It's Not You Claude It's Me

Added: 2026-05-05

[00:00:00]I'm done.

[00:00:01]I'm going on sabbatical.

[00:00:03]I'm taking a break.

[00:00:05]I am just done vibe coding, agentic engineering, whatever people want to call it nowadays. I'm just taking a break from it.

[00:00:15]And the reason is I am just not learning enough. I am just not stimulating my brain enough. I am too abstracted from the code.

[00:00:26]The ease of using vibe coding and voice to prompt and all these different things unfortunately really just bring you farther and farther away from the code.

[00:00:38]Now, I remember when I was working at bigger tech companies not so long ago, our review process for any PR I would raise was as follows. We had two required PR reviewers on the team and trust me, we were messaging every single day and pushing you to review code. A lot of what you would do in a day would be actually reviewing developers code and you would sign off on it. And you would go line by line from these PRs.

[00:01:07]Your PR's too big, guess what? Break it up into two PRs. Your PR isn't using the right import order. Your PR is not syntactically correct. Your PR is using a less optimized approach to finding some type of algorithm. Your PR has a bunch of nitpicks on it. All these would go into the reviews and it would not get merged. You would not be able to put anything into production until or not even production into UAT until you had this sign off. And this would take sometimes just as long as the development.

[00:01:43]Yes, you heard me right. The review process of code would take just as long as it would take you to develop the project or develop the feature I mean.

[00:01:53]And some people might be asking, "Aren't you slowing everything down? Isn't this coming to a grinding halt? What can we do here? You're really just hindering the progress of our application. Our dev velocity is so slow."

[00:02:05]Yeah, we were working with a enterprise level application that was making millions of dollars in revenue a year.

[00:02:12]Do you think they wanted to just generate code fast and throw it up there for their thousands of users to use this application? No.

[00:02:20]It doesn't matter if they were able to fix a feature fast, this review process was robust and in place to make sure that the proper checks were there to make sure that what was delivered in production worked.

[00:02:36]What was written in the code was the correct way to solve the issue at hand or build the feature that was requested by the product owners.

[00:02:49]This is not that long ago.

[00:02:52]We have now devolved into this world where we are mass generating code, zero reviews are happening and my favorite is, "Oh, have AI review it."

[00:03:04]Okay, let's spend money and have AI go through your 90 file PR and scrutinize it, check it off. I've used these tools before, they're not good. They complain about things that really don't even matter, things that have nothing to do with the functionality of this PR. So, no, they don't work great. And then what are you going to do? What is going to happen at the end of this?

[00:03:26]You are going to not be exposed to any of the code.

[00:03:33]You're not going to really even understand the architecture, the state manager or anything going into your application that's happening. What is actually going on?

[00:03:41]You'll have no idea.

[00:03:43]And if you're building any type of application, imagine explaining to your investors or whoever is investing in your app in your app, you whoever's helping you fund this journey, imagine explaining to them, "Oh, I'm not really too sure what's going on here. I really don't know why we're using this function." Because trust me, when you get into these meetings and you're trying to either sell software or you're trying to get more funding into a project, this happens.

[00:04:09]And is there a world where we just say, "Oh, I don't know, AI wrote it."

[00:04:14]I doubt it. Money doesn't fly there.

[00:04:19]It really really doesn't.

[00:04:22]You can't get away with just saying, "Oh, AI wrote it. It should be good. It should have worked fine."

[00:04:27]And I hear this story, there's this one story I heard someone was telling me where this guy, and I believe this is probably hype, this is probably completely fake, but he basically vibe coded out an application um and he brought it to some tech invention and everyone thought it was a bomb or something. Again, this might not be completely accurate, but he basically vibe coded something and everyone was worried about what it was and so they had to bring in an expert.

[00:04:56]Probably knocked on the door of his uh secluded cabin and said, "Please." Cuz I believe it was written in um Was it Pearl? No, I think it was probably written in Rust, so it wasn't a cabin, it was just some uh high-rise apartment somewhere. They knock on the door, they say, "Please come. Please come and review this code. Make sure we don't know what this is doing. This could potentially be dangerous." He comes and reviews it and again, this is 100% written by AI and he says, "You know what? Not only is everything correct here, everything is as great and works as well as if a senior development team were to do this same development work."

[00:05:34]Don't believe the hype. Don't believe the propaganda.

[00:05:38]Don't become abstracted and it starts with yourself. It starts with me.

[00:05:45]I will only be using AI as a glorified Google search for the foreseeable future. I will be going back to line by line through code reviews. I'll be writing code.

[00:05:57]And yes, I will not be abstracted again because you can't afford to be abstracted out there. You can't afford to become less than you were as an engineer. You can't afford to become addicted. The subsidies are going to end.

[00:06:11]And when the subsidies end, the 10 to 100 X a month that is needed, there was a funny comment they said. I said, "Oh, the subsidies, you know, how are they going to make up for all these losses and all these different things?

[00:06:23]They're going to still need hundreds of billions of dollars in investment." And the guy just said, "Oh, they're going to print more money."

[00:06:28]Yeah. Let's go into deep deep inflation due to AI expansion. I mean, come on. At some point, the boomers, people who are throwing money into this are going to have to think, "Wow, I could burn millions and billions of dollars and really get nothing of actual use out of it, nothing that is truly transformative to the scale of my investment out of it." Yes, you can.

[00:06:55]So, as an engineer, I implore you to follow the same path as me. Take a break. Get back to writing it yourself.

[00:07:04]It is that much more frustrating to deal with AI generated errors, AI hallucinations, and things such as that rather than you taking the time to write the code yourself. And a lot of the time, you are faster.

[00:07:20]The rate limiting's happening. Even on my max plan, I'm getting rate limited to crap.

[00:07:25]Trust me. Don't become abstracted. Stay locked in. Get back to writing. Use AI for its best purpose as a super learning tool to help you and a glorified Google search that functions even better.

[00:07:39]If you have a fast computer, like an M5 or something, some beefy computer with a bunch of computer RAM, download um Ollama, get Gemma 4 on there. That thing is blazing fast. It is a joy to work with. I would really recommend that you check local models out if you haven't yet.

[00:08:00]All right. I will keep you guys updated on this journey, but we are going back to basics. No better way to say it. Take care.