AI and the Workforce: What Is Actually Going On Inside American Firms?

This brief is a part of an ongoing series examining the impacts of AI on the U.S. workforce and implications for our education, training, and workforce systems. Explore the AI and Workforce Navigator to learn more.

As Artificial Intelligence spreads through the American economy, usage data only tells part of the story. A previous explainer examined how widely AI is being used across industries, firms, and regions using data from the U.S. Census Bureau’s Business Trends and Outlook Survey (BTOS). This piece goes deeper, drawing on data from new supplemental AI questions asked between November 2025 and February 2026 of more than 117,000 business representatives across all sectors of the economy. They focus on what AI is doing inside the firms that are using it, including:

Which business functions it touches

How it affects tasks, workers, and employment

What holds back firms that haven’t yet adopted it

Our analysis draws on two recent papers, one by Census Bureau researchers and another by a Federal Reserve Board economist.

Workers are Using AI Ahead of Their Employers

The supplemental data show that only 18% of firms reported use of AI during the last two weeks in any of their business functions. Respondents may interpret this question in different ways, but the wording implies more formal integration of AI than incidental or personal use. This likely explains why the figure is lower than some other estimates, such as surveys that ask individual workers about their personal use of AI tools.

(function(){function e(){window.addEventListener(`message`,function(e){if(e.data[`datawrapper-height`]!==void 0){var t=document.querySelectorAll(`iframe`);for(var n in e.data[`datawrapper-height`])for(var r=0,i;i=t[r];r++)if(i.contentWindow===e.source){var a=e.data[`datawrapper-height`][n]+`px`;i.style.height=a}}})}e()})();

Indeed, BTOS data show that worker-level AI use frequently outpaces formal firm-level adoption. Census Bureau researchers find that in 36% of firms where employees use AI for work tasks, the firm itself reports no official AI adoption. Workers are often bringing AI into their workflows independently, through tools like ChatGPT or Microsoft Copilot, before their organizations have developed formal policies or strategies around it.

This “bottom-up” diffusion has practical implications. It suggests that firm-level adoption statistics likely underestimate the true reach of AI through the workforce. It also raises questions about governance: when workers adopt AI tools independently, firms may lack visibility into how it is being used, what data is being shared with external tools, and whether outputs are being adequately reviewed. For policymakers thinking about AI workplace standards or liability frameworks, this dynamic matters.

AI Is Used in a Narrow Set of Business Functions

Survey data have shown that AI use is concentrated in knowledge-intensive sectors, including information, professional services, and finance, while those requiring physical work such as construction, manufacturing, and accommodation and food services lag well behind. A notable insight from the supplemental BTOS data is how narrowly AI is being used even among firms that use it, according to Census Bureau researchers, 57% of firms that used AI in at least one business function in the last six months deploy it in just one to three of the 15 business functions tracked in the survey (see chart below for a list of the functions). This suggests that AI is currently a targeted tool used in specific parts of the business rather than a fundamental transformation of how the organization operates.

Zooming out to all firms surveyed, sales and marketing are the functions most commonly using AI , followed by strategy and business development, IT, and R&D. Functions tied to physical production, distribution, quality management, and production of goods, see the lowest rates of AI use. However, use in these areas is beginning to emerge, for example, in warehouse and logistics operations.

(function(){function e(){window.addEventListener(`message`,function(e){if(e.data[`datawrapper-height`]!==void 0){var t=document.querySelectorAll(`iframe`);for(var n in e.data[`datawrapper-height`])for(var r=0,i;i=t[r];r++)if(i.contentWindow===e.source){var a=e.data[`datawrapper-height`][n]+`px`;i.style.height=a}}})}e()})();

AI Is Displacing Technology More Than Workers, For Now

The BTOS data offers important new insight into how AI is changing the nature of work, distinguishing between the following three types of task effects.

Substitution: the survey asks if AI was used to “perform a task previously done by an employee.”

Augmentation: the survey asks if AI was used to “supplement or enhance a task performed by an employee.”

Creation: the survey asks if AI was used to “introduce a new task not previously done by an employee.”

(function(){function e(){window.addEventListener(`message`,function(e){if(e.data[`datawrapper-height`]!==void 0){var t=document.querySelectorAll(`iframe`);for(var n in e.data[`datawrapper-height`])for(var r=0,i;i=t[r];r++)if(i.contentWindow===e.source){var a=e.data[`datawrapper-height`][n]+`px`;i.style.height=a}}})}e()})();

Among firms that report using AI in at least one business function in the last six months, 44% report using AI to augment worker tasks. Task substitution and creation are less common, each reported by around 10-11% of AI-using firms. Notably, while task substitution remains limited in scope, the intensity is rising, the share of firms reporting a ‘large number’ of substituted tasks more than doubled in the roughly 21 months between the first and second BTOS AI supplements.

These distinctions matter, but employment effects can be complicated. Task substitution, for example, is often associated with job loss, but the emergence of ATMs, which automated many bank teller tasks, actually increased teller employment as banks opened more branches. Understanding which specific tasks are affected and how this shapes the skills needed in specific jobs would help policymakers design education and training responses as the nature of work evolves.

Given these complexities, task-level data alone cannot tell us what is happening to employment overall. The BTOS also asks firms directly about net employment effects, providing a cleaner read on this question. The current data show stability: 95% of AI-using firms report no change in total employment attributable to AI. Employment increases and decreases attributable to AI affect roughly 2% of AI-using firms.

(function(){function e(){window.addEventListener(`message`,function(e){if(e.data[`datawrapper-height`]!==void 0){var t=document.querySelectorAll(`iframe`);for(var n in e.data[`datawrapper-height`])for(var r=0,i;i=t[r];r++)if(i.contentWindow===e.source){var a=e.data[`datawrapper-height`][n]+`px`;i.style.height=a}}})}e()})();

Interestingly, AI appears to be affecting legacy technology more visibly than labor. Census Bureau researchers find that 16% of AI-using firms report replacing existing software or equipment with AI, compared to roughly 2% reporting AI causing any decline in employment.

Why Most Firms Aren’t Using AI For firms not planning to use AI in the next six months, roughly 56% of all firms surveyed, the most common explanation is that AI simply does not seem applicable to their business, cited by 62% of the group.

(function(){function e(){window.addEventListener(`message`,function(e){if(e.data[`datawrapper-height`]!==void 0){var t=document.querySelectorAll(`iframe`);for(var n in e.data[`datawrapper-height`])for(var r=0,i;i=t[r];r++)if(i.contentWindow===e.source){var a=e.data[`datawrapper-height`][n]+`px`;i.style.height=a}}})}e()})();

The second and third most common barriers are a lack of knowledge about AI’s capabilities (22%) and concerns about privacy and security (21%). Notably, cost and regulatory barriers rank near the bottom of the list. This pattern suggests boosting AI literacy should be a priority, to help businesses understand what AI can and cannot do for their operations and how to adopt it sensibly.

Conclusion

The BTOS data offers policymakers a more grounded picture of AI in the workplace than headlines typically suggest. AI use remains narrow even among firms that use it. Most firms aren’t using it, and self-reported employment effects so far are modest. But workers are outpacing their employers, knowledge gaps are limiting adoption, and the intensity of task substitution is rising. Both policymakers and businesses have a role to play in addressing barriers to adoption and keeping pace with how AI is used in the workplace, while staying alert to more disruptive scenarios as AI capabilities continue to advance.

The post AI and the Workforce: What Is Actually Going On Inside American Firms?  appeared first on Bipartisan Policy Center.