The United States spends about 20% of the OECD average on active labor market policies — the training, job-search and transition programs that move a worker from one job to the next. It subsidizes capital investment generously and hiring almost not at all.
Gina Raimondo, who ran the Commerce Department and implemented the CHIPS Act, is not arguing that companies should be forced to keep people. She is arguing that nobody has yet written the tax incentive that would make them want to.
"There's really almost no incentives to invest in people. It's all capex incentives."
Raimondo's father lost his job at the Bulova watch factory in Providence in his mid-fifties when the company moved production overseas; three months ago she launched RAISE US, which has raised more than $500 million of a $1 billion target from Amazon, Microsoft, Bank of America, Anthropic, OpenAI, GM, UPS and IBM.
The full interview is covered here so you can skip it. 42 minutes of audio, 18 minutes of reading.
Here are the 11 arguments that matter.
👤 Guest: Gina Raimondo, former US Secretary of Commerce and former Governor of Rhode Island, who was the chief advocate for the CHIPS Act and ran its implementation, and who now leads RAISE US, the workforce organization she launched in June with Indiana Governor Eric Holcomb
🎙️ Host: David Deming, a Harvard economist who studies the labor market and hiring, and who hosts The Context Window
📰 Published: 16 September 2026 on YouTube and the show's Substack feed
🔴 YouTube | 🟣 Apple Podcasts | 📝 Show notes | ⏱️ 42 min | ✅ Time saved: 24 min
Key Takeaways
The tax code pays companies to buy machines and gives them nothing for hiring people
Full immediate depreciation on capital costs the Treasury about $36B a year
She says the 4.1% unemployment rate is the wrong number to watch
Strip out involuntary part-timers and people who stopped looking and labor force participation is 61%
No company she works with has implemented AI deeply enough to cause mass layoffs yet
What the data does show is a hiring slowdown, concentrated on entry-level workers
The question she puts to every chief executive is what incentive it would take to retrain rather than lay off
RAISE US has raised over $500M of a $1B goal, from every frontier AI lab except Google
The US spends about 20% of the OECD average on active labor market policies
She argues the programs are worth building even if the AI job shock never arrives
Deming's answer to her homework is a refundable federal credit for entry-level hiring
Up to $5,000 per hire over two years, about 4 million eligible hires, roughly $20B a year
Entry-level hiring is inefficiently low because a first employer creates information rivals can free-ride on
Her advice on future-proof jobs is not to look for one, and then: advanced manufacturing, health care and science
1. What RAISE US Is
Deming opened on the organization Raimondo launched in June with Indiana Governor Eric Holcomb.
She described herself as an enthusiast about the technology and a pessimist about the handover. "So I'm an AI enthusiast," she said, and expects artificial intelligence to create jobs and companies nobody can now predict. What worries her is the handover: "But I am very, very worried about, the transition from here to there and having a period of high unemployment."
The mission as she states it has two halves that she says are currently in conflict: the United States should lead the global artificial-intelligence competition with the best technology, and should do so "Without leaving workers behind."
The method is piloting rather than legislating. RAISE US works with bipartisan governors and a coalition of companies to fund new ideas in four pilot states — Arkansas, Connecticut, Maryland and Utah.
The pilots she named: a year of service in Maryland where young people are paid to take a job in health care or teaching and pick up skills; a startup accelerator for people on unemployment insurance, on her expectation that artificial intelligence produces a wave of new companies; a wage-insurance pilot, which tops a worker's pay back toward what they earned before regardless of the new job; and a job-sharing pilot.
Her stated reason for urgency is the last time: "And we don't wanna make the same mistakes we did with the China shock, for example."
Deming's framing of the same design is that it is agnostic about winners. Wage insurance pays out whatever job the worker lands in, which he said is the right structure when nobody knows where the demand goes.
2. 20% of the OECD Average
Raimondo put a number on how far behind the United States starts.
"Yeah, so we spend, as I'm sure you know, about 20% of the OECD average on active labor market policies." Her conclusion is that the country is behind in an objective sense on workforce development and training.
Her complaint is not that the money is missing but that it is pointed somewhere else. The United States spends heavily subsidizing traditional four-year college.
She was blunt about the supply side of higher education, in front of a Harvard professor: "Now, if you're lucky enough to go to a place like Harvard, that's amazing, but there's 4,000 colleges in America, I think half of which probably should go away." That money, she said, should be available for other skill-building that leads to a job.
Deming agreed and said workforce development needs to be more flexible, and that not every college needs to be like Harvard.
3. Employers Have to Change
Asked how the corporate money and the AI labs fit together, Raimondo said the companies are the point, not the funding.
"We can't do this without employers at the table, because they need to change the way they hire, and fire, and train, and retrain, and redeploy."
What she wants from them is a commitment to experiment and an honest answer about price. She said she asks what it would take for a company to retrain and redeploy some share of its back-office or call-center workers instead of laying them off.
She is asking directly, chief executive by chief executive. She described a long meeting with the executive who runs Amazon's retail business, and the script she uses: high unemployment is destabilizing for America, so what are your strategies for implementing artificial intelligence without producing it, and what incentives would you need from government?
