AI vs Software Engineers: Why mass layoffs have not occurred

Arvind Narayanan and Sayash Kappor argue that AI has not, and is unlikely to, trigger mass layoffs among software engineers, presenting evidence and task-level analysis in an essay linked on 14th June 2026. The authors say the available data does not support the narrative that once AI capabilities reach a threshold, it will cause mass job losses, and they point to recent U.S. disclosure data as concrete counterevidence.

The essay and the concrete evidence

The authors note a specific, measurable test case: in March 2025 New York became the first U.S. state to add an AI disclosure checkbox to WARN Act filings. In the full first year after that change, more than 160 companies filed WARN notices, and not a single one checked the AI box. The essay uses that absence of AI-checked WARN filings to challenge the claim that rising AI capabilities have already led to mass layoffs in a sector with few regulatory barriers.

Narayanan and Kappor frame that finding inside a broader claim: "there is enough evidence to reject the narrative that once AI capabilities reach a certain threshold, it will cause mass layoffs." They treat software engineering as a profession particularly exposed to automation because much of the visible work is typing code into a computer, but they argue the reality of engineering work is more complex.

Where AI helps and where it doesn’t

The essay concedes that AI "speeds up the typing-code-into-a-computer phase," but stresses software engineering involves many tasks beyond raw coding. Task-breakdown surveys highlighted activities such as meetings and debugging as bottlenecks. To probe those, Narayanan and Kappor went qualitative and identified three core bottlenecks that resist automation:

1. Deciding and specifying what to build.
2. Verifying and being accountable for what is delivered.
3. The "deep human understanding" of the codebase, the business, and the environment required to carry out both of the above.

The authors write that these three aspects are the real constraints on productivity, not raw code-generation speed. The linked commentary adds a first-person note: the writer finds AI assistance helpful for the deciding and verifying steps, but emphasizes that the value they produce still depends on "how deeply I understand both the problems and the solutions that the agents are building for them."

Prior context and why the paper focuses on software engineering

The essay treats software engineering as a high-exposure sector because it has relatively few regulatory barriers and a visible automation surface: generating code. That makes it a useful test case for the broader claim that AI-caused mass unemployment is imminent. The New York WARN Act disclosure experiment provides a concrete institutional measure: despite a year of WARN filings after the checkbox was introduced, more than 160 notices produced zero self-identified AI-related layoffs in filings.

Narayanan and Kappor also draw on task-breakdown surveys showing non-coding activities dominate engineers’ time. From that empirical starting point they argue that automation of typing is not sufficient to remove the human roles tied to specification, verification, and deep contextual knowledge.

Why it matters

If the essay’s evidence and reasoning hold, the implication is that headlines about imminent, AI-driven mass layoffs may be overstated, at least in software engineering. That matters because software engineering is a low-regulation, high-automation-exposure profession; if it is cushioned by human-centered bottlenecks, other professions with higher regulatory or contextual barriers may be even less vulnerable. The distinction between code generation and the broader responsibilities of engineers reframes conversations about AI’s labour-market impact.

What to watch

Watch subsequent WARN Act filings in New York and other jurisdictions for any increase in companies checking an AI-related box, and watch task-level studies that measure changes over time in the share of engineer hours spent on specifying, verifying, and maintaining deep systems knowledge. Those signals would confirm whether the essay’s observed gap between capabilities and layoffs narrows or persists.