Back in 2022, I wrote about why reading code is a fundamental skill every developer must cultivate. The premise was simple: you spend 10 hours reading code for every 1 hour you write it. Master reading, and you accelerate everything else—understanding codebases, reviewing pull requests, learning from open source.
Three years later, that ratio has flipped in the most unexpected way.

My development process today looks radically different. I don't start by opening an editor and writing code. Instead, I think deeply about the problem, translate that mental model into a detailed specification, and hand it to AI. The AI generates an implementation—often hundreds of lines—in seconds.

Then comes the critical part: reading every line of that generated code. Understanding the approach. Verifying the logic. Reviewing the tests. Ensuring it aligns with our architecture, follows our conventions, and doesn't introduce subtle bugs that will wake me up at 2 AM.

The time I saved writing code didn't vanish. It shifted entirely to reading code.

AI was supposed to make coding easier. And it has—but not in the way most people expected. It didn't eliminate the need for deep technical skills. It amplified the importance of the one skill many developers undervalued: reading and understanding code written by others.

Think about it. When you write code yourself, you inherently understand it—you built the mental model as you typed. But when AI generates it, you're in the same position as joining a new codebase or reviewing a colleague's pull request. You must reconstruct someone else's thinking, verify correctness, and ensure maintainability. Except now, that "someone else" is a probabilistic model that can write brilliant code and nonsensical code with equal confidence.

The developers who thrive in this era aren't necessarily the fastest typists or the ones who know every framework. They're the ones who can read complex code quickly, spot architectural issues, understand subtle edge cases, and verify that what looks correct actually is correct.

Junior developers often ask me: "Should I learn to code if AI can write code for me?" I tell them: absolutely—but the skill isn't just writing syntax. It's developing the architectural thinking, pattern recognition, and debugging intuition that lets you evaluate whether code is good.

You can't review what you can't read. You can't maintain what you don't understand. And you certainly can't ship production code generated by AI without the ability to critically analyze every line.

The technology changed. The tools evolved. The workflows transformed. But the fundamental skill—reading code with comprehension, speed, and critical thinking—remains as essential as ever. Perhaps more so.

In 2022, I argued that reading code faster makes you a better developer. In 2025, I'm arguing that reading code carefully makes you a responsible one. AI didn't obsolete the skill; it changed the stakes. We're no longer just reading our teammates' code during reviews. We're reading the output of systems trained on billions of lines of code, capable of synthesizing solutions we might never have considered—and equally capable of confidently presenting broken implementations.

The 10:1 ratio hasn't changed. But now, at least one of those hours you're reading code written by something that's never been on-call for a production incident. Read carefully. Your future self, and your users, will thank you.

Because in the end, someone still needs to understand the code. And with AI in the loop, that someone is you.