When Nobody Can Name the Safe Jobs 🧰

Two honest playbooks, one moving ladder, and a question neither party’s grandfather had to answer. Not a political piece — a walk around a policy problem with the jerseys left at the door.

---

The Counselor’s Whiteboard 🧭

Picture a workforce counselor at an American Job Center — a composite drawn from the public record, not a real person, but you have met her. For thirty years her job has run on one reliable move: when a mill or call center closes, point to the whiteboard of growing occupations and help people walk toward it. Coding bootcamps. Medical billing. Paralegal certificates. The ladder shook, but it stood still long enough to climb.

This year, two people sit down at her desk in the same afternoon. One is fifty-five, laid off from data entry. The other is twenty-four, with a computer science degree and eleven months of junior developer experience — the exact job the whiteboard told the fifty-five-year-old to train for.

The counselor looks at her whiteboard and, for the first time in her career, does not know what to write on it.

That whiteboard — what goes on it, who pays for the walk toward it, what happens to people mid-walk when it gets erased — is what labor policy is. And right now the honest answer to “what are the safe jobs?” is that nobody can name them with confidence.

That’s not a reason to panic. It’s a reason to redesign.

---

The Facts, Before the Feelings 📋

Start with what’s actually measured, because this debate drowns in vibes.

Something real is happening at the bottom rung. A Stanford Digital Economy Lab study of millions of payroll records — “Canaries in the Coal Mine?” — found that since generative AI arrived, employment for early-career workers (ages 22–25) in the most AI-exposed occupations has fallen roughly 16 percent relative to less-exposed peers, even controlling for firm-level shocks. Older workers in the same occupations: stable or growing. The declines concentrate where AI automates rather than augments. The authors are careful — it’s early, and interest rates compete as explanations — but the canary is at least coughing.

And something real is not happening at the macro level. Nobel laureate Daron Acemoglu’s published estimate is that only about 20 percent of U.S. labor tasks are even exposed to current AI, fewer are profitably automatable, and total productivity gains over the next decade land around 0.66 percent — total, not per year. That is not “the end of work.” That’s “a decent software upgrade.”

Hold both findings at once, because both are honest: the entry-level repricing is measurable now, and the economy-wide transformation is, so far, modest. The credible range runs from “faster than any transition in history” to “merely as big as electricity.” Nobody serious predicts stasis. The argument is about the clock, not the direction — and here’s the policy problem: both parties’ labor playbooks were written for clocks that told time differently. The left’s was forged in deindustrialization — a storm that hit specific industries in specific places, so you could aim help at steel towns. The right’s was forged in the long postwar expansion, where growth reliably minted new rungs faster than old ones broke. Neither was built for a repricing that may hit tasks rather than industries, entry-level rather than end-of-career, everywhere at once, on an unknowable schedule.

Both playbooks deserve their strongest telling.

---

The Case from the Left, at Full Strength 🧣

The left’s core claim is older than any particular program: workers should not bear, alone and unaided, the cost of transitions that enrich everyone else. When a technology raises national productivity while cratering particular households, the winners owe the losers a bridge — not as charity, but as the price of the bargain.

From that root, the serious agenda follows. Modernize unemployment insurance, because the current system is quietly broken: only about 28 percent of unemployed workers were actually receiving UI benefits by 2019, down from roughly half in the 1950s — eligibility rules written for full-time payroll work miss gig workers and part-timers almost by design. A safety net that catches one in four is mostly hole.

Rebuild worker bargaining power, because one worker negotiating alone against a firm deploying AI at scale is not a negotiation. Sectoral bargaining — standards set across an industry, not shop by shop — is the left’s structural answer, with real AI-era logic: it can bargain over how automation is deployed (augmentation versus replacement), not just wages.

Invest in retraining — with a caveat honest advocates state themselves: the evidence record is mixed. The gold-standard randomized WIA evaluation found that intensive, staff-assisted services raised earnings 7 to 20 percent and paid for themselves — while other rigorous work found training gains for regular adults but not for dislocated workers, exactly the people a transition policy targets. The left’s honest position isn’t “retraining works”; it’s “the parts that demonstrably work — human navigation, staffed guidance — are the parts we starve.”

Beneath it all, some argue for a job guarantee: if the market can’t offer work, the public should — because work distributes belonging, not just income.

