A set of counterintuitive data appears in the "Talent Status Report" recently released by the venture capital firm SignalFire: the overall hiring volume of large tech companies in 2025 is expected to decrease by 25% compared to 2019, but the proportion of engineers among all new hires has surged from 46% to 55%. NVIDIA CEO Jensen Huang also stated that after the adoption of intelligent AI by everyone, "software engineers are busier than ever" and that the idea of AI replacing engineers is "utter nonsense." In the current environment where the "AI replacement theory" is rampant, the robust resilience of engineering positions starkly contrasts with the drastic declines faced by other foundational roles. What exactly has happened behind this?
Shrinking HC of Big Tech and the Counter-Trend Expansion of Engineers
In the past two years, news of layoffs in the tech industry has been almost nonstop. From social media to financial news, AI has been frequently cited as a major reason for tech companies reducing their workforce. However, the tracking data from SignalFire, which includes 12 tech giants like Alphabet, Meta, Apple, Amazon, Microsoft, and NVIDIA, reveals a fact obscured by emotional narratives: these companies are not halting recruitment; they are undergoing extreme structural optimization.
The overall hiring volume of tech giants in 2025 is expected to drop by 25% compared to 2019, which signals macro-level downsizing. However, if we break down the hiring data by function, the differentiation is drastic. The hiring volume for engineering positions only fell by 11%, far below the average. More critically, the proportion of engineers among all new hires increased from 46% in 2019 to 55%. This means that for every two new hires, more than one is now an engineer.
In stark contrast to the resilience of engineering positions is the shrinkage of entry-level and non-technical roles. The SignalFire report noted that the proportion of new graduates being hired by Big Tech is only 7%, a drop of over 50% from pre-pandemic levels in 2019. At the same time, non-technical functions like recruiting, product development, and sales continue to shrink. Conversely, early-stage startups are showing a different extreme, having hired 7% more engineers compared to 2019.
This structural optimization's underlying logic is not complicated. Before the explosion of generative AI, tech companies maintained large personnel scales to support business expansion and functional iterations. Many entry-level engineers and execution staff handled basic development tasks, test case writing, and daily maintenance. As AI tools can now perform fundamental coding, copy generation, customer service responses, and even some sales lead filtering with exceptional efficiency, businesses no longer require large execution teams. On the contrary, to deeply integrate these AI capabilities into existing business lines and develop new AI-based products, there has been a surge in demand for experienced engineers who possess system architecture skills and AI tool usage experience. Asher Bantock, research director at SignalFire, stated that if AI were truly replacing engineers, engineering hiring should decline first, but the reality is quite the opposite.
For investors and industry observers, this data conveys a clear signal: the capital expenditures and human resource investments of tech giants are shifting from "horizontal expansion" to "vertical deepening." In the past, it was about piling up human labor to seize territory; now it’s about concentrating resources to build the underlying infrastructure and application ecosystem for the AI era. Engineers are no longer merely cost centers, but leverage points that determine whether a business can survive in the AI wave.
The More Efficient, the More Scarcity: The Verification of Jevons Paradox in the World of Code
Why has the demand for engineers not decreased despite the efficiency gains from AI in programming? SignalFire's analysis suggests the presence of a typical Jevons Paradox phenomenon.
19th-century economist William Stanley Jevons discovered while studying coal consumption that improvements in steam engine efficiency did not reduce coal usage—instead, because steam engines became more economically practical and were applied in broader scenarios, global coal consumption surged. This paradox has been perfectly validated in today’s coding world.
AI-assisted programming tools have made the cost and time of writing basic code nearly zero. Companies have not laid off programmers to reduce development volume; rather, "software has become extremely cheap," prompting them to integrate AI into all business lines. The demands that were previously shelved due to high development costs have been released in large numbers, leading to exponential growth in system complexity. The boundaries of engineers' work have been infinitely expanded.
In recent public statements, Jensen Huang detailed this change. He pointed out that after the company's personnel started using intelligent AI, AI is now "micromanaging" employees, and software engineers are "busier than ever." AI has not replaced engineers; rather, it has pushed them towards higher-level creative and architectural work. Engineers are no longer mere "code typists" but have transformed into "AI foremen" and system architects, having to handle more agent orchestration, system integration, and code review tasks.
In this process, a key technical concept is reshaping engineers' daily workflows: Harness. In the current AI engineering context, Harness typically refers to "toolchains, scaffolding, or orchestration frameworks." It is responsible for encapsulating, scheduling, and implementing the general capabilities of large models into specific business flows. As the capabilities of large models become homogenized, the main battlefield of competition has shifted to the Harness layer outside the models themselves. One of the core responsibilities of engineers is to build and maintain these Harness systems, ensuring that AI agents can operate stably and securely within corporate environments. This involves a large amount of interface integration, access control, exception handling, and context management tasks, significantly increasing the complexity of system architecture.
