AI reached 18% of U.S. firms in late 2025 (32% when weighted by employment), concentrated in large, knowledge-intensive companies but typically confined to a few functions and tasks. Broader AI deployment correlates with stronger firm performance, while only functional-level integration — not day-to-day worker task use — associates with modest employment declines after controlling for investments.
The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks
Using a nationally representative 2026 BTOS AI supplement, the paper documents that 18% of U.S. firms (32% employment-weighted) used AI in late 2025, concentrated in large and knowledge-intensive firms with limited within-firm scope, and finds positive correlations between broader AI integration and firm performance while functional deployment (but not worker-task use, once controlled) is associated with employment declines.
Using novel, nationally representative data from the 2026 AI supplement to the U.S. Census Bureau’s Business Trends and Outlook Survey (BTOS), we characterize AI diffusion across three layers: firm-wide adoption, business-function deployment, and worker-task use. During Nov 2025–Jan 2026, 18% of firms used AI in at least one function (32%, employment-weighted), with adoption expected to reach 22% within six months. Use is concentrated in large firms and knowledge-intensive sectors, reaching 50%–60% (60%–70%, employment-weighted) among very large firms in Information, Professional Services, and Finance. Among adopters, scope remains limited: 57% use AI in three or fewer functions, most often Sales and Marketing (52%), Strategy (45%), and IT (41%). Worker-level use appears in 23% (41%, employment-weighted) of firms, primarily for writing, document analysis, and information search; 65% restrict use to three or fewer tasks. Evidence suggests both top-down and bottom-up diffusion: worker use can occur without firm adoption, and vice versa. Most firms (66%) use AI for task augmentation, while employment reductions are rare (2%). Regression results show a positive relationship between firm performance and AI integration breadth. However, functional deployment and operational investment are associated with employment declines, while worker-task use is not once these factors are controlled for.Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org.
Summary
Main Finding
Using the 2026 AI supplement to the U.S. Census Bureau’s nationally representative Business Trends and Outlook Survey (BTOS), the paper documents early but uneven AI diffusion across three layers—firm-wide adoption, business-function deployment, and worker-task use. As of Nov 2025–Jan 2026, AI use is already nontrivial but concentrated: 18% of firms (32% employment-weighted) reported using AI in at least one function, with expected adoption rising to 22% within six months. Adoption is skewed toward large and knowledge‑intensive firms; within adopters, scope is usually limited and focused on augmentation rather than outright automation. Regression analysis links broader AI integration to better firm performance, but finds employment declines are associated with deeper functional deployment and operational investment rather than with worker-level task use once controls are included.
Key Points
- Adoption prevalence (Nov 2025–Jan 2026):
- 18% of firms used AI in ≥1 business function (32% when weighted by employment).
- Expected to reach 22% of firms within six months.
- Concentration:
- Large and knowledge‑intensive sectors (Information, Professional Services, Finance) show the highest adoption.
- Very large firms in those sectors: 50%–60% of firms (60%–70% employment-weighted) report AI use.
- Functional deployment (among adopters):
- 57% use AI in three or fewer functions.
- Most common functions: Sales & Marketing (52%), Strategy (45%), IT (41%).
- Worker-task use:
- Reported in 23% of firms (41% employment-weighted).
- Primary worker tasks: writing, document analysis, information search.
- 65% of firms restrict AI use to three or fewer tasks.
- Diffusion pathways:
- Evidence of top-down (firm-led) and bottom-up (worker-driven) diffusion: worker use can occur without formal firm adoption and vice versa.
- Effects on work and employment:
- 66% of firms primarily use AI for task augmentation.
- Reported employment reductions are rare (2% of firms).
- Regression results: broader AI integration correlates positively with firm performance. Functional deployment and operational investment correlate with employment declines; worker-task use does not once those firm-level factors are controlled for.
Data & Methods
- Data source: 2026 AI supplement to the U.S. Census Bureau’s Business Trends and Outlook Survey (BTOS). Nationally representative cross‑section covering Nov 2025–Jan 2026; employment-weighting used to show exposure of workers.
- Measurement: Three-layer framework—(1) firm-wide adoption indicator, (2) function-level deployment across business functions, (3) worker-level task use.
- Analyses:
- Descriptive statistics on prevalence, sectoral and size heterogeneity, and concentration of functions/tasks.
- Regression analyses linking AI adoption breadth and depth to firm outcomes (performance metrics) and employment outcomes, controlling for firm characteristics and layers of adoption to disentangle where employment effects appear.
- Limitations (implicit in design):
- Cross-sectional, self-reported survey data—limits causal inference and is subject to measurement/expectation biases.
- Short-run snapshot; expected near-term adoption captured but medium/long-term dynamics require follow-up.
Implications for AI Economics
- Distributional concentration: Rapid uptake concentrated in very large and knowledge-intensive firms implies growing productivity differentials across firms and potential increases in within- and between-firm inequality. Employment-weighted adoption figures show many workers are already exposed, mostly via large employers.
- Augmentation vs automation: Most firms report augmentation rather than substitution. However, deeper functional integration and operational investments are the channels associated with job reductions, signaling that the mode of integration matters for labor outcomes.
- Diffusion mechanics: The coexistence of worker-driven and firm-driven adoption highlights multiple diffusion pathways. Policies and models should account for bottom-up diffusion (e.g., employee-initiated tools) as well as centralized rollouts.
- Measurement priorities: Functional deployment and operational investment are key predictors of employment change—surveys and empirical work should track deployment depth (which functions and processes) not just binary adoption.
