AI can write code, what will newcomers use to level up?

Mon, 27 Apr 2026 23:18:36 +0800

In the last few months, while writing code using tools like Claude or Codex, my most striking realization wasn’t that “programmers are obsolete,” but rather that many tasks that used to be given to newcomers for practice can now generate a basic first draft themselves. Whether it’s creating a scaffold, adding several tests, or making small modifications on the fly—after running through these operations, the speed is genuinely fast, so fast it feels almost bittersweet.

For someone like me, who graduated ten years ago, frankly, this is more about increasing efficiency. Because I generally know where it’s reliable and where it isn’t; where something looks functional but actually has pitfalls hidden further down the line. But for fresh graduates, this topic isn’t so straightforward. AI isn’t just here to take over a few hours of manual labor; it feels more like it is compressing the traditional path of how a newcomer goes from zero knowledge to proficiency. This is also why I wanted to write about it separately.

What’s actually fading away isn’t the programmer, but mobile phone skill training

In the past, our team always had a batch of tasks that weren’t technically complex but were perfect for new hires. These included modifying pages, integrating interfaces, completing CRUD operations, fixing minor bugs, and debugging by following logs little by little. The work itself was small, but it was enough to take someone from merely “knowing how to write code” (syntax) to truly understanding “how the live system actually works.”

Currently, this type of work is the most susceptible to being taken over by AI.

The U.S. Bureau of Labor Statistics (BLS), in its updated Occupational Outlook for 2025, provided a very interesting comparison. On one hand, it states that roles such as software development and testing will continue to grow over the next decade; but on the other hand, it clearly indicates that the position of computer programmer—which is more focused on “writing and executing code”—is declining, as many repetitive programming tasks will continue to be automated. This change is critical.

It’s not that the software industry doesn’t need people; rather, the value derived solely from “taking tasks and writing code” is increasingly diminishing. Corporations, of course, are focused on cost efficiency: if AI can generate a first draft, followed by an experienced person to refine it, why do they still need to staff a slew of junior positions merely for slow mentorship, like in the past?

So, what might be truly frightening may not be layoffs themselves, but often the fact that companies are no longer replenishing staff or expanding roles. The door is still there, but the opening has become narrower.

Why Seasoned Professionals are Best Positioned to Capitalize on AI Dividends

It’s quite paradoxical. On the surface, code generated by AI might not be better than that written by fresh graduates. However, those who truly master using AI are often seasoned professionals who have learned through many pitfalls and failures.

The reason is also not complex.

First, seasoned professionals know that “looks runnable” is not the same as “truly deployable.” In Anthropic’s Economic Index from January 2026, when they analyzed software development in isolation, they found that although these requests are highly standardized, the task success rate was only about 61%. Furthermore, most scenarios still require back-and-forth iteration; it cannot simply be resolved by dumping the entire thing to AI once. METR conducted an even more sobering experiment in July 2025, allowing open-source maintainers who had been familiar with the repository for years to use cutting-edge AI tools of that time. The result was actually a 19% overall slowdown. This indicates that when the context is complex, quality standards are high, and boundaries are fuzzy, AI is not autonomous driving; it is more like a very talkative intern.

Secondly, experienced developers have something they call “code taste.” This concept might sound somewhat abstract, but it is very tangible. Whether an interface should be split this way, whether exceptions should be swallowed like this, whether testing is merely a trick to fool the CI, or if refactoring will preemptively bury pitfalls for the next week—these are things that matter. AI makes mistakes now, and many of those mistakes aren’t syntax errors; they are conceptual errors, abstraction errors, or boundary condition errors. Honestly, it is even harder for someone who hasn’t experienced traditional programming to identify these types of mistakes.

So, in the age of AI, what is most valuable is not typing speed, but critical thinking (or judgment).

Newcomers aren’t without paths, they are just out of the old ones.

I don’t believe that newcomers will be without opportunities. The World Economic Forum’s January 2025 report still ranks software and application developers among rapidly growing roles, and the U.S. Bureau of Labor Statistics’ long-term outlook for software development positions also shows growth. This suggests that the demand has not disappeared entirely.

But the entry method must have changed.

The previous default path was to start with some grunt or difficult work, learning as you went while coding, and relying on sheer time and effort to build your intuition. Now, companies are more likely to expect that upon joining, you already know two things.

One thing is to treat AI as a tool, not as an answer. You must be able to clearly articulate the problem, break down the requirements, and prompt it to give you usable drafts first.

Another thing is being able to review it effectively. It’s not enough to just say, “This is wrong”; rather, you need to know where the mistake lies, why it is wrong, and how to modify it so that it aligns with the project context.

This is troublesome. Because the ability to “review code” is something that usually takes years of working experience to develop, now that opportunities to practice are decreasing, yet newcomers are instead required to possess this skill much earlier. How should I put it? It’s like a game that dismantled the newbie village, but the boss is still waiting for you up ahead.

What Will It Be Like in the Future

My current judgment/assessment is relatively simple.

AI will continue to boost the output of experienced professionals while also continuing to diminish those parts of work in entry-level positions that are easiest to standardize and break down. These two things are likely to happen simultaneously, and they do not conflict. The ILO’s report from May 2025 was quite measured, suggesting that generative AI is more likely to change the structure of tasks rather than eliminating entire jobs all at once. The issue is that when the task structure changes, the first things to be eliminated are often those low-risk tasks originally designed for training new employees.

Therefore, the most scarce talent going forward is not merely someone who “codes best,” but rather someone who possesses foundational abilities, can harness AI, and takes ownership of both business outcomes and quality.

For seasoned professionals, this feels like adding a sharper shovel—it’s still taxing, but significantly more efficient. However, for beginners, the issue isn’t whether they should use AI; it’s that if they rely on it from the start, who will tell them when it writes well, and when it is producing convincingly plausible nonsense?

Neither the school nor AI can provide this answer right now. Ultimately, you’ll probably have to figure it out yourself. Ugh, the door isn’t closed, but getting through it is much harder.

References

Writing Notes

Original Prompts

Some thoughts on AI programming. As a recent graduate, my coding abilities are certainly not comparable to Claude or Codex. For experienced developers, AI can boost efficiency, but it has also led companies to significantly cut the positions of junior programmers. I graduated ten years ago and have gone through the process of starting as a beginner and reaching senior levels. What about those who enter the field now? How will they fare? The older generation lived through the days of “old-school” programming, so frankly speaking, they can recognize whether AI wrote good code or bad code, because current AIs still make mistakes.

Summary of Writing Ideas

Center the main argument on the idea that “AI is compressing the learning ground for newcomers, not shrinking the entire software industry.”
In the first half, first discuss why experienced workers benefit more from the AI dividends, and then explain how this can coexist with newcomer anxiety.
Use materials/data sets like BLS, WEF, Anthropic, METR, and ILO to firmly establish two points: “structural job changes” and “the necessity of human judgment in AI.”
Intentionally avoid elaborating on specific job search strategies or creating a ‘how newcomers should learn’ list of motivational advice.
The conclusion must return to the core reality: it is not that there are no opportunities, but rather that the traditional “newbie training ground” is disappearing.

Software Development on Uncle Xiang's Notebook