What Do You Owe the People the Robots Replace?

Automation just became contested terrain. The companies that win it will have a labor strategy before they need one.

EXECUTIVE SUMMARY

For twenty years, "what do we owe displaced workers?" was a question for economists and op-ed pages. In one week this July it became a question for your general counsel and your board. When automation reshapes jobs, the companies that treat the transition as a negotiated deal — not a severance line item — are buying something their competitors can't: speed, labor peace, and the right to automate without a fight.

// Hyundai workers staged the first strike in automotive history over humanoid robot deployment — a partial walkout in mid-July after the company, which owns Boston Dynamics, announced humanoids for its factory floors. BMW and GM are running similar experiments. Labor analysts expect this fight to spread.

// Twenty-six Meta employees sued over AI-assisted layoff selection, alleging that activity dashboards, keystroke data, and AI token-usage metrics used to rank 8,000 layoff candidates structurally penalized workers on medical and parental leave. Meta says people, not AI, made the calls. The court will decide; your playbook shouldn't wait.

// AI is now the leading stated reason for U.S. job cuts — four consecutive months — 101,743 AI-attributed cuts through June 2026, 31% of June's total.

// History says the deal is buyable. The 1960 longshore mechanization agreement, the UAW's 2023 contract, and the ILA's 2025 port deal all traded worker security for automation rights — and the employers got what they paid for.

The leaders who win this won't be the ones who automated fastest. They'll be the ones whose workforce let them.

// The Deep Dive

So in the early 2000s I was Editor-in-Chief of LinuxWorld, and the question I got asked more than any other — at trade shows, in reader mail, once by a guy who cornered me at a booth in the Javits Center — was some version of: "Is this thing going to take my job?"

The "thing" was Linux. The people asking were Solaris admins, NetWare engineers, folks who had spent a decade getting certified on proprietary systems that open source was about to steamroll. And I'd love to tell you I gave them a wise answer, but mostly I said what everybody said: it'll create more jobs than it kills. Learn the new thing. You'll be fine.

Here's what actually happened. Open source did destroy those jobs. The shrink-wrapped Unix business collapsed. The NetWare certification that made a guy in Harrisburg the highest-paid person in his IT department became a punchline. And open source also created vastly bigger jobs — the Linux admins became the devops engineers became the cloud architects. I rode that wave myself: I ran marketing and community at Cloud.com, we got acquired by Citrix, and the whole cloud economy that employed millions was built on the software those worried admins were afraid of.

Both things were true. The aggregate was a boom. The transition was brutal — and it was brutal to specific people. The Solaris admin who retrained in 2003 caught the wave. The one who waited until 2008 mostly didn't. Same skills, same talent, same zip code. The variable was whether anyone — an employer, mostly — invested in moving them across the gap.

I think about Amara's Law a lot here: we overestimate a technology's impact in the short run and underestimate it in the long run. It has an unwritten labor clause. We also overestimate how fast the robots arrive — and underestimate what we owe the people standing where they land.

That clause came due the week of July 14.

Three things happened in a single news cycle that, taken together, moved "what do we owe displaced workers" from the ethics panel to the risk register.

First, Hyundai workers in South Korea staged a partial strike over the company's plan to deploy humanoid robots on its factory floors — the first labor action in automotive history aimed specifically at humanoids. Hyundai isn't a bystander in robotics; it owns Boston Dynamics and has been showing off humanoid capabilities all year. The strike bundled the robot grievance with wage and bonus disputes, which is exactly how these fights work: automation anxiety doesn't show up as a philosophy seminar, it shows up as leverage at the bargaining table. Analysts quoted in the piece expect similar resistance at other automakers — BMW and GM are running their own humanoid pilots.

Second, 26 Meta employees sued the company in federal court in Oakland over how it selected roughly 8,000 people — about 10% of its workforce — for layoffs announced in May. The complaint alleges Meta used keystroke monitoring, activity-tracking dashboards, AI token-usage metrics, and algorithmically assisted rankings to score employees, and that those scores "cannot be accumulated by an employee who is on protected medical or family leave." Eight of the plaintiffs were on maternity leave. The claims run through the FMLA, the ADA, and Title VII disparate impact. Meta's response: "Workforce management decisions were made by people, not AI."

Maybe so. But notice what the lawsuit is really about. It's not a complaint that layoffs happened. It's a complaint about how the selection worked — and whether anyone owed the selected an explainable, humanly defensible answer. That's a labor-strategy question wearing a litigation costume.

