The Fed Counted the AI Layoffs. There Weren't Many.

Every small business owner we talk to has absorbed the same background message about AI: it is coming for the headcount. The question is only whether it arrives this year or next. That message shapes real decisions, including some bad ones, like deferring a hire the business genuinely needs or buying a tool on the theory that it will pay for itself in salary.

There is now survey data on what employers are actually doing, and it does not match.

What the survey found

On September 1 the Federal Reserve Bank of New York published an analysis by Jaison R. Abel, Richard Deitz, Natalia Emanuel and Nick Montalbano drawing on the bank's Regional Business Surveys, the Empire State Manufacturing Survey and the Business Leaders Survey, fielded in August across New York and northern New Jersey.

Adoption climbed steeply. Among service firms, 61 percent reported using AI this year, up from 40 percent last year and 25 percent in 2024. Among manufacturers the jump was larger in proportional terms: 51 percent this year against 26 percent last year and 16 percent in 2024. Manufacturing has roughly caught up to where services were twelve months ago.

Then the employment numbers. Four percent of service firms reported laying off workers in response to AI over the previous six months, up from 1 percent in last year's survey. No manufacturers reported AI-driven layoffs this year or last year.

Two other figures matter more than the layoff number. About 15 percent of service firms said they hired fewer workers than they otherwise would have, which is the quieter and more common version of the same effect. And roughly 13 percent of service firms hired *more* workers to support AI implementation.

Those two are nearly the same size. In this sample, the number of service firms adding people because of AI is close to the number holding back on hiring because of it.

The largest single response was neither. Just over a third of service firms and more than 20 percent of manufacturers reported retraining workers in response to AI.

Read the limits honestly

This is a regional survey, covering New York and northern New Jersey, not a national census. The industry mix in that footprint is heavier on finance and business services than the country as a whole, and a firm in Phoenix or Tuba City is not automatically described by it. It is a snapshot from August, and the trend line over three years is steep enough that next August could look different.

What makes it useful anyway is that it asks employers what they did rather than what they expect to do. Most of the AI-and-jobs material in circulation is forecasting. This is a count.

What we would take from it

The dominant corporate response to AI is training, not termination. That is the finding with the clearest operating consequence. If more employers are retraining staff than are doing anything else, and your plan contains no training line, your plan is out of step with what your competitors are actually doing. Training is also the cheapest part of an AI initiative and the first thing cut when a budget gets tight, which is precisely backwards.

The hiring freeze is the effect to watch, not the layoff. Four percent cutting staff is a small number. Fifteen percent quietly hiring fewer people is not, and it is nearly invisible from outside because there is no announcement when a role simply does not get posted. For a small business, this is the honest version of the question: not "who do I let go," but "which of the roles I was planning to add can I hold off on, and for how long." That is a real decision and it deserves to be made deliberately rather than by drift.

Somebody has to make the tools work. The 13 percent hiring to support AI implementation is the finding most often left out of the summaries. Software does not deploy itself into a business. Someone has to pick it, configure it, connect it to the systems that hold your data, write down how it is supposed to be used, and answer questions when it behaves oddly. In a company of twelve people that is not a new hire, it is a portion of someone's week, and it should be named and budgeted as such rather than absorbed silently by whoever is most curious.

Rising adoption is not evidence that adoption is working. The survey measures use. It does not measure return. A 61 percent adoption rate tells you what your peers have turned on, not what it earned them. We would not let that number drive a purchase, and we would be skeptical of any vendor who quotes it at you as though it does.

The useful reframing is this. The interesting question stopped being whether to adopt, because the answer is now visibly yes across most of the market. It became what you are willing to spend on making adoption produce something, and the data says the money is going into people rather than out of them.

If you are working out where AI fits in your business and want an assessment from someone with nothing to sell you but the advice, we are happy to have that conversation.