"I've never felt this much behind as a programmer." - Andrej Karpathy

So there's a pretty viral post on X from Andre Kaparthy where he talks about, "I've never felt this much behind as a programmer." If you don't know who Andre

programmer." If you don't know who Andre is, basically he's the one who helped contribute to OpenAI and build out all these LMS that we're using today. He's a

founding member who played a key role in developing the GPT models, right? And so

he has a lot of experience with AI and how to basically do all the machine learning and stuff. But I think it would be interesting to kind of talk about his post and I can give you my thoughts on where I think this industry is going.

Now, if you remember, this is the first guy who coined the term vibe coding, right? Okay. And so this is where the

right? Okay. And so this is where the vibe coding term comes from. It's from

this guy. He's been working with, you know, open AI and all this stuff. Let's

just go and read through this post. I

think it's kind of interesting. I have

never felt this much behind as a programmer. The profession is being

programmer. The profession is being dramatically refactored as the bits contributed by the programmers are increasingly sparse in in between. I

have a sense that I could be 10 times more powerful if I just properly string together what has become available over the last year. And a failure to claim the boost feels decidedly like a skill issue. There's new programmable layers

issue. There's new programmable layers of abstractions to master in addition to the usual layers below involving agents, sub aents, their prompts, contexts, memory modes permissions tools plugins skills hooks MCP

LSP/Comands, workflows, IDE integrations in a need to build an all-encompassing mental model for strengths and pits of fundamentally stoastic, failable, in unintelligible, and changing entities

suddenly intermingled with what used to be a good old-fashioned engineering.

Woo, that was a long sentence. Clearly,

some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it while the resulting magnitude 9 earthquake is rocking in the profession. Roll up your sleeves to not fall behind. Okay, so

he's basically talking about how all these models are coming out and understanding how to properly use them does take some time to ramp up, right?

There's a lot of tools that are involved with using like cloud code and hooking into MCP servers, how to integrate with various IDs that keep coming out every month and I think it's a really good topic to discuss. Now, this is a really

good post and I do want to hook into one more post that I found from him. Very

good questions. In my opinion, experienced devs have a real advantage, but only if they rapidly progress through the grief cycle and adapt now and onwards. Categorically, rejecting or

and onwards. Categorically, rejecting or ignoring the new layer would be a mistake. So, he's basically saying, hey,

mistake. So, he's basically saying, hey, uh, you know, every developer is going to go through this grief cycle. And I

honestly had this like a couple of months ago on my channel. I made a video how I'm just I'm tired of making content. I feel like AI is ruining the

content. I feel like AI is ruining the joy for programming. And I feel like that was my grief cycle, right? I

realized that these AIs can turn out more code than I can in like one minute from one prompt than I could in like 30 minutes, right? Building out React

minutes, right? Building out React components, building out whole API endpoints, building out all these various things. And AI can do it much

various things. And AI can do it much faster and often it can do it much better than you, right? If you hook in the right claw skills, you can have the output be very high quality like senior engineer level. If you do a security

engineer level. If you do a security review and a code review after you generate the initial code, you can even start iterating and make it better and better. And so I do think that there is

better. And so I do think that there is something that a lot of developers are going through right now. A lot of it's denial. People are denying that these

denial. People are denying that these models are even good. And when you start using Opus 4.5, which is I think the catalyst of a lot of the stuff, like a lot of people are waking up and starting to realize that Opus 4.5 is actually

really good. You can prompt it one

really good. You can prompt it one thing, you let it cook, you come back and it basically implemented all the features. And I've been using Opus 4.5

features. And I've been using Opus 4.5 since it came out. And honestly, like it's pretty amazing. I built out a full video editor right here, which uses AI to analyze where I have dead air in my my videos. And I can just go through

my videos. And I can just go through here and I can cut them out. I can edit all this stuff. I can export it. It's

using FFmpeg to export. And this was like two hours of work. I literally used Automaker. I let it run all night to

Automaker. I let it run all night to create the initial project. Then I came back through and I started sprucing it up manually with cursor and cloud code.

And so the future is literally AI. You

do need to know how to use these tools.

