From "Stop Pretending You Understand Your Codebase"

"I got a hot take for you guys. If you truly fully understand all of the code in the projects that you're working on, you're probably not working on things that are that important or serious..."

"...frameworks and libraries we use, things like Postgres itself. Nobody understands all of the code in these things. And I feel like we pretend that's not the case. As an industry, we've always liked to think of the best programmers as these super geniuses that know exactly where everything is in the codebase always. And those people feel like they exist, but I'm going to be real with you, they don't."

Yeah, right. Maybe nobody knows all the code at once, but that doesn't mean they don't understand it. Or, as Theo insinuates, that the good engineers just have the intuition for where the architectural parts should click.

After his several-year-long streak of going through JS and TS frameworks, while promising the "next big thing" - sponsored at the time by Vercel or something similar - Theo slowly but surely reached his final form - now without the mustache and blond hair from the framework era (btw, I liked that aesthetic, and maybe I'm just pissed off that cool look is lost), and as always, using the standard "surprised" face in his thumbnail.

He has an "interesting theory" this time, masked as a "hot take" at the beginning of the video: You really don't need to understand all of your code. Don't waste your time trying to understand all of it. Theo insinuates that maybe it's enough if you have some notions, and that it depends on what you're doing. You know, maybe it's enough if you know stuff like "+ adds", "print goes brrrr", and "computer computes". I don't know, I'm just guessing where we can stop understanding.

But then - why bother understanding anything at all? I mean, he doesn't say it yet, but why not? Live a happy life and just enjoy the magic slop machine. It seems like a logical next step. This week we are finally "in the AGI era", as OpenAI declared, and Jensen confirmed, with no vested interest, of course. This time, yeah, this time for real, it's AGI until next week, when it sucks because there is a new "AGI" model. Yeah - this time for real; loop while false. And Theo, once again, serendipitously has a "hot take" after his "this feels like a taste of AGI" thumbnail.

The intro is 100% Theo. The video itself is not so hardcore (as usual). He kinda shields himself by doing a read-along of a Goedecke post, dulling everything step by step until his "hot take" is almost diluted. It's not about that - I mean, the guy is always like that. He could reuse the first thumbnail with "AGI". He has several of those going back to when OpenAI released o3. If you check the thumbnails, Theo has already been living in an AGI-verse for years, through different degrees of "vibe coding", but now, with his mega hot take and a few small brushstrokes, we are reaching a new cusp: he is explaining to us how to live - how we will "code" more and be more productive - perhaps happier too, I suppose - by just having an LLM generate a ton of whatever.

I usually don't mind this stuff - Theo is trying to get all the views he can and all the best sponsors. But now, he introduces the idea that if you're working on something really important and interesting, you can stop pretending that you know things. Embrace the ignorance, relax. Just sing your ideas to your favorite slop generator, and it will barf the best shit you've ever seen - well, you haven't really seen it. Theo, for example, says he doesn't need to understand the 60k lines of Swift code it generated.

So, wait... By that logic, his Swift code must then be very interesting and serious, and Theo therefore had to ignore his own ignorance? What's next? Because if 60k lines of code are already "too interesting and too serious" to understand, where is the limit? Who gets to establish these rules? If an unfriendly auditor turns your app into a beehive of exploits, infecting users, what would you say with that surprised-thumbnail face? "An AI made this, so..."

I don't care if you use AI to program, btw. Programming is not a cult. You can program using whatever language, tool, or editor you choose, even though there are a lot of talibans for certain languages. Maybe you should use AI to program. Maybe you will make wonderful things or something useful that you'll later have to rewrite manually to make sure that the next, duller model won't break everything. Maybe it will just turn you into a 2026 Theo - a guy who intuitively organizes his stuff without caring about the code. I just don't recommend generating code, at least not as your only source for learning programming.

Dudes and dudettes working on Postgres are a README file or a pgsql-hackers thread away from understanding what this or that part does at any given moment. The reasoning is there - it wasn't some free-range prompt or a random-seed decision statistically deduced at a given moment. It's not a slop code that nobody understands.
Just go to the repo and learn from it, or even better - download it while it still exists (Tinfoil alert! Tinfoil alert mf!) https://github.com/postgres/postgres

Either way, another fallacy is that if a codebase is not a behemoth, it is uninteresting and unimportant. So what about curl? Or FFmpeg? SQLite anyone? zlib? You realize the entire world runs on these things, don't you? Or maybe your idea, as I've been seeing lately in some circles, is that an LLM should reinvent the wheel each time you could use a battle-tested library.

Used carefully, AI can maybe help "augment" your knowledge. Not faster- knowledge acquisition depends on how well-trained your brain is. Put the rep(os) in. But going a step further and saying that nobody understands the entire codebase is a low bar, even for clickbait.