Thoughts on "Control the Ideas, Not the Code"

Jul 13, 2026

I’ve been trying to gain more clarity on the debate on whether we should be reading code or not. Antirez’s blog post is interesting. After reading it, I actually feel more puzzled than ever. The first point I’d like to explore is this quotation:

Then I compared the implementation, for correctness, to other systems, finding that other implementations sometimes contained more errors. I researched more, and found that the local inference world is full of subtle errors that accumulate and damage the model output, issues in the attention implementation causing performance slopes after the context is over a certain limit because indexed attention implementations are broken (do more work than they should, for instance), and so forth.

I agree 100% that AI helps a lot with that. But I feel like there’s something missing between “AI helps a lot with that” and “I believe reading code is mostly pointless”. I genuinely wonder how the above can be accomplished without reading any code. And I get more puzzled when Antirez says he actually reads code, even though it’s for different reasons. So, reading code is important, but not for the reasons we thought? And to be fair, it’s not like we have always had a deep reason on why we read code. We read (or used to) mostly because there’s no separation of writing and reading when we manually write (or used to) code. I say mostly because we’ve been copying/pasting code for quite some time now, and copying/pasting can be an example of writing and not reading. Now that we’ve decoupled reading from writing, it is fair to wonder: what role does reading play in software development? To add, in a PR review, we have code we didn’t write. How much of that code have we really been reading (over the years), and why?

The second point is what is truly meant by “control the ideas”. I see where he’s getting at, and I believe there’s some truth there. But I mean, how is a team of software developers (or idk product managers) supposed to control the ideas? How do they share understanding about the system, and who is the ultimate saying of what is going on in that system? We’ve developed a bunch of methods and practices for that over the years, and AI will help a lot there as well. But suddenly reading code does not have a role there? The code is itself the reality of how a system works. Saying reading code is pointless feels like discouraging inspecting reality with your own eyes.

I am not certain about how much the slider of “how much we should be reading” shifts towards “almost never”. Antirez claims that it is “almost never” but for some reason this blog post was not enough for me to really believe that. I would love to know more of his development process to see where the gap lies.