Census data shows AI use rising but uneven. Large firms report materially higher use, while most adopters remain concentrated in a small number of functions. Engineer View’s conclusion: the differentiator is increasingly the organization around the model—workflow ownership, data access, evaluation, change capacity, and decision rights.

Key takeaways

Firm size is a proxy for deployment capacity, not merely software access.
Narrow functional use suggests experimentation has not yet become an operating system.
Governance should accelerate bounded use—not become a detached compliance layer.

01 · The capacity divide

Scale changes the ability to absorb new technology.

Larger organizations can fund data integration, security review, domain evaluation, procurement, and training in parallel. Smaller firms may have faster decision cycles but less spare capacity to redesign work.

Figure 1. Reported business AI use by scale.

All firms, point estimate from research paper18%

250+ employees, May 2026 story37%

Sources: U.S. Census Bureau working paper and Business Trends and Outlook Survey story. Measures come from related but distinct Census products and are shown directionally, not as a controlled comparison.

02 · Functional breadth

The real transition is from tool to workflow.

Census researchers report that 57% of adopting firms use AI in three or fewer functions. Leading functions include sales and marketing, strategy and business development, and information technology. Breadth matters because durable value usually requires handoffs across people, data, controls, and systems.

52%Sales & marketing

45%Strategy & business development

41%Information technology

Source: Census working paper CES-26-25. Percentages refer to AI-adopting firms in the study.

03 · Operating model

Build the deployment loop.

Frame

Name the decision, user, evidence, and failure mode.

Integrate

Connect governed data and the actual system of work.

Evaluate

Test quality, risk, latency, and human escalation.

Operate

Assign ownership, monitor performance, and improve.

A pilot becomes an operating capability only when its result is measured, exceptions are handled, and someone owns the feedback loop.

04 · Governance

Control should be proportional to consequence.

NIST’s AI Risk Management Framework provides a voluntary structure for managing AI risks. For operating teams, the practical translation is to classify use cases by consequence, establish evidence requirements, define human authority, and monitor behavior after deployment.

Deployment test

Can the organization answer these five questions?

Who owns the outcome?
Which evidence can the system use?
How is quality evaluated?
When must a human intervene?
What changes when performance drifts?

Sources & method

Research record.

This note synthesizes two current Census publications and the NIST AI RMF. Organizational implications and the deployment loop are Engineer View analysis.

General research only. Survey estimates are subject to sampling and response limitations. Engineer View frameworks are analytical tools, not certification standards.