This transformation has been microscopically verified in the rising heat of tools within OmniTools and developer sentiment analysis. Between 2024 and 2025, views and favorites of AI programming and code assistant tools saw a steep upward trend. However, in the emotional distribution of user comments, the feedback is not overwhelmingly optimistic. Positive feedback centers on "basic CRUD code no longer needing to be handwritten" and "prototyping speed doubled." Negative feedback, on the other hand, points directly at the changing intensity of work.
In developer communities like Reddit and Hacker News, many experienced developers feel helpless about Jensen Huang's statement about "engineers being busier." They point out that while AI has saved time on writing basic code, it has brought about endless "mountains of code reviews." The code generated by AI often includes hallucinations and hidden logical flaws, requiring engineers to expend significant energy on complex prompt debugging and system-level bug checking. Specifically, when AI-generated code snippets are integrated into large projects, issues like context loss or type mismatches often arise. Engineers cannot blindly trust this code; they must review its logical correctness, security, and performance impacts line by line. The cognitive load of this review is far greater than writing code from scratch, as they need to understand the "black box" reasoning paths of AI.
Furthermore, debugging workflows for AI agents also represent an extremely mentally taxing task. A complex agent may involve multiple rounds of dialogue, tool calls, and external API interactions. When an agent exhibits abnormal behavior, engineers find it challenging to set breakpoints for stepwise execution like they would with traditional code. Instead, they need to analyze a massive amount of logs and intermediate states, and sometimes repeatedly modify system prompts to correct the agent's behavioral deviations. The nature of the work has shifted from repetitive "manual labor" to high-intensity "mental exhaustion." The demand for engineers has not decreased, but the cognitive load per unit time has significantly increased.
Some developers also express pessimism about the long-term validity of Jevons Paradox. They believe that when AI reaches the "singularity" of being able to write code, self-verify, self-debug, and understand global business logic, Jevons Paradox will fail, as AI itself becomes the "architect," ultimately decreasing the absolute need for human labor. However, at least by 2025, AI remains a lever that requires human engineers to operate, rather than an independent creator.
The Cliff Collapse of Basic White-Collar Positions: Who is Truly Being Replaced by AI?
The counter-trend expansion of engineering roles is just one side of the occupational shockwave from AI. On the other side, basic white-collar positions are facing substantial replacement.
Third-party data agency Bloomberry, based on tracking analysis of 5 million public posts on the Upwork platform via Revealera, revealed that within 15 months after the release of ChatGPT, the number of writing positions on the platform decreased by 33%, translation roles decreased by 19%, and customer service roles fell by 16%. In sharp contrast to these shrinking positions, backend development roles grew by 6%, frontend roles increased by 4%, while demand for chatbot development surged by 2000%. Although this third-party data analysis does not represent the entire job market, it accurately reflects the real changes in this most sensitive area of the freelance market.
Why are these three positions at the forefront? The core reason lies in the structural clarity of tasks. Writing, translation, and customer service involve highly rule-defined and repetitive core tasks. AI, when handling information conversion and rule-matching tasks of this nature, has extremely low marginal costs and quality has reached usable levels. Companies find that using AI to manage these basic tasks is much more cost-effective than hiring human workers.
This replacement is not only occurring on freelance platforms but is also quietly happening within companies. Anthropic CEO Dario Amodei has warned that AI could eliminate 50% of entry-level white-collar jobs within the next five years. This is not an alarmist statement but is based on predictions regarding current AI agent capabilities.
In specific business scenarios, AI agents are substantially taking over traditional execution teams. In the sales field, OmniTools previously analyzed the case of AI employee Viktor. This product, which has no human sales team, has secured 30,000 enterprise clients and generated $20 million through the AI agent. It has replaced traditional entry-level sales, SDRs (sales development representatives), and implementation teams. The core responsibilities of these positions involve following leads with established scripts, entering data, and answering basic questions—all areas where the AI agent excels. Viktor can handle massive volumes of leads around the clock and automatically adjust communication strategies based on customer feedback, making efficiency and consistency levels unattainable by human teams.
