Introduction

Measurement of artificial intelligence (AI) diffusion has often struggled to keep pace with organizational deployment. Existing research frequently provides fragmented views, focusing either on firm-level adoption or individual worker-task usage. This paper addresses the gap by using a unified framework to link formal adoption with worker-task usage. It identifies two simultaneous diffusion channels: a “top-down” channel where firms deliberately implement AI through functional deployment, and a “bottom-up” channel starting with decentralised worker experimentation. This distinction is critical as formal adoption and worker use may have different connections to firm outcomes like productivity and employment.

Data and measurement

The study utilises the 2026 AI supplement to the U.S. Census Bureau’s Business Trends and Outlook Survey (BTOS). This survey provides near real-time measurements from a nationally representative sample of approximately 117,000 distinct firms. The framework maps survey content into three layers: firm-level adoption, functional deployment, and worker-task integration. This approach allows researchers to distinguish between the extensive margin (binary adoption) and the intensive margin, such as the breadth of deployment and the depth of task replacement intensity.

Patterns of firm AI use

Between November 2025 and January 2026, the firm-level AI use rate was 18%, or 32% when employment-weighted. Adoption is expected to reach 22% within six months. AI use is heavily concentrated in large firms and knowledge-intensive sectors like Information (38%), Professional Services (34%), and Finance (30%). In very large firms within these sectors, diffusion rates reach the 50–60% range. While aggregate AI use is sticky, larger firms appear more proactive in initiating and maintaining adoption compared to smaller enterprises.

AI use within firms

A notable decoupling exists between institutional adoption and operational use. In 36% of firms where workers utilise AI, there is no indication of formal firm-level adoption, suggesting a grassroots trajectory. Conversely, 19% of firms with formal adoption show no evidence of worker-task use, indicative of implementation lags in top-down strategies. Within adopter firms, scope remains limited; 57% use AI in three or fewer business functions, most commonly Sales and Marketing (52%) and Strategy (45%).

What AI does to work

AI impact is currently dominated by task augmentation rather than substitution. Approximately 44% of AI-using firms report augmenting worker tasks, and 66% of those experiencing task effects do so exclusively. Pure task substitution and task creation are rare, occurring in only 5% of firms each. Consequently, AI-driven employment changes are modest; only 5% of using firms report a headcount impact, split nearly equally between increases (2.3%) and decreases (2.0%). However, AI is serving as a substitute for legacy capital, with 16% of firms replacing equipment and software.

Organisational adjustments to use AI and barriers to adoption

Over half of AI-using businesses (64%) report making no institutional adjustments, suggesting a reliance on “off-the-shelf” tools or a lag in restructuring. Common changes include training current staff (15%) and developing new workflows (15%). The primary barrier to adoption is the perception that AI is not applicable to the specific business (65%). Lack of knowledge regarding AI capabilities (22%) and concerns about privacy or security (20%) are also cited as significant impediments.

Firm-level indices of AI use and impact

To manage survey dimensionality, the researchers constructed indices for functional breadth, worker-task integration, operational investment, and impact. These components are aggregated into a holistic AI integration index. Data reveal a highly skewed distribution, where only a small fraction of firms show high integration levels. Active users show significant heterogeneity, indicating that functional breadth and task integration do not always scale in a uniform or linear fashion, as firms make distinct strategic choices.

AI use and firm outcomes

Regression analysis shows that broader AI integration is connected to superior firm performance. A one-standard deviation increase in the overall use index correlates with a 3.2 percentage point (pp) increase in the likelihood of high current performance and a 6.4 pp increase in sales growth probability. While functional breadth and operational investments are associated with current and future employment declines, worker-task use shows no statistically significant relationship with headcount reductions, functioning instead as a productivity enhancer.

A comparison with AI use rate estimates in related literature

The BTOS results are compared with other surveys like the Survey of Business Uncertainty (SBU) and the Real-Time Population Survey (RPS). Differences in estimates are often attributed to sampling frames, question wording, and respondent types. For instance, SBU estimates are higher because they focus on large firms in AI-intensive sectors. When the BTOS sample is restricted to similar cohorts, adoption patterns become broadly aligned, highlighting the concentration of AI use in specific industries and large organisations.

Conclusion and future work

The findings indicate that AI diffusion is in a nascent stage, primarily characterised by task augmentation in knowledge-intensive sectors. Proactive organisational adjustments are identified as critical for traversing the “productivity J-curve”, though such actions carry short-term implications for headcount. Future research aims to link these survey measures to administrative microdata to examine realised outcomes such as earnings, productivity, and firm dynamics. Identifying the specific mechanisms of impact remains a critical objective for understanding AI’s structural footprint in the economy.