Despite AI's growing capabilities, software engineers are not being replaced—because writing code is only a small part of the job.

What happened

A new essay by Arvind Narayanan and Sayash Kappor argues that the evidence does not support the narrative of mass AI-driven job losses in software engineering. In the first year after New York added an AI disclosure checkbox to WARN Act filings in March 2025, more than 160 companies filed WARN notices, but not a single one cited AI as the reason.

Why it matters

Software engineering is uniquely exposed to AI disruption because writing code—the most automatable part—is not the bottleneck. The real work lies in deciding what to build, verifying the results, and maintaining deep understanding of the codebase, business, and environment. If software engineers are safe from displacement, most other professions with stronger regulatory barriers and less automatable core tasks are likely to be even more protected.

What to watch

The essay identifies three persistent bottlenecks that resist automation: (1) deciding and specifying what to build, (2) verifying and being accountable for deliverables, and (3) the deep human understanding required to carry out both. One engineer quoted in the piece notes that while AI assistance helps with deciding and verifying, the value produced still depends on how deeply the person understands the problems and solutions.