whirrwork
Insights

Robots and jobs: what the evidence actually says, and how to roll out robots without wrecking trust

What peer-reviewed studies from the US, Germany, Spain and 17 countries found about robots and employment, why they seem to disagree, and a practical playbook for introducing robots to your team honestly.

By The Whirrwork team7 min read
  • automation and jobs
  • labour economics
  • change management
  • workforce
  • unions
  • robot rollout

The short version

  • The best studies don't agree on a single number, and that's not a cop-out: they're measuring different things. Areas exposed to robots can lose jobs and wages (US evidence). Firms that adopt robots often grow and hire, while their competitors shrink (Spain). Whole countries see productivity gains with little change in total hours, but fewer low-skilled jobs (17-country study).
  • The cost often lands on people who never get hired rather than people who get fired, and on job quality: one study links robotisation to lower sense of meaning and autonomy at work.
  • Our view: whether a robot rollout builds or destroys trust is mostly in your control. Say the word "robot", explain early, redeploy before you reduce, watch the pace of the remaining jobs, and put commitments in writing.

We sell nothing but information, and we'd like more people to use robots where they make sense. So we've tried hard to read this research without squinting.

What the studies found

StudyWhere and whenWhat it found
Acemoglu & Restrepo, Journal of Political Economy, 2020US local labour markets, 1990–2007One more robot per 1,000 workers reduced the employment-to-population ratio by about 0.2 percentage points and wages by about 0.42% (JPE)
Dauth, Findeisen, Suedekum & Woessner, JEEA, 2021GermanyRobots cut manufacturing jobs, but losses were offset by new service jobs; the adjustment fell mainly on young people entering the labour market, while incumbents often stayed and moved into new roles (JEEA)
Graetz & Michaels, Review of Economics and Statistics, 201817 countries, 1993–2007Robots raised annual labour productivity growth by about 0.36 percentage points, with no significant effect on total hours but a lower share of low-skilled workers (paper)
Koch, Manuylov & Smolka, Economic Journal, 2021Spanish manufacturing firms, 1990–2016Robot adopters grew output 20–25% and employment about 10% within four years; non-adopting competitors lost jobs (VoxEU summary)
Nikolova, Cnossen & Nikolaev, Research Policy, 2024Workers in 14 industries, 20 European countries, 2005–2021Doubling robotisation was associated with about 0.9% lower sense of meaningful work and 1% lower autonomy (Research Policy)

Why they seem to disagree (they mostly don't)

They zoom in at different levels. The US study looks at whole local labour markets. The Spanish study looks inside firms. Both can be true: a firm that automates grows and hires, while its slower rivals shed workers, and the town as a whole ends up with fewer jobs than it would have had.

"Losing jobs" isn't the same as "firing people". The German evidence suggests much of the adjustment happened through fewer young people being hired into manufacturing, while existing staff were kept on and moved into new roles. That's gentler for the people already there, and invisible in a layoff count.

The skill mix changes. The 17-country study found robots shifted employment away from low-skilled workers even where total hours held steady. Averages hide who pays.

Old robots aren't new robots. Most of this research covers industrial arms in factories through the 2000s and 2010s. Mobile robots in warehouses, drones in fields and robotaxis on streets are newer, and the research hasn't caught up. Anyone telling you with certainty what they'll do to jobs is guessing. (Including us, which is why the next section is labelled.)

The newer evidence: pace and quality

Two things the job-count debate tends to miss:

The jobs left behind can get harder. Reveal's 2020 investigation reported that serious injury rates at Amazon's robotic warehouses were more than 50% higher than at its non-robotic ones from 2016 to 2019, linked to faster pick rates (The Verge). The robot didn't hurt anyone directly. The pace did.

Work can feel less like yours. The 2024 Research Policy study above links robotisation to lower meaning and autonomy, especially for workers doing routine tasks.

New jobs appear, but you have to staff them. Amazon says its newest robotics-heavy fulfilment centre needs 30% more people in reliability, maintenance and engineering roles (Amazon). In agriculture, the number of FAA-approved drone spraying operators in the US rose 58% in a year to 1,710 (ASDC). These are real jobs, but not automatically for the people whose tasks were automated.

A case study in how not to talk about it

In October 2025 the New York Times reported on internal Amazon documents suggesting the company hoped automation would let it avoid hiring more than 600,000 US workers by 2033. The same reporting said the company had considered avoiding words like "automation" and "AI" in favour of terms like "advanced technology", and calling robots that work alongside people "cobots". Amazon said the documents reflected one team's view, not its hiring strategy, and that executives weren't told to avoid certain terms (The Verge, summarising the NYT).

Whatever the full story, the lesson for every other employer is clear. Your team will find out what the robot is for. Nobody in history has been reassured by the phrase "advanced technology".

A case study in putting it in writing

The January 2025 deal between US East and Gulf Coast dockworkers (ILA) and port employers (USMX) allowed semi-automated equipment while ruling out fully automated terminals, and it set up structures like a joint New Technology Committee and staffing commitments tied to new equipment (CNBC; USMX-ILA master contract). You don't need a union to borrow the idea: decide how technology decisions get made, and write it down before the first robot arrives.

Whirrwork's rollout playbook

This is our view, drawn from the evidence above and from watching a lot of robot projects. It's not a study. It's in the spirit of MIT's Work of the Future task force, which concluded that technology both replaces and creates work, and that institutions and employers decide how the gains are shared (MIT, 2020).

  1. Say "robot". Explain what it will do, in plain words, before the delivery truck arrives.
  2. Start with the jobs people hate. Heavy, repetitive, dangerous or dirty tasks. If your team would happily give the task away, you're starting in the right place. (This is also where robots tend to work best.)
  3. Tell people what it means for their job, early. Ambiguity is worse than bad news. If roles will change, say how.
  4. Redeploy before you reduce. Natural attrition, retraining and moving people into the new technical roles. The German evidence suggests firms can absorb automation largely by keeping incumbents and hiring fewer new entrants; that's a choice you can make on purpose.
  5. Pay for new skills. If someone learns to supervise or maintain the robots, that's a more valuable job. Treat it like one.
  6. Watch the pace. Measure injury rates, strain and turnover before and after, not just throughput. If the robot speeds up the line, check what it does to the people on it.
  7. Give people a say in the details. Routes, schedules, where the robot parks, what happens when it gets stuck. The people on the floor know where the wet patch is.
  8. Put commitments in writing. Even a one-page note: what the robot is for, what won't change, how concerns get raised.
  9. Share the gains. If productivity rises, some of it should show up for the people who made it work.

Bot or Not?

A robot is likely to build trust when…It's likely to damage trust when…
It takes over a task people dislike or find dangerousIt's introduced as a surprise
People hear about it from their manager firstPeople hear about it from a delivery driver
Roles change with training and better payThe remaining jobs just get faster
There's a written planThe plan is "we'll see"

And the Ladder Test applies to people decisions too: if the job is small and occasional, a robot won't save enough to be worth the upheaval.

What to ask a supplier

  1. What training do you provide for operators and maintenance staff, and how many hours?
  2. What do the remaining human tasks look like at your reference sites?
  3. Can we speak to staff, not just managers, at a reference site?
  4. How do you measure ergonomic impact?

Then see our due diligence checklist and hidden costs pieces.

Sources

All accessed October 2026.

More Insights