In partnership with
The labour data finally has a number attached to it
For two years, AI's effect on hiring has mostly lived in survey answers and executive soundbites. Goldman Sachs has now put a figure on it. In a report published on 19 August, the bank's economists analysed employment growth across more than 800 occupations in major developed economies and found that industries with higher exposure to AI automation have posted slower job openings growth since the second half of 2022, a pattern most pronounced in the United States, Germany and Australia.
Call centre employment in the US now sits 39 percent below its historical trend line, with Canada 33 percent below and Germany 27 percent below. Software publishing, management consulting and advertising show the same divergence.
The headline finding sits inside the detail. Across France, Canada and the US, a 10 percent increase in occupational AI exposure corresponds to only a 0.1 percentage point drag on overall annual headcount growth. For entry-level workers specifically, the same 10 percent exposure increase is associated with a decline exceeding 0.6 percentage points in Australia and more than 0.2 percentage points in the US. The effect is real but narrow, concentrated in the roles that have traditionally functioned as the industry's on-ramp.
Goldman Sachs senior global economist Joseph Briggs, who co-leads the bank's global economics team, has been tracking this shift for months. Discussing the wider labour picture on the bank's Exchanges podcast, he framed the risk in terms markets understand:
"If we see some job losses pulled forward, that sets the stage for potential underperformance relative to our forecast, and that may lead the Federal Reserve to cut rates."
His broader research puts the scale of long-run displacement at roughly 9 percent of the US workforce, or around 15 million workers, over a ten-year transition, while maintaining that job creation should eventually offset most of the disruption.
Why the entry-level squeeze matters more in construction than most sectors
AEC firms already run one of the industry's tightest pyramids. Associated Builders and Contractors puts the 2026 shortfall at roughly 349,000 net new workers needed just to keep pace with demand, rising to 456,000 in 2027 as spending accelerates. The average US construction worker is 42.5 years old, only 16 percent of the workforce is under 35, and around a fifth of electricians are already over 55, a demographic profile that leaves the industry unable to simply absorb a slower entry-level pipeline the way a mature, stable sector might.
Every graduate quantity surveyor, junior planner or assistant estimator who does not get hired this year is one fewer senior estimator or project manager in a decade.
The mechanism Goldman describes, in which AI absorbs the routine, repeatable work traditionally assigned to juniors, has already been observed inside firms.
Sandy Rezendes, head of corporate learning and development at Degreed, has studied the effect directly. "The risk isn't simply that AI changes aspects of entry-level hiring," she said. "It's that it may reduce some of the foundational on-the-job learning that comes with the cognitive struggle and tasks inherent in entry-level work that people need to grow into experienced subject matter experts and future leaders."
A separate industry survey she drew on found that 30 percent of employers are shifting hiring toward mid-level talent using AI to complete tasks once given to juniors, and that 56 percent report a reduction in basic work being delegated to early-career staff because generative AI now does it faster.
For a QS team or a design office, that finding should sound familiar. Quantity take-offs, first-pass clash detection, RFI logging and early-stage cost checks are precisely the tasks AI tools already handle competently, and precisely the tasks through which junior staff historically learned to read a drawing set, price a package or spot a coordination error before it became a change order.
The adoption curve is still shallow, which is the window that matters
Goldman combined eleven separate surveys to estimate that AI adoption in major developed economies now sits at roughly 15 to 20 percent, led by France, the US, the Netherlands and the UK, while Italy, Japan and New Zealand trail. Emerging markets sit lower still, at 10 to 15 percent.
Adoption in construction specifically remains lower again. Firms report enthusiasm running well ahead of deployment, with roughly 70 percent of AEC executives believing AI will transform the industry against a low double-digit share who say they have actually adopted it at scale.
This gap is the opportunity. With the labour effect still concentrated in a narrow band of tasks and roles, firms retain room to decide deliberately how AI enters the entry-level workflow, rather than having the decision made for them by attrition and hiring freezes. A few practical shifts are worth weighing now:
Redesign junior roles around judgement and verification rather than pure production, so AI handles the first pass and people own the sign-off.
Protect a minimum volume of hands-on take-off, drafting and site documentation work for graduates, even where AI could do it faster, because the learning happens in the doing.
Track internal promotion pipelines against AI adoption by task and by headcount, so a slowdown in junior hiring does not silently become a slowdown in future senior capacity.
The UK counterpoint shows how easily this gets misread
British commentary on AI and construction employment has tended to run in the opposite direction. Lloyds research published this week found more than half of British firms reporting that AI had created jobs inside their organisations, a genuine finding but one measuring net positions among existing employees rather than the number of new doors into the industry.
The UK is simultaneously responding to youth unemployment with AI-focused boot camps, an approach that assumes entry-level roles still exist in sufficient numbers to be trained into. Goldman's data is a useful corrective here. It measures the door, not the room, and the door appears to be narrowing faster than the boot camps assume.
Takeaway
The Goldman data confirms a five-year problem is compounding beneath this week's headline. A firm relying on today's mid-level bench in 2031 needs today's graduate intake to be healthy now.
Treat entry-level task design as a strategic decision, not an efficiency afterthought. The roles most exposed to automation are also the roles that build the judgement senior staff rely on.
Watch adoption rate as closely as revenue growth. At 15 to 20 percent penetration across developed economies, the labour effect Goldman describes is still early, which means firms retain more control over the outcome than the headlines suggest.
Do not assume national data generalises. The UK's net job creation figures and the US entry-level drag are both true simultaneously, measuring different things.
Firms that get this balance wrong will not see it in this quarter's numbers. They will see it in 2031, when the promotion pipeline runs dry. Project Flux tracks exactly these structural shifts each week, connecting labour data, technology deployment and project delivery so you are not caught relying on a bench that quietly stopped refilling. Subscribe to the newsletter and get the next read before your competitors do.
Links and Stuff
4x your communication output. Same quality. No burnout.
The bottleneck isn't what you want to say — it's how long it takes to type it. Wispr Flow removes the bottleneck.
Speak naturally and get polished, send-ready text for executive summaries, client updates, board recaps, investor notes, or just the 30 Slack messages you're behind on. Flow strips filler, formats numbers and lists, and preserves your tone.
Used by teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works in every app on Mac, Windows, and iPhone.
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
1