Why hasn't AI caused large-scale unemployment among software engineers? Latest research: humans are irreplaceable in judgment and accountability

Technology Column "Normaltech.ai" recently released the latest research report indicating that although AI coding capabilities are rapidly growing, software engineers have not ushered in a "massive unemployment wave." The report reveals that many companies' layoffs are actually driven by financial considerations under the guise of "AI cleaning (AI Washing)." In the three-layer architecture of software development—"decision, execution, delivery"—AI can only compress the "execution" phase; humans remain the irreplaceable core in judgment and accountability, and future market demand for software engineers may even increase accordingly.
(Background note: Attention — the photos you take, the words you say, Google now wants to store them for AI training (how to turn off training))
(Additional background: Anthropic CEO: Governments should have the authority to veto high-risk AI, mandatory testing before deployment, three main propositions clash with Trump’s deregulation route)

Table of Contents

Toggle

• The Truth About the Tech Giants' Layoff Wave: It’s Actually "AI Washing"
• The "Sandwich Architecture" of Software Development: Human Judgment and Accountability Are Hard to Replace
• Say Goodbye to "Intuitive Coding," Embrace Agentic Engineering
• Jevons Paradox Strikes: Future Job Openings Not Decreasing but Increasing

The fear that AI will eliminate software engineers is spreading in the tech industry, but actual data paints a completely different picture. According to the specialized tech column "Normaltech.ai," focused on AI trends, published on the 10th of this month, despite the rapid rise of AI applications in code generation, there is no evidence supporting the claim that "software engineers are facing large-scale replacement." The report uses in-depth industry data and case analysis to strongly counteract market hype and doomsday panic.

The truth about the tech giants' layoff wave: It’s actually "AI washing"

The report straightforwardly states that many headlines about "AI replacing human labor" layoffs are not credible upon closer inspection. For example, Block (led by Jack Dorsey) announced laying off 4,000 employees and attributed part of the reason to AI, but the truth is that over-expansion during the pandemic and financial pressure were the main causes. Internal staff even said that the productivity boost from AI was minimal. Similarly, the large-scale layoffs at Snap and Intuit were driven more by aggressive investor cost-cutting pressures rather than direct AI impact.

Data from investigations further ruthlessly expose this illusion: up to 59% of recruiting managers admit to exaggerating AI’s role in layoffs just to appear "forward-looking" to investors. Under the U.S. WARN Act (which requires disclosure of AI-related layoff factors), despite thousands being laid off, almost no companies officially reported AI as the reason.

The "Sandwich Architecture" of Software Development: Human Judgment and Accountability Are Hard to Replace

The report introduces a core "Decide-Execute-Deliver" sandwich framework, precisely analyzing the essence of software engineering:

• Decide: Involves problem framing, requirement specification, and business goal planning.
• Execute: Design and coding implementation, which is currently AI’s strongest suit.
• Deliver: Testing, validation, system integration, and security accountability.

AI has indeed greatly compressed the "execute" stage. GitHub research shows that AI can increase code output by 8 times, but the final software release volume only increased slightly by 30%. The reason is that the "decide" and "deliver" ends require deep contextual understanding, flexible business judgment, and human "accountability." The report points out that developers spend only 9% to 61% of their time on pure coding, with the rest consumed by complex architecture issues, and supervising AI agents is highly mentally taxing.

Say Goodbye to "Intuitive Coding," Embrace Agentic Engineering

The report further distinguishes two current AI development modes: "Vibe coding" (intuitive programming) and "Agentic engineering." The former relies on casual prompts and lacks human review, which is extremely risky, with a vulnerability rate 9 times higher, and only about 44% of code survives into projects.

In contrast, the mainstream professional practice is "agentic engineering," where humans must maintain control, carefully review code, and bear ultimate safety responsibility. This means companies cannot rely on non-engineers lacking professional architecture skills to release production-level software critical to business operations.

Jevons Paradox Strikes: Future Job Openings Not Decreasing but Increasing

Looking ahead, the report adopts a cautiously optimistic attitude. According to the economic "Jevons paradox," when the costs and barriers to software construction are significantly lowered, market demand for software will grow exponentially. For example, modern cars already contain over 100 million lines of code, and future societal demand for software applications has virtually no short-term limit.

Therefore, although AI tools are changing work patterns and even slowing some companies’ recruitment pace due to increased development efficiency, overall demand elasticity suggests that the total employment need for software engineers will remain stable or even grow. This powerful "normal technology" will elevate engineers to higher-level decision-making roles rather than bring about a doomsday unemployment wave.