AI News

When AI starts writing code, what kind of programmers are still needed?

ScienceNetUpdated Nov 16 19:32

AI summary

A BairesDev survey of 705 developers shows that AI saves them an average of 13 hours of programming time per week, nearly double that of a year ago; 42% of respondents say AI already participates in writing at least half of the code, compared with only 12% a year ago. But 67% of respondents say the time spent reviewing AI-generated code has instead increased, and nearly 60% need to invest more effort in identifying hard-to-spot errors. Lightcast data shows that the mention rate of Python in software engineer job postings rose from 14% in 2015 to 41% in 2026.

Body · AI translation

Home | News | Blog | Academicians | Talent | Conferences | Funds & Projects | Papers | Illustrations | Video & Live | Xiaoke Robot | Medical Science Popularization

Author: Kong Jiaxing, Han Yutong Source: The Paper Published: 2026/10/24 10:31:38

In 2026, technology services company BairesDev surveyed 705 software developers and found that AI saves them an average of 13 hours of programming time per week, nearly double that of a year ago. But at the same time, 67% of respondents said the time spent reviewing AI-generated code has instead increased.

October 24 is Programmers' Day. As one of the first professions to use AI on a large scale to transform workflows, the changes programmers are undergoing may allow us to see in advance how AI will change more people's work.

The BairesDev survey shows that 42% of responding developers said AI is already involved in writing at least half of the code, up from only 12% a year ago. Now, only 21% of respondents still spend more than half of their weekly working time writing new code.

Writing code is no longer the programmer's main time expenditure. 67% of respondents said the time spent reviewing AI-generated code has increased; nearly 60% of respondents need to invest more effort in identifying hard-to-spot errors in AI code; and 58% of respondents said they bear more responsibility for code not entirely written by themselves.

Reviewing code is only part of it. More than half of responding developers also said pressure to deliver work quickly has increased. When AI can complete development tasks faster, companies' expectations for delivery speed may also rise accordingly. But delivering faster does not mean less responsibility. Programmers still must ensure that AI-generated code runs properly and does not affect other functions.

More intriguingly, after programmers save time with AI, they do not keep themselves idle either. About two-thirds of developers save at least 6 hours of programming time per week through AI. But at the same time, they also need to constantly learn new AI tools and technologies, with nearly 30% of respondents investing more than 10 hours per week in this regard.

Zhang Liang (pseudonym), an engineer at a technology startup in Shanghai, said in an interview with The Paper that programmers are now increasingly like the "gatekeepers" of AI code. He believes that the code written by AI is already quite good, and people may not necessarily be able to spot many problems. The reason human review is still required is more about process, rather than AI's ability to write code being insufficient.

When writing code is no longer scarce, what abilities become more valuable?

The threshold for writing code has become lower, but that does not mean programming languages are no longer important. Online recruitment data platform Lightcast sorted through the 20 skills that most frequently appeared in software engineer recruitment from 2015 to 2026. One notable change is that companies' demand for programming languages has diverged: the mention rate of Python rose from 14% to 41%, while the mention rates of some traditional development languages such as JavaScript declined.

At the same time, companies' demand for capabilities such as system scalability, application programming interfaces, artificial intelligence, and workflow management is increasing. Compared with specific technical tools, some cross-domain capabilities also show continued importance: for example, the mention rate of communication skills in software engineering job recruitment rose from 33% in 2015 to 42% in 2026.

Job structure also shows a similar trend. Engineer talent data tracked by SignalFire shows that over the past four years, the frequency of front-end engineer positions fell by nearly a quarter, and mobile engineers also decreased significantly; positions such as AI engineers, research engineers, and deployment implementation engineers rose significantly.

Under the requirements of reducing costs and increasing efficiency, programmers regardless of seniority need to adapt to the changes brought by AI. For experienced engineers, AI can amplify their abilities, allowing one person to complete some tasks that previously required collaboration among multiple people; while for programmers who originally specialized in a certain technical field, AI may also force them to step out of their familiar division of labor and take on more work they have not previously encountered.

In US software developer employment data, one noteworthy change is that the employment trends of young people and middle-aged people are diverging.

The employment data dashboard jointly established by the Stanford Digital Economy Lab and ADP Research shows an obvious age divergence. Using November 2022 as the base period, as of August 2026, the employment index for software developers aged 22 to 25 fell by nearly 20%; in contrast, the 41 to 49 age group grew by nearly 20%.

This set of data comes from a sample of US companies using ADP payroll services. It does not represent the entire US labor market, nor can employment changes be simply attributed entirely to AI. But the career divergence of developers with different levels of seniority is still worth attention.

The risk of a talent gap facing programmers today may also appear in more industries such as media, law, and finance in the future. Young reporters learn judgment by organizing materials and verifying information, and junior lawyers become familiar with the business by searching cases and organizing documents. This seemingly repetitive labor is both work and professional training. If AI continuously takes over junior tasks, companies may be able to improve efficiency and reduce hiring needs in the short term, but may also cause newcomers to lose opportunities to enter the industry, trial and error, and accumulate experience.

In September 2026, an article published by the American Bar Association proposed that rather than retaining inefficient basic work, it is better to redesign the way newcomers are trained. For example, let young lawyers participate earlier in client communication, witness interviews, and case strategy discussions, and learn professional judgment under the guidance of senior lawyers, rather than only completing document organization and drafting.

These issues have also begun to enter the public policy field of vision. On October 23, the General Office of the State Council issued the "Opinions on Adapting to the Development of Artificial Intelligence Technology and Helping to Build an Employment-Friendly Development Model." It pointed out that the employment substitution impact of AI applications should be effectively mitigated, and efforts should be made to achieve inclusive development in which technological progress, industrial upgrading, employment promotion, and livelihood improvement are coordinated.

In this article, Zhang Liang is a pseudonym Special statement: This article is reprinted solely for the purpose of disseminating information and does not mean that it represents the views of this website or confirms the authenticity of its content; if other media, websites, or individuals reprint and use it from this website, they must retain the "source" indicated by this website and bear their own legal liability for copyright and other matters; if the author does not wish for the article to be reprinted or wants to contact us regarding reprint fees and other matters, please contact us.

2

BioMedInformatics: The "last mile" of brain imaging AI: How to make deep learning segmentation and classification more robust and more general?

FCS Nanjing University and Tencent AI Lab team launch DivHF, automatically constructing behavior descriptors through human feedback to solve difficult problems in complex scenarios

5

Ran Yang/F. Chiavaioli/Guan Baiou et al. Light | Multifunctional integrated all-in-one optical fiber diagnosis and treatment technology

6

Peng Dengfeng/Ju Minggang/Gan Jiulin/Wang Feng et al. Light | Self-recovering mechanoluminescent simple oxide: AI2O3:Cr

7

The 2026 Nanjing offline academic editor exchange meeting of Sustainability journal was successfully held

8

Symmetry: Interpretation of Nobel laureate Kenso Soai's article: The role of asymmetric autocatalysis in explaining the origin of homochirality in organic compounds

Cover article | Mycology: Unlocking the hidden "colors" of fungi

Source

ScienceNet