Third — and this is the one executives keep misreading — Anthropic published research on what happens when you hand Claude a robot. The headline finding cuts both ways. On one hand, current models are nowhere near general-purpose robot brains: direct-control manipulation success ran 0–5.5%, no model could stand a collapsed humanoid up, and real-time control needs roughly two orders of magnitude more speed than today's inference delivers. On the other hand, the researchers note that "a general-purpose chat model with no robotics training can already, on a good run, write and download its own tools to slowly walk a quadruped through a maze or pick a plate off a counter" — and when they changed the interface, giving the model higher-level controls and pretrained policies, success on one real-robot task suite jumped from 6% to 32%.

Read that as an executive, not a roboticist. The robots aren't ready to take the night shift. But capability is a function of scaffolding, and scaffolding improves on software timelines, not hardware ones. Amara's Law, again: the Hyundai strikers are probably early relative to the technology — and probably right about the direction. Which means you have time to build a labor strategy. Less time than you think. More than the panicked headlines suggest.

And the displacement pressure isn't hypothetical while we wait for the humanoids. AI has been the leading stated reason for U.S. job cuts for four consecutive months, per Challenger, Gray & Christmas — 101,743 AI-attributed cuts in the first half of 2026, with tech alone shedding 139,156 jobs year to date, up 83% from last year.

Here's the part almost nobody in an AI strategy meeting knows: American industry has already negotiated the automation transition, several times, on the record. The deals are public. The results are measurable.

The 1960 longshore deal. When containerization threatened to gut waterfront work, the ILWU under Harry Bridges and the Pacific Maritime Association signed the Mechanization and Modernization Agreement. The trade was explicit: employers got the freedom to introduce labor-saving machinery and eliminate restrictive work rules, and in exchange the registered workforce could not be laid off — early retirement was funded, and machines were to be introduced "to lighten the burden of hard and hazardous work." Employers paid into the transition instead of fighting through it. The result: the Pacific ports containerized fast, decades ahead of what a war of attrition would have allowed. The deal wasn't perfect — protections tiered off for partially registered "B men," resentments built, and a 134-day strike still erupted in 1971 over containerization's second-order effects. But the core bargain held: management bought the right to automate by making the current workforce whole.

The UAW, 2023. The stand-up strike ended with the union winning a 25% base wage increase through April 2028 and — a historic first — the right to strike over plant closures and product commitments. Stellantis reopened the idled Belvidere plant, restoring 1,200 jobs and adding over 1,000 more at a planned battery facility. Shawn Fain's line: "Not only did we not lose those 5,000 jobs, we turned it all the way around." The mechanism to notice isn't the raise. It's that job security became enforceable — closures and investment decisions moved inside the contract, where they can be negotiated rather than announced.

The ports, again, 2024–2025. The ILA's October 2024 strike shut Atlantic and Gulf ports for three days over exactly the question in this edition's title — semi-automated cranes. The settlement, ratified by nearly 99% of members in February 2025, delivered a 62% wage increase over six years plus protections on automation, with labor peace locked in through September 2030. Employers got what they actually needed — predictability and the ability to modernize under agreed conditions — and they paid for it in wages and guarantees.

Three deals, three eras, one pattern. In each case the company (or employer group) that proactively paid for the transition got to automate faster than the ones that tried to impose it. Hyundai is currently running the control group for us: impose first, negotiate after, lose production days in between. The strike is what not having a deal costs.

You don't need a union on the other side of the table to learn from this. The pattern is the point: the transition is purchasable, the price is knowable, and buying it early is cheaper than buying it after the walkout or the lawsuit.

This is the claim that gets waved around to justify doing nothing, so let's run it honestly.

The bull case is real. The World Economic Forum's Future of Jobs survey projects 170 million new roles created by 2030 against 92 million displaced — a net gain of 78 million jobs. Even in this year's grim cut announcements, hiring plans are up 10% over the first half of 2025. And history genuinely is on this side: the web destroyed the travel-agent economy and created something far larger; open source destroyed the proprietary Unix business and created the cloud. I lived the second one. The optimists aren't making it up.

The bear case is also real — and it's more specific. Stanford's Digital Economy Lab, using payroll records from ADP, found that early-career workers aged 22–25 in the most AI-exposed occupations have already seen a 16% relative decline in employment since generative AI adoption took off — while older workers in the same occupations held steady. The declines concentrate precisely where AI automates work rather than augments it. The canaries are already quiet. And per the same WEF survey the optimists cite, 41% of employers plan workforce reductions due to automation.

Here's how I score it: both cases are true at different altitudes. The bull case is about the economy over a decade. The bear case is about your employees over the next eight quarters. A net-positive 2030 is cold comfort to the 92 million on the wrong side of the netting — and "the economy will create new jobs" was exactly true and exactly useless to the NetWare engineer in 2004. Nations experience aggregates. People experience transitions. Companies are where the two collide, which is why this is a strategy question and not a debate topic.

