Will AI replace office staff? What the data actually says

A loud debate with quieter data

Ask ten people whether AI will replace office staff and you'll get two answers. One camp says the order desk, AP and customer service will be run by software within a few years. The other says nothing has changed and this is hype. If you run a 40 to 400 person manufacturer or distributor, neither answer helps you plan next year's headcount.

The public data is better than the debate. In the last few months the government released its ten-year job projections, the New York Fed surveyed manufacturers about AI, Stanford updated its payroll study, and the Fed's Beige Book asked businesses in all 12 districts what they're seeing. Here is what each one says, where they disagree, and what it means for an office your size.

What the government projects for office jobs

Every two years the Bureau of Labor Statistics projects jobs ten years out. The 2025 to 2035 projections, released Aug. 27, 2026, expect the economy to add 5.9 million jobs, growth of 3.5%. Office and administrative support is the exception. BLS projects it to fall 4.0% and lose 752,100 jobs, the most of any major group. It says AI-powered automation tools are likely to reduce demand for several office jobs.

The list of fastest-declining occupations reads like an office org chart.

Projected change in jobs, 2025 to 2035

U.S. Bureau of Labor Statistics, Table 1.5, fastest declining occupations. These are projections, not counts.

1. Word processors and typists-34.4%
2. Data entry keyers-25.5%
3. Order clerks-17.5%
4. Payroll and timekeeping clerks-15.9%
5. File clerks-15.8%
6. All occupations+3.5%

Source: BLS, Fastest declining occupations, 2025 and projected 2035.

In plain numbers, BLS expects data entry keyer jobs to go from about 131,800 to 98,200, and order clerk jobs from about 78,900 to 65,100. Those are big drops. They're also spread over ten years, across the whole country, so a 17.5% decline works out to less than 2% a year.

BLS is careful here too. The same release says AI is expected to support demand for some jobs while productivity gains may dampen demand for others. A projection is a direction, not a forecast for your company.

What manufacturers are actually doing

The New York Fed asks manufacturers and service firms in New York and northern New Jersey about AI each summer. Its Sept. 1, 2026 write-up carries the title "Businesses are using AI to transform work, not cut jobs," and the numbers back it up:

• Use is up fast. 51% of manufacturers said they used AI, up from 26% a year earlier and 16% in 2024. Among service firms it was 61%.
• Spending is small. More than 90% of manufacturers called their AI investment minimal to modest.
• Few people use it yet. Among manufacturers using AI, the median share of workers using it was 7%.
• No AI layoffs at manufacturers. No manufacturers reported laying off workers because of AI, this year or last. Among service firms, 4% did.
• Some hiring slowed. About 15% of service firms said they hired fewer people than they would have without AI. A handful of manufacturers said the same.
• Retraining is common. More than 20% of manufacturers and just over a third of service firms said they were retraining workers.

One detail stands out for anyone running a back office: a firm in the survey put AI on accounts payable and receivable with a person in the loop to oversee it. That's the pattern that works: software does the reading and typing, a person checks and approves.

The limits matter. This is one region, not Texas, and the post doesn't give a sample size. The Fed itself says the patterns could shift as use matures.

Where the change shows up first: hiring, not firing

Stanford's Digital Economy Lab tracks payroll data from ADP, the payroll company. In the August 2026 update of its "Canaries in the Coal Mine" paper, Erik Brynjolfsson and his co-authors found that employment of workers aged 22 to 25 in the jobs most exposed to AI is 19% below where it would be if it had kept pace with less-exposed peers. Experienced workers show no such gap.

Two more findings matter for an office manager. First, the gap comes mainly from hiring fewer young workers, not from letting people go. Second, the drop is concentrated in jobs where AI mostly does the task for the person. Where AI mostly helps the person do the task, employment is flat or rising. The authors also say they find no evidence of widespread job loss across the economy, and they list caveats: the pattern weakens when they control for education, and some trends started before generative AI.

The Fed's October 2026 Beige Book, published Oct. 14, points the same way. Five of the 12 districts reported that some firms had cut staff through attrition, delayed hiring, or expected to create fewer jobs after getting more done with AI. In the Dallas district report, an accounting contact said some large accounting firms were delaying new hiring contracts for entry-level staff, partly because of AI. A staffing firm said AI agents ran more than 180 hours of interviews in two days, and it hired more recruiters to review that work and make the calls.

So far, AI shows up as the job that doesn't get posted, not the person who gets let go.

The other side: layoffs that name AI, and economists who see nothing yet

It would be wrong to say AI isn't costing anyone a job. Challenger, Gray & Christmas counts announced layoffs and the reason employers give. In its Oct. 1, 2026 report, AI was cited in 120,136 announced job cuts so far this year, about 21% of all cuts, making it the leading reason this year. Those are real people. Keep two things in mind: these are plans announced by employers, often large ones, and the reason is the employer's own label. Technology companies alone account for 165,925 of this year's cuts.

On the other side, the Yale Budget Lab tracks the whole labor market each month. As of Oct. 14, 2026, it says the mix of jobs isn't yet changing in ways that clearly line up with AI, and that measures of AI use show no link to changes in employment or unemployment. In an earlier explainer it adds two warnings: office and administrative support is the most common job group among the most exposed, and the picture could change fast.

Both can be true. Big companies are announcing cuts and pointing to AI, while across the whole economy the effect is still too small to see clearly. For a mid-size company, the useful signal is the narrower one: entry-level, task-heavy office work is where change starts.

What this means for a 40 to 400 person office

Most office jobs at a manufacturer or distributor aren't one task. An order entry clerk types orders, but also answers "where's my order?" calls, catches the PO with the wrong price, and knows which customer always means the black version. The data says the typing part is what shrinks.

Here is how we'd plan around it:

1. Split each job into tasks. List what the person does in a week and roughly how long each part takes. Typing from emails and PDFs is the part to automate. Judgment, phone calls and fixing problems aren't.
2. Plan around growth and turnover, not layoffs. The pattern in the data is roles not refilled and new work absorbed. If sales grow 20% and the desk handles it with the same people, that's the win.
3. Keep a person approving. The firms in the Fed survey that put AI on AP kept a person in the loop. Software drafts the order or the invoice; someone you trust approves it.
4. Retrain on purpose. More than 20% of the surveyed manufacturers were already doing it. The person who typed orders is the best person to check the software's work, because they know the customers.
5. Measure items per person. Orders or invoices handled per person per day tells you if it's working, better than any vendor's claim.

Talk to your team before the rumor does

People in your office read the same headlines you do. If you bring in automation without saying what it's for, they'll fill in the blank themselves. Be specific: which tasks go away, what stays, and that the goal is handling growth without burning people out. We wrote more about that gap in how bosses and staff see AI rollouts differently.

If you're deciding right now whether to post a job, the hire-or-automate calculator runs the numbers on your own salary and workload, and before you hire covers what to check first.