Goldman Sachs Puts AI Job Displacement at 9% of U.S. Workers: What It Means for 2027 Hiring Plans

AI job displacement could reach something like 9% of U.S. workers over the next decade, which works out to around 15 million people, and Goldman thinks slower hiring could shave roughly another 3% off employment on top of that. Their old number was 6 to 7%, so yes, it went up.

When I first saw the headline, I assumed it meant a wave of layoffs sometime next year, and I suspect a lot of hiring managers had the same reaction, but that’s honestly not what the note says once you get past the first paragraph of the coverage.

So if you’re the person who has to sign off on 2027 headcount, this is what the forecast does and doesn’t tell you, and what I’d actually change in the plan because of it.

Key Takeaways

• It’s a ten-year number, and it counts people who change jobs, not people left unemployed.
• The extra 3% from slower hiring is the bit you’ll likely feel first.
• Displacement tends to show up in recessions, less so in good years.
• Go through your existing roles before you commit to a 2027 headcount figure.
• Retraining people you already have is usually cheaper than replacing them.

What Did Goldman Sachs Actually Say?

The note is by Joseph Briggs, one of Goldman’s economists, and it came out on October 22.

The whole thing hangs on an assumption that AI eventually makes workers in the U.S. and similar economies about 15% more productive, and then he goes back through older productivity jumps to see how many people got pushed out of their jobs each time and scales that forward.

There are a couple of things in there that matter more than the 9% itself:

• One is that it’s gross displacement, so it counts everyone who leaves their current role, including the (probably large) group who land somewhere else fairly quickly.
• Another is the timeline, which is about a decade, not a year or two.
• And the third, which I think is the most important one for planning, is that he doesn’t expect it to happen evenly. Fortune’s write-up of the note quotes him saying AI-related disruption could amplify the business cycle, and in plain English that means the cuts mostly come when the economy is already going the wrong way.

It’s also worth remembering how early all of this is. Only about 23% of U.S. establishments have AI built into their regular workflows, and AI-attributed layoffs actually hit their high point in the spring and have come down since then, which isn’t what you’d guess from LinkedIn.

Goldman isn’t way out on its own here either. If you line up the bigger AI job replacement forecasts side by side, they mostly sit in the same rough range:

So for a hiring plan I’d spend less time on “how many” and more on “when.”

Why Does the 3% Hiring Drag Matter More for 2027?

Goldman actually splits the impact into two pieces. There’s displacement, people losing the job they’ve got. And then there’s the hiring side, which is just jobs that never get opened in the first place, and that one’s much harder to notice because there’s no announcement and no headline – the req just quietly doesn’t happen.

According to the note, job growth in the industries most exposed to AI has run about 10,000 below its long-run trend over the past three months, and tech, management consulting, and graphic design are among the areas feeling it first.

In a real company, it tends to look pretty boring. Someone on the support team hands in their notice and, instead of the backfill getting approved as it would have in 2023, finance comes back and asks if the AI tools can pick up the slack for a quarter or so.

Entry-level postings get fewer, since drafting and first-pass work is exactly what people hand to a model first. And the roles that do get approved drift more senior, because what teams want now is someone who can look at AI output and tell what’s wrong with it.

Apollo’s list of AI-replacing jobs statistics has a lot of the hiring and layoff data behind this in one place, and I’d honestly read it with your current open reqs pulled up in another tab, just to see which of them look like the ones already slowing down elsewhere.

Which Signals Should You Watch Before Locking 2027 Headcount?

Any forecast is kind of useless until you know what would tell you it’s wrong. Briggs actually suggests using AI-related layoffs as a real-time indicator, his reasoning being that job losses move employment more than hiring slowdowns do.

I wouldn’t overcomplicate this. Four things, looked at once a month, is plenty:

• AI-cited layoffs. Challenger, Gray & Christmas data has AI as the most cited reason for announced cuts so far in 2026, about 21% of them, and the 2026 layoff statistics page on this site breaks it out by industry if you want to see your own sector.
• Payroll growth in the exposed industries compared with their own trend, rather than the national headline, which smooths over too much.
• Your own adoption. If your teams are well beyond that 23% figure, expect your hiring needs to change before the wider market’s do.
• Signs of a recession, since that’s when displacement has historically bunched up, and so it’s the one that should actually change your plan.

Underneath nearly all of that is the monthly jobs data, and the trouble is how quickly it gets spun once it’s out, the same soft month described as the start of an AI jobs crisis by one commentator and as evidence that everyone overreacted by the next. Mostly you just want the number, the revisions to the previous month, and the industry detail.

Unbiased Headlines is one I’d add to the routine. It’s an independent news organization that writes its own original stories from 110+ sources, and its jobs report news coverage shows the sources behind every story, which helps when you need to trace a figure back before it goes into a board deck.

