The output data strengthen the compression argument. By 2025 Q3, computer systems design output was 12.6% above its January 2024 level, while aggregate hours were 1.9% lower. Information output was 11.6% higher, while aggregate hours were 1.2% lower. That is not what a plain downturn looks like. In a demand slump, output and hours usually weaken together.
The pattern is closer to a productivity transition, though the data do not identify the technology responsible. Companies in these industries have been investing heavily in software, cloud infrastructure and generative AI tools, while also pushing cost discipline. When output rises and hours fall, the same volume of work is being produced with less measured labor input. That can come from automation, from better tools for existing staff, from outsourcing, from higher-skilled worker mix, or from stopping low-return projects.
For workers, the distinction is practical. A company does not need to announce an AI layoff for AI to change the job market. It can slow replacement hiring, ask fewer workers to handle more output, shift junior work to software, and keep payroll spending concentrated on experienced staff. Those changes lower entry points into white-collar careers before they produce a visible jump in unemployment.
The official data therefore answer the first part of the question with a narrow claim. AI-exposed sectors have not collapsed. They have become less labor-intensive. The burden of proof now moves to 2026: whether the early-year announcements and corporate language mark a new step down in employment, or simply continue a compression cycle already underway in 2024 and 2025.