That’s the real case. It is not “pay people to disappear.” It is: we did deindustrialization badly, we watched what abandonment did to whole regions, and we refuse to run that experiment again at national scale.

---

The Case from the Right, at Full Strength 🧢

The right’s core claim is also older than any program: the best worker policy ever discovered is a growing, dynamic economy that nobody has to be assigned into. Every prior technology panic ended with more jobs, in categories no planner foresaw. This isn’t faith; it’s measured: about 60 percent of the jobs Americans did in 2018 did not exist in 1940. No committee invented “app developer” or “solar installer.” Growth did. The right’s first question about any transition policy is therefore brutal and fair: does this speed people toward the new work, or park them comfortably outside it?

And the right has receipts on the parking lot. The federal government’s own evaluation of Trade Adjustment Assistance — the flagship program for trade-displaced workers, the closest analogue to an “AI adjustment” program — found that four years in, participants were earning about $3,300 less than comparable non-participants. Years in subsidized training were years not spent building experience in the actual labor market. Compassion that lowers lifetime earnings is not compassion; it’s sedation with paperwork. (Honest advocates note the gap was narrowing by year four, and the minority who finished training and found in-field work did markedly better — but the headline finding stands.)

The affirmative agenda follows: remove the barriers we built ourselves. Occupational licensing now covers more than one in five American workers, up from about five percent in the 1950s — so a displaced worker who wants to cut hair or fit pipes in a new state faces months of fees and hours-requirements that mostly protect incumbents. In a transition where people must move between kinds of work quickly, every licensing wall is a tax on adaptation, paid by the least credentialed.

Add work as dignity, not just income — the right takes seriously that unemployment wounds people in ways a check doesn’t heal, and reads the social science on idleness and despair as one long warning against making non-work comfortable. And humility about prediction — if nobody can name the safe jobs, a retraining apparatus aimed at last year’s whiteboard is a slow cannon aimed at a fast target. Let millions of employers and workers discover the new work through prices and experiment, the way they discovered the last sixty percent.

That’s the real case. It is not “let them eat dynamism.” It is: the new rungs have always come from the churn, the graveyard of failed adjustment programs is real, and the kindest thing you can do for a displaced worker is a labor market hungry to hire her.

---

The Reframe: Floors and Hinges 🧰

Set the two playbooks side by side and notice: each answers a different failure mode, and both failure modes are real. The left guards against abandonment (which happened). The right guards against entrapment (which also happened). But both share one load-bearing assumption the AI transition may not honor: that you can tell, in advance, which workers and which industries the storm will hit. Deindustrialization policy knew its target: steel, autos, the Midwest. Growth policy knew its promise: the new rungs will appear over there. When nobody can name the safe jobs, category-based policy — help sorted by industry, occupation, or region — is aiming at a map that redraws itself.

Here is the ReThink proposal set — to be argued with, not obeyed:

1. Trigger help by event, not by category. Stop asking “is your industry on the list?” and start asking “did displacement happen to you?” A documented layoff plus a demonstrable earnings drop opens the toolkit — wage insurance, retraining funds, relocation support, a UI system that actually catches you — whether you were a welder, a paralegal, or a 24-year-old developer. Event-based triggers don’t require anyone to predict the storm’s path; they fire wherever it lands. The right should notice: no eternal industry lists to lobby onto. The left should notice: it finally covers the people every industry list misses.

2. Make benefits portable — cut the cord to the employer. Health coverage, retirement contributions, training accounts that follow the person through job changes, gig stints, and reinvention. When the median worker may cross more kinds of work in a decade than her parents did in a career, benefits welded to one employer are a tax on exactly the agility both parties claim to want. Conservative flexicurity fans and progressive gig-worker advocates keep arriving at this plank from opposite doors.

3. Rebuild retraining around honest signal. Fund what the evidence says works — the staffed, intensive navigation that raised earnings 7–20 percent — and be truthful about what doesn’t. Then change what’s taught: not “train for the safe occupation” (nobody can name it) but AI fluency inside the work you already know — the plumber who quotes with it, the nurse who documents with it — plus real-time public data on which skills are repricing, so the whiteboard updates itself instead of lying at a decade’s delay.

