Not Replaced, Rebuilt: What Three Years of Survey Data Say About AI and Jobs
For three years now, the same headline has reappeared every few months: artificial intelligence is coming for the jobs. The projections are dramatic, the timelines short, and the tone somewhere between breathless and apocalyptic.
Meanwhile, a quieter body of evidence has been accumulating — and it tells a different story. Not a reassuring one, exactly, but a more specific one. And specificity is what most of us actually need when deciding what to do about our own careers.
What the surveys actually found
At the start of September, economists at the Federal Reserve Bank of New York published the third year of a survey asking businesses in New York and northern New Jersey a simple question: are you using AI, and what has it done to your workforce?
The adoption numbers are striking. Among service firms, 61 percent reported using AI in their business processes, up from 40 percent a year earlier and 25 percent in 2024. Manufacturers went from 16 percent in 2024 to 26 percent in 2025 to 51 percent this year. In two years, AI use roughly tripled in both groups.
If adoption at that speed were driving mass job losses, you would expect the layoff numbers to be climbing just as fast. They aren’t. Four percent of service firms said they had laid off workers because of AI in the previous six months — up from one percent the year before, but still a small minority. No manufacturers reported AI-related layoffs in either year.
The hiring picture is more interesting than a simple yes or no. About 15 percent of service firms said they had hired fewer people than they otherwise would have because of AI. But roughly 13 percent said they had hired more people, specifically to help the business use it. Those two effects are close to cancelling each other out.
What firms are doing instead, consistently and across all three years of the survey, is retraining. Just over a third of service firms using AI reported retraining staff in response to it, along with more than 20 percent of manufacturers. The researchers noted that retraining spans the full educational spectrum rather than concentrating at any one level.
The unglamorous reality behind the adoption numbers
Here is the part that rarely makes the headlines. Most of this AI adoption is shallow.
Three-quarters of service firms and more than 90 percent of manufacturers described their AI investment as minimal to modest — ranging from simply using free tools to allocating a small slice of overall spending. Only about 5 percent of service firms called AI adoption a major strategic investment.
And within the firms that have adopted it, use is concentrated in a small group. Among service firms using AI, the median share of employees actually using it was 17 percent. Among manufacturers, 7 percent.
So the accurate summary is not “AI has swept through the workplace.” It is closer to: most companies have now tried AI, a few people inside each company use it regularly, and almost nobody has bet the business on it yet.
The reasons non-adopters gave are worth noting too. Cost was among the least cited obstacles. About half said the nature of their work simply doesn’t suit AI. Roughly a quarter said the technology isn’t good enough yet to help them. More than a third raised concerns about data privacy or security, a similar share worried about accuracy, and about a third said they didn’t have staff with the skills to use it well.
That last one is the interesting gap. A third of companies that haven’t adopted AI say the blocker is people, not technology and not money.
What this means if you’re the one holding the job
Three things follow reasonably directly from the data.
First, the near-term risk to most roles is task displacement, not role elimination. The pattern across three years of surveys is that firms adjust by changing what people do, not by removing them. That shifts the question from “will my job exist” to “which parts of my job are moving, and what am I doing with the time that frees up.”
Second, retraining is currently the employer’s default move – and the content of that retraining is revealing. Firms described teaching basic AI literacy, tool-specific instruction, automating routine tasks, prompt engineering, and job-specific applications. But many also emphasised something less obvious: teaching people to verify outputs, recognise bias, follow data security rules, and avoid leaning on the tool too heavily. The skill being trained isn’t just use the AI. It’s know when and how far to trust it.
Third, there is a real caveat, and it points at entry-level workers. The New York Fed authors flagged a recent Stanford study suggesting that workers at the start of their careers may be affected significantly, because AI can substitute for exactly the routine tasks that junior employees have traditionally been handed. The same technology that makes an experienced professional faster may be removing the rungs of the ladder beneath them.
That deserves more attention than it usually gets. The optimistic framing – AI augments rather than replaces – may be true in aggregate while still being false for the twenty-four-year-old trying to get their first analyst job.
The useful takeaway
The gap between the headline narrative and the survey data isn’t a reason to relax. It’s a reason to redirect attention.
Worrying about whether AI will eliminate your profession is mostly unproductive; you can’t control the answer and the evidence doesn’t yet support the panic. Working out which specific tasks in your week are automatable, what you’d rather be doing with those hours, and whether you can actually judge when an AI output is wrong – that is both answerable and within your control.
The firms in this survey are making that bet on their own people. The reasonable move is to make the same bet on yourself, and not wait to be enrolled in someone else’s training programme.
One honest limitation: these findings come from a regional survey covering New York and northern New Jersey, so they reflect one economy rather than the world. But the authors note their results line up with the broader research literature, which has also found limited labour-market effects from AI adoption so far. The patterns are consistent – and the researchers themselves are careful to say that could change as the technology matures.
Sources
- Federal Reserve Bank of New York, Liberty Street Economics: “Businesses Are Using AI to Transform Work, Not Cut Jobs” (1 September 2026) – libertystreeteconomics.newyorkfed.org
- Stanford Digital Economy Lab: “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”