Similar erosion is happening in the administrative and HR sectors. The rise of office agents like Tencent's WorkBuddy has opened up cracks in the restructuring of basic support positions. The process-oriented tasks of initial resume screening by HR, schedule coordination in administration, and reimbursement approvals are being taken over by office agents. The roles are not being directly cut but are shrinking through natural attrition and "invisible compression." The workers in these roles are not explicitly told that "you are replaced by AI"; rather, their work content is being stripped away by agents, ultimately leading to the worthiness of the positions themselves vanishing. When an HR official can complete 80% of their work by sending a single command to WorkBuddy, the existence of that position now solely depends on whether the remaining 20% of non-standard tasks justifies keeping a full-time employee.
62% Premium and Elimination: Two Tracks Split by AI
The macro labor market has not collapsed due to the impact of AI; rather, it is presenting a "dual-track labor market" as defined by PwC in the "2026 Global AI Employment Barometer."
According to PwC data, the growth rate of jobs requiring AI skills is eight times that of the overall market (69% vs. 9%). More critically, the wage differentiation is significant, with a premium of up to 62% for AI skills. This is not merely a short-term dividend from a technical upgrade, but a reconstruction of the labor value assessment system.
On the first track are those who master the leverage of AI. They may be senior engineers skilled in using AI programming tools or business architects capable of designing complex agent workflows and harness systems. AI amplifies their output capacity, allowing one person to accomplish the work of several others, thereby enabling companies to pay a higher premium. When tech giants mentioned in the SignalFire report choose to reduce overall hiring to preserve engineering roles, they are essentially paying for this leveraging effect. The costs companies are incurring are no longer about hourly labor costs but are investments in system design capabilities and complex problem-solving skills.
On the second track are those engaged in "rule-defined" tasks that are easily automated by AI. Their jobs are being replaced by AI at very low marginal costs, and demand is falling off a cliff. The shrinkage of writing, translation, and customer service roles on platforms like Upwork is a true reflection of this track. When tasks can be clearly broken down into input, processing rules, and output, the irreplaceability of human labor approaches zero.
This is no longer a competition between humans and AI, but a competition between "those who use AI" and "those who do not use AI," a competition between high-leverage positions and low-leverage roles. The market is being split by AI, and the middle ground is disappearing. For industry observers, this indicates that future corporate organizational structures will trend towards a "dumbbell model": at one end are a few core architects and strategists, while at the other end are the AI systems themselves, with the large execution layer in the middle being significantly reduced.
Rejecting Chicken Soup: Career Deadlocks and High-Intensity Survival in the AI Era
Faced with this differentiation, vague "guides to surviving in the AI era" or "humans will eventually triumph over AI" sentiments are meaningless. We must confront the harshest realities in the current job market.
First is the career deadlock caused by the "experience paradox." The SignalFire report mentions a harsh reality: companies are all looking for high-level independent contributors who can produce output independently, and to save money and pursue efficiency, they are actually hiring senior engineers to fill entry-level roles or simply allowing AI to do basic work. This has led to a deadlock for new graduates and career changers: "without experience, you cannot find a job; without a job, you cannot gain experience."
This deadlock has a fatal impact on entry-level developers. In the past, entry-level developers became familiar with codebases and business logic by writing basic code and fixing simple bugs, following the traditional "apprenticeship" training model. Now, these foundational tasks have been taken over by AI, and entry-level developers have lost the soil in which to accumulate experience through practical work. Companies expect new employees to be able to conduct AI code reviews and agent orchestrations right away; however, these advanced skills require substantial foundational experience to support them. The drop of over 50% in the hiring ratio of new graduates is not just a numerical decline but represents a rupture in the conventional talent development chain. Breaking this deadlock is a systemic risk that the entire industry must address in the next five years. If a new mechanism for cultivating entry-level talent cannot be established, the industry will face a crisis of talent depletion at the grassroots level.
Secondly, engineers are being expanded but experiencing an unlimited increase in work intensity. The Jevons Paradox has ensured engineers keep their jobs, but it has not guaranteed their quality of life. Transitioning from "manual labor" to "mental exhaustion" is not a blessing. Reviewing AI hallucination code, debugging complex agent workflows, handling the exponentially growing system integration demands—these high cognitive load jobs are reshaping engineers' daily routines. Security in the AI era does not come from ease but from irreplaceable architectural capabilities and high-intensity human-machine collaboration. Engineers need to adapt to the shift from "creators" to "reviewers" and "orchestrators," requiring higher abstract thinking skills and a system-level perspective.
AI has not killed the engineer; rather, it has made them the most valued core asset for tech giants. However, it has accurately hit those foundational white-collar roles that are rule-defined and highly repetitive, and in this process, it has split the labor market into two distinct tracks. In this market, the middle ground is disappearing, and professional value is being redefined. Understanding this differentiation is more important than blind panic or optimism.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