- Policy and firm strategy:
- Labor-market policy: early focus on upskilling, task reallocation, and safety nets in sectors/firm size classes where deep functional deployment is happening.
- Competition and productivity: concentration of AI in large firms suggests monitoring competitive dynamics and market power implications.
- Research agenda: need for longitudinal and causal studies to separate selection (better-performing firms adopt more) from causal productivity and employment effects; granular task-level and process-level measurement will clarify mechanisms.
- Takeaway: Early AI diffusion is meaningful but uneven; short-run labor impacts appear limited and concentrated where firms integrate AI deeply into functions and operations. Understanding long-run economic effects requires tracking depth of deployment and causal identification of performance and employment channels.
Assessment
Paper Typecorrelational
Evidence Strengthmedium — The study uses a novel, nationally representative supplement to the Census BTOS with employment weights, providing strong, generalizable descriptive evidence on AI diffusion across firms, functions, and tasks; however, causal claims (e.g., on employment effects or productivity) rely on cross-sectional associations and controls rather than quasi-experimental variation, leaving results vulnerable to omitted variable bias and reverse causality.
Methods Rigormedium — Survey design, national representativeness, and employment-weighting indicate high descriptive rigor; regression analyses include controls for firm size, sector, and investments, but there is no exogenous variation or longitudinal identification to credibly isolate causal effects, and key measures are self-reported which may introduce measurement error.
SampleNationally representative sample of U.S. employer firms from the 2026 AI supplement to the U.S. Census Bureau’s Business Trends and Outlook Survey (BTOS), fielded Nov 2025–Jan 2026; includes firm-level reports on firm-wide AI adoption, function-level deployment, and whether workers use AI for specific tasks, with employment weights to reflect aggregate exposure.
Themesadoption productivity
IdentificationNo causal identification strategy; results are based on cross-sectional descriptive statistics and multivariate regressions relating AI adoption and deployment measures to firm outcomes while controlling for observable firm characteristics and investments.
GeneralizabilityU.S.-only — may not generalize to other countries or regulatory environments, Snapshot (Nov 2025–Jan 2026) during a fast-moving adoption period — patterns may change quickly, Self-reported firm-level measures — potential reporting bias and limited granularity on intensity of use, Cross-sectional design — limits causal inference and dynamics of diffusion over time, May underrepresent very small, informal, or non-employer businesses if outside BTOS frame
Claims (14)
| Claim | Direction | Confidence | Outcome | Details |
|---|---|---|---|---|
| During Nov 2025–Jan 2026, 18% of firms used AI in at least one function. Adoption Rate | positive | high | firm-level AI adoption (use in at least one function) |
18%
0.5
|
| Employment-weighted adoption rate was 32% (i.e., 32% of employment is in firms using AI in at least one function). Adoption Rate | positive | high | employment-weighted firm AI adoption |
32%, employment-weighted
0.5
|
| Adoption is expected to reach 22% of firms within six months. Adoption Rate | positive | high | expected firm-level AI adoption within six months |
22% within six months
0.3
|
| AI use is concentrated in large firms and knowledge-intensive sectors, reaching 50%–60% (60%–70%, employment-weighted) among very large firms in Information, Professional Services, and Finance. Adoption Rate | positive | high | firm-level AI adoption by firm size and sector |
50%–60% (60%–70%, employment-weighted) among very large firms in Information, Professional Services, and Finance
0.5
|
| Among adopter firms, scope remains limited: 57% use AI in three or fewer functions. Task Allocation | negative | high | number of business functions using AI per adopting firm (breadth of functional deployment) |
57% use AI in three or fewer functions
0.5
|
| Among adopter firms, AI is most often used in Sales and Marketing (52%), Strategy (45%), and IT (41%). Task Allocation | positive | high | functional deployment proportions (Sales & Marketing, Strategy, IT) |
Sales and Marketing (52%), Strategy (45%), and IT (41%)
0.5
|
| Worker-level AI use appears in 23% of firms (41%, employment-weighted), primarily for writing, document analysis, and information search. Task Allocation | positive | high | presence of worker-level AI use within firms and primary tasks where used |
23% (41%, employment-weighted)
0.5
|
| Among firms with worker-level AI use, 65% restrict use to three or fewer tasks. Task Allocation | negative | high | breadth of worker-task AI use per firm (number of tasks) |
65% restrict use to three or fewer tasks
0.5
|
| Evidence suggests both top-down and bottom-up diffusion: worker use can occur without firm adoption, and vice versa. Adoption Rate | mixed | high | co-occurrence (or lack thereof) of firm-wide adoption and worker-level AI use |
0.3
|
| Most firms (66%) use AI for task augmentation rather than replacement. Task Allocation | positive | high | reported primary role of AI (task augmentation vs replacement) |
66%
0.5
|
| Employment reductions attributable to AI are rare: only 2% of firms report employment reductions. Job Displacement | negative | high | reported employment reductions due to AI |
2%
0.5
|
| Regression results show a positive relationship between firm performance and breadth of AI integration. Firm Productivity | positive | high | firm performance (as related to AI integration breadth) |
0.3
|
| Functional deployment and operational investment in AI are associated with employment declines. Job Displacement | negative | high | employment change associated with functional deployment and operational investment |
0.3
|
| Once functional deployment and operational investment are controlled for, worker-task use is not associated with employment declines. Job Displacement | null_result | high | association between worker-task AI use and employment change conditional on other AI deployment measures |
0.3
|