A new essay by Princeton researchers Arvind Narayanan and Sayash Kapoor says the idea that artificial intelligence is rapidly eliminating software engineering jobs is not supported by the evidence. Instead, they argue that AI tools are changing parts of the job while leaving the core work of engineers largely intact.
The pair, writing in their series on AI as a normal technology, say software engineering offers a useful test case because AI adoption has been fast in the field and the technology is especially capable at generating code. Even so, they contend there is still not enough evidence to conclude that AI has triggered mass displacement of programmers. Their broader argument is that most knowledge work is harder to automate than headline-grabbing claims suggest.
Much of the essay focuses on recent layoffs at tech companies that were publicly linked to AI. Narayanan and Kapoor say those cases often do not hold up under closer scrutiny.
They point to Block’s February layoffs, which the company’s leadership connected to AI and a push for smaller teams. The authors say later reporting painted a different picture, describing financial pressure after rapid pandemic-era headcount growth. They also cite employee accounts suggesting AI delivered only modest productivity gains.
Snap is another example they examine. The company cut about 1,000 jobs in April, and CEO Evan Spiegel said AI played a major role while also claiming the company’s code generation was heavily automated. Narayanan and Kapoor say the timing and structure of the cuts fit a broader cost-cutting drive, including pressure from an activist investor, rather than a clean substitution of engineers by AI.
They give a similar reading of Intuit’s May layoffs. Although the company had also announced partnerships with major AI firms, its chief executive said the cuts were not caused by AI and instead were aimed at reducing coordination-heavy roles and management layers.
The authors describe this pattern as a form of “AI washing,” where companies emphasize AI in public explanations for layoffs because the rationale is more palatable to investors, employees or the public than financial restructuring. They also cite survey data suggesting many executives and hiring managers are quick to invoke AI even when no mature system is ready to replace the affected workers.
Narayanan and Kapoor say the clearest official labor data does not show software engineering employment in free fall. They point to Federal Reserve research that finds U.S. software engineer jobs are still growing, though at a slower pace than a no-AI counterfactual would imply.
They also note a New York disclosure requirement for certain layoffs. In the first year after the state added an AI checkbox to WARN filings, the authors say, nearly none of the companies reporting mass layoffs identified AI as the cause. To them, that suggests direct AI-driven cuts are still rare.
At the same time, they acknowledge that AI can affect job demand indirectly. Tools that reduce demand for products like homework help or community Q&A can shrink staffing needs at those businesses even if the AI is not directly doing the same job. They also say companies that sell AI may shift workers away from older lines of business toward newer products. But they argue those are examples of ordinary restructuring, not a simple story of machines replacing engineers.
A central claim in the essay is that coding is not the main limiting factor in software work. The authors say engineers spend only a portion of their time writing code, with much of their work involving planning, debugging, meetings, verification and understanding the larger business and technical environment.
They describe software development as a “decide-execute-deliver” process. AI may make the execution step faster by generating code, but Narayanan and Kapoor argue it does not remove the need to decide what should be built, check that it works, and take responsibility for the result. Those tasks, they say, depend on deep human context that current AI systems do not replace.
Their conclusion is cautious rather than alarmist. AI may change how software is built and slow the pace of hiring in some places, but they say the evidence so far does not support claims that it is on the verge of wiping out software engineering jobs altogether.