you guys are so resistant to it because this is crazy stuff. Anyway, I

just want to make a quick video to kind of discuss this because I think this is crazy stuff and this is just kind of the future, right? So, if you want to kind

future, right? So, if you want to kind of join us all in the future, go check out myicjumpstart.com course. I'm going to hopefully finish it

course. I'm going to hopefully finish it by the end of the year. I have a bunch of videos teaching you how to set up cloud code and cursor, but more specifically, I'm teaching you how to prompt, how to think like an agenda coder, how to use certain models for

different scenarios like fixing bugs or adding new features or it's going to be good. So, go ahead and join the waiting

good. So, go ahead and join the waiting list and I will talk to you later. Have

a great day and happy coding.