4. Clear the road while you build the bridge. Every floor above pairs with a barrier tear-down: licensing reciprocity across states, so the displaced worker who retrained as an electrician can work next month, not after another year of fees — and a standing rule that any new trigger or benefit sunsets unless re-justified, so the right’s ratchet worry gets an answer in the architecture, not a promise. A transition policy that only adds programs isn’t neutral; it’s half a toolkit.

5. Grade everything on the people with no runway. The Federal Reserve finds 37 percent of American adults could not cover a $400 emergency with cash. For that third of the country, “retrain for six months” is not advice; it’s arithmetic that doesn’t close. Any transition policy that assumes runway is a policy for the people who least need one.

Run the set against this series’ one test. Does it move toward abundance? Yes — every plank speeds adaptation to the new tools, and none taxes the engine building the abundance. Does it put the most vulnerable at the center? That’s plank 5’s whole job — event-triggers and portability reach the gig worker and the no-savings household first, because they’re the ones category-based systems currently drop. Does it keep the both-and — market engine plus dignity floor? That’s the architecture itself: the market keeps discovering the new work, while the floor stops being a bet on anyone’s forecast. Floors that don’t require predicting the storm; hinges that open when life actually turns.

---

Where This Could Be Wrong ⚖️

Honesty in both directions, because the reframe has real failure modes too.

If the skeptics are right, and Acemoglu’s modest numbers hold for decades, then new trigger systems and portable-benefit plumbing spend real political capital on a storm that stays offshore — the right would fairly say the churn we already have needed licensing reform more than new architecture. (Counter-thought: triggers in a calm decade cost little because they don’t fire — that’s what insurance is. But setup costs are real, and defenders of the status quo aren’t crazy.)

If the aggressive forecasts are right, the reframe may be too small. Event-triggers assume displacement arrives as discrete, documentable events. A slow repricing (hours thinning, raises vanishing, new grads never hired at all) might never trip the wire. The Stanford canary data is exactly this shape: the 24-year-old who never gets the first job files no displacement claim. Floors-and-hinges may need a floor that doesn’t wait for a trigger at all — which reopens the very universal-benefits fight this framework tried to route around.

And two design worries. “Documented displacement” invites gaming and adjudication bureaucracy — who decides your earnings drop was the economy and not you? The left should also press: floors without bargaining power can become quiet serfdom — comfortable, covered, voiceless — and nothing here rebuilds voice by itself. The right should press back: event-triggers have a way of ratcheting into permanent entitlements the moment a recession hits. Both worries are historically literate. Neither has a clean answer yet.

---

For the Person Holding a Policy Pen 🖊️

Three things to carry out the door, whatever your jersey:

Stop demanding the safe-jobs list before acting. Its nonexistence is the finding. Policy that requires the list is policy that waits forever; design for not-knowing instead.

Fund the measurement now. The Stanford payroll work exists because private data happened to be available; a country navigating a repricing of work shouldn’t be learning about it from one lab’s access to one payroll processor. Real-time, public, occupation-level signal is cheap compared to flying blind.

Argue about triggers and floors, not headlines. “Is AI taking jobs?” is a bar argument. “Should transition help fire on events or categories, and how thick is the floor for the 37 percent?” is a policy argument. The second can actually be won — by either side, in public, with evidence.

---

Honest Fine Print 📎

This is not a political verdict. The left’s fear (abandonment) and the right’s fear (entrapment) have both actually happened to real American workers, and both camps are full of people trying to protect the same families. We’ve given each side its strongest voice; if you finished unable to tell which way this essay votes, it worked.

The timelines are genuinely unknown. Everything here sharpens if the fast forecasts land and relaxes if the slow ones do — and nobody, including us, knows which. The honest position holds the entry-level data and the macro skepticism at once.

And this is a proposal for debate, not a plan anyone has voted on. Floors and hinges is a sketch slid across the table — argue with it, improve it, tear out the planks that don’t survive contact. Just don’t keep running either legacy playbook unexamined while the counselor stares at her whiteboard.

Because the two people at her desk — the fifty-five-year-old and the twenty-four-year-old — aren’t on different sides of anything. They’re in the same line.

---

Facts verified against sources current as of July 2026, linked inline. Companions: The Thousand-Day Question (the personal side of this transition) · The Eighteen-Year Bet (the next-generation side).

— The ReThink · firstfruits 🌱 · truth first, hope on top