Is AI actually replacing workers, or just changing what they do?
For technology and business leaders trying to plan next year’s headcount, the answer has so far come mostly from vendor projections and alarming headlines. The latest wave of research — Federal Reserve surveys, a national small-business poll and labor market trackers from California and Stanford — offers something more grounded: jobs data and what employers themselves say is happening.
The freshest entry comes from the Federal Reserve Bank of St. Louis. Businesses across the Fed’s Eighth District — a swath of mid-America covering Arkansas, most of Missouri and parts of five neighboring states — report AI-related efficiency and output gains, while nearly half expect no noticeable AI-related staffing effect over the next year.
Regional adoption trends and barriers to use
The findings come from special questions in the bank’s May 2026 edition of its quarterly Economic Conditions Survey. According to the researchers, AI adoption is wide but shallow: about one-third of respondents (34%) say a small share of employees use AI regularly, and another third (36%) say employees are either testing AI without regular use or report no adoption at all.
It was the limited-or-no-use group that the researchers probed on barriers to AI: of those, 38% cite gaps in skills, data or technical infrastructure, and 34% say available tools do not address a business need. Individual respondents also cite hallucinations, insufficient controls, and limited training.
Output gains and altered hiring plans
Behind the percentages, respondents offer a ground-level view of where the AI gains are landing. A respondent from a law firm estimates that AI-supported research, deposition summaries and first drafts saved each attorney two to three hours a week. A leader at a professional services firm credits AI adoption with supporting 35% revenue growth over the past year without additional staff, while a counterpart at another firm credits AI with a 15% rise in revenue per employee.
Others tie the gains directly to hiring plans. A respondent from an accounting firm reports a 30% to 40% increase in output capacity, reducing the need for a planned staff expansion, and a restaurant operator says automated order-taking has eliminated the need for a drive-thru position.
However, those AI-driven headcount reductions are the exception rather than the rule: among respondents to the staffing question, 49% expect no noticeable staffing effect during the next 12 months, nearly 20% anticipate just a slight reduction in staffing needs and 18% expect skill changes rather than headcount changes.
The researchers wrote that AI is enabling firms to “do more with the same workforce.”
Mixed workforce effects and a focus on retraining
The St. Louis findings echo what the New York Fed saw a year earlier in its August 2025 regional business surveys, which found only 1% of AI-using service firms reported layoffs during the previous six months, while 12% said they had hired fewer workers. At the same time, 11% of AI-using service firms and 7% of manufacturers said they had hired more workers because of AI.
About one-third of AI-using service firms reported retraining employees. The researchers concluded that AI was “more likely to result in retraining than job loss.”
A national Ipsos poll of 750 small-business owners and operators, conducted May 19 to June 4, 2026, for the US Chamber of Commerce Foundation, adds detail on how employers say AI-related gains are used. Among respondents reporting faster or better work, 65% say employees produce more or higher-quality output; 56% cite benefits to learning, planning or review; and 36% said employees avoid overtime or extra hours.
The Ipsos/USCCF poll also found that 73% of AI-using employers report changes to roles and responsibilities, 70% report changed performance expectations and 65% say AI has influenced hiring or personnel decisions.
Tracking actual job displacement and unemployment claims
Employer surveys capture what business leaders say; unemployment data shows what’s happening to workers.
On that front, California’s AI-Unemployment Tracker found no statewide surge in unemployment-insurance claims from workers in highly AI-exposed occupations through May 2026. The underlying California Policy Lab analysis also found no meaningful increase in the proportion of claims coming from AI-exposed workers compared with before the pandemic.
The researchers cautioned that occupational exposure indicates whether tasks could be or have been performed by AI, not whether AI caused a particular employer to eliminate a job.
In its July 22 update, the Stanford Digital Economy Lab’s Canaries Dashboard reported that employment growth in its sample was slowest in the two most AI-exposed occupational groups, although the overall differences were modest. Declines were concentrated among workers aged 22 to 25 in exposed roles such as software development and customer service.
Study limitations and the broader economic picture
Stanford said its ADP-based sample does not represent the entire U.S. labor market and measures correlation between occupational exposure and employment trends rather than a causal relationship.
The St. Louis Fed findings carry separate limitations. Its published analysis does not disclose the number of respondents to the special questions or independently verify reported revenue, capacity or staffing effects. The regional survey documents employer-reported changes in output, hiring plans and skill requirements, while the California and Stanford sources measure different worker outcomes. None establishes an economywide employment effect caused by AI.
For that, business leaders will need to keep watching. But an early answer to the main question is taking shape: so far, AI isn’t replacing workers so much as changing what they do — and reshaping hiring plans going forward.