AI's Net Job Drag Fell to 11,000 a Month. Construction Is Hiding the Truth

The number that says AI is hurting the job market less than it was is the number you should trust the least right now.

According to Fortune's June 1 reporting on new Goldman Sachs research, AI's net drag on US monthly payroll growth has eased to roughly 11,000 jobs, down from about 16,000 earlier in the year. On its face, that reads like the worst is passing. It isn't. The improvement comes almost entirely from hard hats, not from white-collar work recovering.

If you run a company and you set your 2026 headcount plan off the headline, you'll plan for a labor market that is stabilizing. The data underneath says something more uncomfortable, and it changes where you should spend your AI budget.

What Goldman's Numbers Actually Say

Goldman's economists pulled apart AI's effect on the labor market into two opposing forces. One is substitution: roles where AI does the work instead of a person. The other is augmentation: roles where AI makes a person more productive and the company keeps hiring.

Over the past year, substitution wiped out around 25,000 US jobs a month. Augmentation added back about 9,000. Net it out and you get the roughly 16,000-a-month drag that ran through most of the year. The work was led by Goldman economist Elsie Peng, who scored occupations on a displacement measure paired with an international complementarity index, then weighed the Jevons paradox: when AI makes a task cheaper, total demand for the output can rise and pull some jobs back.

That decomposition is the part worth keeping. The single net number hides it. Two companies can post the same flat headcount and be in completely different positions, one quietly bleeding substituted roles, the other adding augmented ones. The net tells you nothing about which one you are.

Key Facts

• AI substitution cut roughly 25,000 US jobs a month over the past year, while augmentation added back about 9,000, for a net drag near 16,000 (Goldman Sachs research, via Fortune, June 1, 2026)
• Goldman's latest AI Adoption Tracker puts the net drag at about 11,000 jobs a month, but data-center construction has added 212,000 jobs since 2022, roughly 9,000 a month (Fortune, June 1, 2026)
• Goldman estimates AI has lifted US unemployment about 0.1 percentage point so far, with the rate drifting toward 4.5% in 2026 from 4.3% in January (Goldman Sachs research)

Why the Better Number Is a Mirage

Here's the catch in the improvement from 16,000 to 11,000. The gap didn't close because AI stopped substituting white-collar roles. It closed because the AI build-out is hiring construction crews.

Data-center construction has added about 212,000 jobs since 2022, and it's currently generating roughly 9,000 new positions a month. That's the same physical, on-site, judgment-heavy kind of work that sits on the augmentation side of Goldman's ledger. So the boom that is automating away claims clerks and data-entry staff is, at the same time, pouring concrete and pulling conduit wire. The construction line item is papering over the office line item.

For a CEO, the practical reading is blunt. The white-collar substitution that hit your administrative, support, and entry-level functions has not reversed. It's been netted against a construction surge that has nothing to do with your org chart. Strip the hard hats out and the office-job drag looks worse than 11,000, not better.

Substitution vs Augmentation Is the Real Signal

The most useful thing in Goldman's work isn't the macro count. It's the map of which roles fall on which side, because that map is a planning tool you can apply to your own company this quarter.

On the substitution side, the highest-risk roles are routine and rules-based: telephone operators, insurance claims clerks, bill collectors, customer service reps, and data-entry staff. On the augmentation side, the job-gaining roles share a thread the spreadsheet can't fake, physical presence, human judgment, or real interpersonal work. Goldman's examples include education workers, judges, and construction managers.

That split gives you a simple instrument. Call it the substitution-augmentation ledger. Take every function in your business and ask one question: are you deploying AI here to remove people, or to make the people you keep produce more? Score each role. The pattern that shows up in your own ledger tells you where you're harvesting a one-time cost cut and where you're compounding output.

The reason this matters beyond a tidy framework is that pure substitution keeps backfiring. We've covered how more than half of AI-driven layoffs are already boomeranging on the companies that made them in the AI babysitting-tax breakdown, where the hidden cost of supervising and correcting AI eats the savings. Goldman's augmentation number, a real and positive 9,000 jobs a month, is the quieter signal: the companies leaning into augmentation are still adding people and getting more out of them.

There's a talent-pipeline wrinkle in the data too. Goldman found that a one-standard-deviation rise in a job's exposure to AI substitution widens the wage gap between entry-level and experienced workers by about 3.3 percentage points. Entry-level hiring in professional services has already cooled sharply, which lines up with the broader collapse of the entry-level job we flagged earlier. If you stop hiring juniors because AI covers the bottom rung today, you also stop building the seniors you'll need in three years.

What This Changes for Your 2026 Headcount Plan

You don't need to react to the macro number. You need to run your own version of it. Three moves.

Move 1: Build your substitution-augmentation ledger before your next planning cycle. List your functions. Mark each as AI-substituting (headcount falling) or AI-augmenting (headcount steady or rising, output up). Don't average them into one number, the way the macro stat does. The mix is the insight. If your ledger is almost all substitution, you're optimizing for a one-time cut and you should expect the babysitting tax to show up next.

Move 2: Protect the junior rung you'd cut first. The roles AI eats earliest are the entry-level ones, and the wage data says that's exactly where the squeeze is sharpest. Decide deliberately whether to keep a junior pipeline rather than letting AI quietly delete it by attrition. Federal programs can offset the cost, which is why we walked through whether AI apprenticeships beat hiring on price.

Move 3: Stress-test your workforce plan against more than one future. Goldman's base case is a roughly 10-year adoption curve with 6 to 7 percent of workers displaced during the transition and unemployment up about 0.6 percentage point long term. That's a wide cone of outcomes. A single point forecast will be wrong; a plan that holds up across scenarios won't. We laid out that exercise in the four futures for jobs checkpoint.

The headline will keep bouncing month to month, and it will keep being noisy because construction hiring and office automation are pulling in opposite directions inside it. The signal that's stable is the one you can act on: substitute and you cut once, augment and you compound. Build the ledger that tells you which one you're actually doing.

Learn More

• Why 55% of AI Layoffs Are Backfiring on CEOs
• AI Is Breaking the Entry-Level Job
• Four Futures for Jobs by 2030: The Mid-2026 Checkpoint for CEOs
• Can Federal AI Apprenticeships Fix Your Skills Gap Cheaper Than Hiring?
• Scenario Planning

FAQ

How many jobs is AI cutting per month in 2026?

Goldman Sachs estimates AI has been a net drag of roughly 11,000 US jobs a month recently, down from about 16,000 earlier in the year. Under the surface, substitution removed around 25,000 jobs a month while augmentation added back about 9,000. The improvement in the net figure comes largely from data-center construction hiring, not from a slowdown in white-collar automation.

Is AI's effect on the job market actually getting better?

Not in the part that matters for most office employers. Goldman's net number improved mainly because data-center construction has added about 212,000 jobs since 2022, roughly 9,000 a month. That physical, on-site work offsets the continued substitution of administrative and support roles in the average, so the white-collar drag looks milder than it is.

Which jobs are most at risk from AI, and which are most protected?

Goldman's research puts routine, rules-based roles at the highest substitution risk: telephone operators, insurance claims clerks, bill collectors, customer service reps, and data-entry staff. The most protected, job-gaining roles involve physical presence, human judgment, or interpersonal work, such as education workers, judges, and construction managers.

Source: Fortune (June 1, 2026), reporting on Goldman Sachs research. Background: Goldman Sachs Insights.