I’d read Unbiased Headlines alongside the actual Bureau of Labor Statistics release, which takes maybe five minutes a month.

How Should You Sort Your Existing Roles?

Before anyone puts a number on 2027, I’d go through the jobs you already have. A rough three-way split tends to actually get used, whereas a big risk matrix with forty rows gets built once and then nobody opens it again after the first quarter.

Exposed Roles

These are the jobs where AI can already handle most of the core work, things like routine content, first-line support, data entry, and some of the junior analysis. I’d plan on hiring fewer of them next year and on moving people out of them where it makes sense.

Augmented Roles

Most technical work ends up here, so software, data, cloud, security, and so on. AI speeds these people up, but somebody still has to decide what’s correct, and demand for them can actually go up because more output means more systems to keep running.

The skills inside the job keep shifting though, and the World Economic Forum’s Future of Jobs Report found employers expect around 39% of workers’ core skills to change by 2030, which is a lot of change for one role to absorb.

Insulated Roles

This is hands-on work, regulated work, or work that depends on relationships, so field engineers, clinical technologists, supply chain roles tied to physical sites, and senior client leads. For these, I’d keep hiring the way you normally would.

It works better if the sort happens as part of your normal IT workforce planning rather than as a separate project, and if finance and the team leads are in the room and not just HR, since audits that only HR owns tend to quietly fizzle out.

How Do You Build a 2027 Hiring Plan With Recession Triggers?

A single headcount number for 2027 probably won’t survive one bad quarter. Goldman’s own view is that the transition should be manageable if the economy stays healthy and a lot harder if it doesn’t, so it makes sense to plan for both cases.

Imagine you’re running an engineering and IT group of about 60 people. I’d write three versions of next year and agree in advance what moves you from one to the next.

Steady Growth

Payrolls are close to trend, and layoffs are flat. You hire the augmented and insulated roles you’d planned, and you’re just a bit slower to backfill exposed ones.

Soft Patch

A couple of weak payroll months back-to-back, and AI-cited cuts going up. At that point I’d freeze the exposed roles, move some budget over to contract talent, and push retraining along faster.

Recession

Obvious contraction, layoffs everywhere. Protect the augmented roles first, stop net new hiring, but keep a short list of good candidates warm so you’re not starting from nothing when things pick up again.

The reason to write the triggers down now is that plans made in the middle of a downturn almost always overcorrect, and letting go of technical people you’ll need back in 18 months costs far more than it saves.

Contract and contract-to-hire setups make it much easier to slide between those versions, and a tech staffing partner who watches the market can usually warn you which skills are about to get scarce, so you’re not scrambling for data engineers six months into a soft patch.

Should You Reskill Before You Replace?

When an exposed role shrinks, what usually happens is a layoff, and then a few months later a new posting for whatever the newer skill is. It’s normally the slowest and most expensive way to do it.

The people you already have know your systems and your customers and all the strange undocumented workarounds nobody ever wrote down. Moving a support specialist into running AI-assisted workflows, or a junior analyst into data quality, might take a few months, while hiring and ramping someone new into the same seat can take the best part of a year.

What tends to make this work is fairly practical. Start with the exposed roles that are closest to augmented ones, because those moves are the easiest. Fund the training out of the hiring budget, since every req you don’t open frees up money for it. If you’re buying software for this, the enterprise learning and development platforms that connect to your HRIS will at least show you who’s ready to move.

There’s a real business case behind this and not only a goodwill one. A Coruzant Technologies piece on AI leadership makes the argument that the net effect of AI on jobs depends a lot on how organizations adapt, and retraining is a large part of how they do.

What Does This Mean If You Are Job Hunting in 2027?

From the candidate side, the same forecast reads a bit differently. If you’re looking for work next year, you’ll probably see fewer junior postings in the exposed fields and more people going after each one.

The best bet is to move toward augmented work, showing that you can use AI tools to get more done and, maybe more importantly, that you can catch what they get wrong.

It’s a fair bet that judgment will count for more than raw speed in 2027 interviews, and most of what’s been written on AI impact on jobs points to hybrid roles, technical skill mixed with a traditional function, as where the growth is.

And if a layoff does happen, it helps to remember what the 9% is counting. For most people, it describes moving to another job, not the end of a career.

Conclusion

AI job displacement at 9% of U.S. workers is a ten-year forecast, but it still deserves a place in your 2027 plan. What you’ll most likely notice first is the slower hiring, the 3%, with bigger cuts coming later and mainly if the economy turns.

• Treat the 9% as gradual and tied to the business cycle, and the 3% as something already happening.
• Sort your roles into exposed, augmented, and insulated before you set a headcount figure.
• Agree on recession triggers now, and retrain people before you replace them.

If you get those three done, the next forecast that comes out shouldn’t send you back to a blank spreadsheet.

Frequently Asked Questions