A long post recently went viral on Juejin (掘金) with a thesis that grabbed us: the better AI gets at writing code, the more programmers should understand system design. After reading it, our editorial team thinks this deserves two minutes of attention from traditional industry managers and enterprise IT leaders too—because it's not just about programmers, but about how every "technical person" in the organization should reposition themselves going forward.
What this is
The author's core argument: programming has been "abstracting upward" for decades—from machine code and assembly, to C and Java, to Python and low-code. With each shift up the abstraction ladder, people worried that entry-level programmers would be wiped out, but the actual result was that humans weren't replaced; the center of gravity moved. AI coding is the next link in this abstraction chain: in the future, humans may no longer write source code, but instead describe goals, constraints, and context, and AI generates the executable program directly.
The key counter-intuitive point: the higher the abstraction level, the higher the comprehension requirement on humans—not lower. A bug in code might just mean one broken function. But if you get the top-level design wrong—system boundaries, module responsibilities, permission models—AI will generate an entire system along the wrong trajectory. What's truly dangerous has never been failing to write one line of code; it's failing to realize your system design is flawed.
The author therefore believes programmers' attention will shift from "how to write the code" to "why the system is designed this way"—how clear are the boundaries, how reasonable the responsibilities, how stable the interfaces, how simple the data flow, and whether failure scenarios have been considered. AI can explain code, break down modules, and fill in boilerplate logic—but it cannot judge whether a system is worth designing this way.
Industry view
Plenty of people back this read. A common line we hear: AI coding tools (GitHub Copilot, Cursor, and the various domestic code assistants) are turning "writing code" from a scarce skill into a baseline capability, and what will truly be scarce going forward is the ability to define problems, decompose systems, and own the outcome—commonly called "architecture capability" and "domain understanding." We've noticed that in recent job postings from major Chinese and international tech firms, the frequency of phrases like "system design," "cross-team collaboration," and "business modeling" has clearly risen.
But there's no shortage of pushback. One common objection: today's AI is still far from "directly generating executable programs"—it routinely makes mistakes even maintaining medium-sized codebases, so talking about "the compiler disappearing" is premature. A sharper critique: elevating programmer value to "system design" sounds lofty, but in most companies there are only a handful of actual system-design roles, and the rest of the programmers still have to write code, debug, and patch production environments—AI won't liberate everyone; it'll fatten the top and squeeze the middle. Others point out that AI-generated code still needs people who understand code to review and backstop it, so the notion of "not looking at the code" simply doesn't hold up in real engineering. The optimistic judgment in the Juejin piece, the critics argue, carries a heavy dose of survivor bias.
Impact on regular people
For enterprise IT: If a team really starts generating large volumes of code with AI, the focus of code review will shift from "is this line correct" to "is this design sound." That means the team's composition has to change—not as many people who only write code, but one or two who can actually decide where the system is heading.
For individual careers: The window where you can make a living purely on "fluency in a given language" may be narrowing, but the ability to clearly articulate why a system is designed a certain way is something AI won't replace in the short term. For non-programmers, this is actually an opening: going forward, working in product, business, or even management may no longer require knowing how to code—but it will require knowing how to describe a requirement precisely enough for AI to implement it correctly.
For the consumer market: All kinds of "AI auto-generates an App" and "one-click website builder" tools will keep getting more polished, but the quality gap of what comes out of them will get more pronounced—the output between someone who knows how to use them and someone who doesn't will differ by orders of magnitude. For ordinary users, the yardstick for judging whether an AI tool is reliable may no longer be "what can it do," but "can I understand what it's doing."