Set ethics aside entirely — I don't, but let's pretend you do — and the numbers still push the same direction.

Start with what churn-and-replace actually costs. Companies end up rehiring about 5.3% of the very people they laid off, per Visier's analysis of 2.4 million employee records — in finance it's 7.5%, and the boomerangs come back at a 3% salary premium. Recruiting and ramping a replacement runs 50–200% of the role's annual salary. And the people you keep watch how you treat the people you cut: 74% of layoff survivors report a productivity decline afterward. Every one of those numbers is a tax on the "just cut and move on" strategy — before you get to the litigation exposure Meta is now modeling for everyone, or the production days Hyundai is losing this month.

Now the other column. 77% of employers say they plan to reskill and upskill their existing workforce by 2030 — and 41% plan automation-driven reductions. Most large companies will do both. The strategic question is sequencing: redeploy-then-reduce companies get to keep institutional knowledge, harvest the rehire premium they'd otherwise pay, and — this is the underrated part — automate with their workforce's consent instead of against its resistance. The ports proved consent is worth paying for. Severance buys silence. A transition buys speed.

That's the competitive advantage hiding in this week's ugly headlines. In a market where AI is the leading stated reason for job cuts, the employer with a credible, funded, public answer to "what happens to me?" recruits better, retains better, and deploys automation faster than the one whose answer is a rumor and a package. Labor strategy is becoming what safety records became in manufacturing: a thing you compete on.

Misstep 1: Waiting for the fight to start the negotiation. Hyundai announced the robots, then discovered what its workforce thought about them. The 1960 longshore employers did it in the other order and containerized the coast. By the time you're negotiating during a strike — or discovery — you've already paid the premium price for the same deal.

Misstep 2: Governing by the net number. "AI creates more jobs than it destroys" may well be true and it staffs exactly none of your Tuesday shifts. The WEF's net +78 million contains 92 million displacements, and the Stanford data shows the losses land on specific, identifiable cohorts. Averages are for economists. You employ individuals.

Misstep 3: Treating severance as the whole answer. A package closes the legal file and opens three others: the 50–200% replacement cost when you rehire the skill next year, the 5.3% boomerang rate that proves you cut capability you still needed, and the productivity slump among the 74% of survivors who now update their résumés on company time. Severance is a door charge, not a strategy.

Misstep 4: Planning on the robot's timeline instead of the workforce's. The Anthropic research says general-purpose physical automation is years out — and that capability jumps an order of magnitude when the scaffolding changes. Companies keep pairing aggressive automation pilots with zero redeployment planning, betting the technology arrives slowly enough to figure out the people part later. Amara's Law says you'll be wrong in both directions at once.

// Key Takeaways

Write your automation-labor deal before you need one. Every durable precedent — 1960, 2023, 2025 — was cheapest for the employer who came to the table before the crisis. Decide now what you'll guarantee (redeployment first? funded retraining? no algorithmic-only selection?) and put it in writing your workforce can read.

Budget the transition as a line item, not a rounding error. The longshore employers funded early retirement; Stellantis funded Belvidere; port operators funded a 62% raise. Price your version — a fixed percentage of every automation initiative's budget reserved for redeployment and reskilling — and make it as non-negotiable as the compute spend.

Track your own canaries. Stanford found the damage concentrating in early-career, high-exposure roles while everyone watched the averages. Instrument your workforce the same way: entry-level hiring ratios, AI-exposure by role, attrition in automating functions — reviewed quarterly, at the same table as the AI roadmap.

Sequence redeployment before reduction. Post every AI-adjacent opening internally first, fund the bridge training, and only then cut what's genuinely surplus. The rehire and replacement math says this is cheaper, and it's the difference between a workforce that adopts your automation and one that organizes against it.

So here's the playbook I'd take into your next planning cycle. One: an automation impact map — every role, its AI exposure, its 24-month outlook, its redeployment path. Two: a named transition fund, sized as a percentage of automation capex, that pays for retraining and bridge periods. Three: a selection standard for any AI-influenced workforce decision that your general counsel has read and would defend in Oakland. Four: the canary dashboard from the takeaways. Five: if any part of your workforce is organized — or could be — open the technology conversation this year, on your initiative, while it's still a conversation.

The robots are coming either way. The only open question is whether your people meet them as partners in a deal you wrote together — or as plaintiffs, picketers, and the 74% who stayed but stopped trying. The companies that answered "what do we owe them?" with a number and a plan got to automate. The ones that answered with silence got a strike. That's the whole lesson, sixty-six years running.