And I think getting a handle on how to utilize cloud skills and how to utilize these sub agents, how to utilize MCP servers is going to give you that uh advantage, that edge over other developers who still don't even use AI.

Now, I find it very interesting because I have literally been using AI since like Cursor Composer came out last year in 2024. And I was blown away even with

in 2024. And I was blown away even with the the terrible models back then. I was

watching it go through my codebase and modify all these files and sometimes it would actually get stuff working. Fast

forward to today, we have cloud code opus 4.5. You give it one prompt, it can

opus 4.5. You give it one prompt, it can sit there and modify 30, 40 files. You

load up your UI and it all just works and the feature has been implemented.

And so we've reached this tipping point, this threshold where these tools are so good you cannot ignore them. And the

developers who are still pushing back against AI tools and are just refusing to accept them and you know they're living in the old ways. I kind of feel like they're almost like dinosaurs and you have to kind of start waking up and at least playing around with these

tools. Don't be sleeping on these tools

tools. Don't be sleeping on these tools every day when a new model comes out.

Try it out. Test it out. Try prompting

it on your existing project. Try

building out a brand new project. Really

analyze these tools because I have met people who still don't use AI and they will spend hours trying to fix simple bugs that Opus or Gemini or, you know, GPT 5.2, which is also a really good

model, can just knock out in one prompt.

Now some of the things he mentioned here do not take that much time to learn but others you know they do take some time to ramp up like how do you properly configure and use sub agents how do you properly prompt how do you understand the context window like a memory layer

hooked into claude code like how do you actually do that type of stuff cla skills that's other thing that just came out recently which people are starting to use and I will say that all these tools I am covering in my agentic jump start course for example I have a whole

video in my course talking about claude skills I walk you through what claude skills are how to get them set up and how to set up a UIUX pro skill to make your user interfaces and your user experiences with what Claude is

generating that much better. So

definitely go check out Agentic Jumpstart if you're interested. I have a course with over 74 videos, 11 and 1/2 hours of content. I teach you how to use Cloud Code, how to use cursor, the basics of all this type of stuff, the

context window, how to do prompting, how to do automatic code reviews on your pore request using like code rabbit. And

at the very end, we build out a full stack web application using my workflows, deploying it out to railway.

So you can actually have one real life full stack application deployed out using my workflow. And then as time goes on, I'm throwing in bonus videos as I learn new things about AI. I will say this industry is changing so fast that you need to be on top of AI because you

can wake up one month and be completely out of the loop on how powerful a new model is that just dropped or how efficient or how fast a model is that basically for example composer one I use it all the time for refactoring code

because it's so fast and it works pretty well. So I highly recommend go check out

well. So I highly recommend go check out a gentic jumpstart.com if you want to be on top of all these changes that are happening in the industry which is exactly what Kaparthi is basically talking about. He feels like he's behind

talking about. He feels like he's behind as a developer, as a programmer. And I

would say that a lot of people are currently way behind the bell curve. The

people who are really embracing AI and who are on the tip of the title wave, these are the people who are going to be able to ship so much code extremely fast. And there's going to be cope

fast. And there's going to be cope that's going to come out that say, "Well, the AI generates insecure code."

Yeah, every once in a while it does. And

that's why you need to understand how to review the code AI is generating or how to hook in to all these other things so you can run them through a security review and then refactor the code if it finds any type of security

vulnerabilities. And you can hook this

vulnerabilities. And you can hook this up into a chain where you can have one agent invoke another agent that invokes another agent. Really clean up the code

another agent. Really clean up the code and make it as clean and performant as possible. Right? So, I guess my final

possible. Right? So, I guess my final thoughts, if the guy who made the LLM is feeling behind as a programmer with with how much all this stuff is changing, it's probably a good sign that you might feel that way as well. So, definitely

don't sleep on these things. Make sure

you're always playing around with these models. Test them out as they come out.

models. Test them out as they come out.

You want to be on top of, you know, these changes so that you're not kind of left behind as a developer and these other AI engineers are just running circles around you. All right, that's about it. Have a good day and happy

about it. Have a good day and happy coding.