Where government is not ready, she wants philanthropy to prove the model. In the short term, she said, philanthropy can fill the hole where an incentive should be, demonstrate that it works, and then government can take it on.
Her example of an idea worth testing is job sharing: instead of firing someone, move them to half time and keep paying them.
Her view is that the constraint is will, not capability. She said American companies have done things previously thought impossible, and that if she went to any chief executive and said "Your life depends upon it. Your share price depends upon it. Your job depends upon it. Figure it out. You gotta both use AI and not fire 10% of your workforce," they would work it out.
Deming pressed on the gap between the obligation and the mechanism. Every chief executive will agree they have an obligation, he said; the question is how RAISE US turns that into a commitment to reinvest productivity gains. Raimondo's answer was that at the moment she is only asking what it would take.
4. How They'll Measure It
Deming asked how the organization will decide which pilots to scale and which to drop.
Raimondo said the metrics are simple and the execution is not. "Do you get a job? At how many, how much wages? How quickly do people transition? Can we reduce the number of days of unemployment between one job and the next?"
She corrected herself on the record about the word "easy." "No, I'm sorry, it's hard to do. It's not easy to do." — hard to execute, she said, but easy to know whether you are executing.
Per-intervention metrics exist; an overall target does not yet. She said the organization is having conversation after conversation about the most ambitious version of what is possible, and has so far hit its own interim goals on fundraising, state partners and team building.
She ruled out the headline metric as overreach. Deming raised the option of a goal expressed as the national unemployment rate being some number of points lower because of the work; Raimondo called that "too grand" for a nonprofit.
Deming praised the habit of setting public targets as a management strategy she has used before.
5. Under the Hood of 4%
Deming put the strongest available objection to her: if you plot the unemployment rate or the youth unemployment rate, there is no visible AI shock to point at.
Her first answer is that it is too early. "I really don't know any companies that have implemented AI so sufficiently that it would lead to major layoffs," she said, and she works with a lot of companies. They are only now redesigning work and org charts around it.
Her second is that the headline number is measuring the wrong thing. Everybody points at 4% unemployment, she said, and "You've gotta look under the hood."
The specific measurement objection is part-time work: "I personally think it's a little bananas that our unemployment rate includes people who are part-time employed." A college graduate working 30 hours a week with no benefits, below their capacity, does not feel employed — and she said the economists who taught her at Harvard would explain why the definition is defensible.
The number she prefers: "If you strip out all the people who are desperate to work but have given up, there's 61% labor force participation rate." Her verdict on it was that it is not a good picture.
She was careful not to blame artificial intelligence for it. What the data does show, she said, is a hiring slowdown: "Maybe not a lot of big RIFs yet, give it a year or two as they really implement AI, but they're slowing their hiring." A reduction in force is a layoff.
The people it lands on are new entrants. More young and entry-level workers are either not getting a job or taking a part-time job when they want full-time work.
Deming explained the mechanics to the audience. To be counted as unemployed you have to be looking for work, so a 25-year-old graduate working part-time is classified on the basis of whether that was a choice — which he said is genuinely hard to read.
6. A No-Regrets Bet
Deming asked whether the agenda stands up even if the AI shock never arrives.
Raimondo agreed without qualification, answering "1,000%."
She said Kara Swisher had pushed her on the same point on a podcast the night before, and that her answer was that this is all work that probably should have been done already.
Her theory of change is that nothing moves without a crisis. "But in my experience, big things only happen in times of crisis or other big changes." If artificial intelligence is the outside force that gets the country to fix workforce policy, she said, so much the better.
Deming's version is the consultant's phrase she reached for: these are no-regrets moves, building resilience in case the labor market does deteriorate.
Both had signed the We Must Act Now letter on AI, which Deming raised as common ground: whatever the quibbles with it, this is not a normal transition and it needs the attention of policymakers.
7. Capex Gets the Tax Credit
The core of the interview is a concession Raimondo makes against her own politics.
"Companies respond to incentives." That is the premise she keeps returning to.
She stated her opposition to the tax law first and in the strongest terms: "For example, I am extremely opposed to the Trump tax cuts that were just, re-extended." She wanted to be crystal clear about it.
Then she credited the part of it that worked. The law allows 100% depreciation for capital, and she said that has in fact led to pretty significant investment by companies.
The asymmetry is the whole argument. "There's really almost no incentives to invest in people. It's all capex incentives." Deming restated it: there is a tax credit for the depreciation of capital and nothing equivalent for people.
She handed the problem to him on air. "Go get some good ideas for how to do that, and then, m- maybe I could go to a state and see if they would be willing to do it." She said the work is properly federal but could be piloted in a state.
Her longer-run bet is that federal AI legislation arrives within five to seven years, and she wants tested ideas on the shelf when it does.
On why a nonprofit is the vehicle, she said good ideas are cheap and coalitions are not. An idea backed by the chief executives of 30 or 40 of the largest American companies and a bipartisan group of governors, she said, has a shot.
8. Growth Is the Only Answer
Deming raised the scaling problem in labor-market research: programs that place people into better jobs work well in small trials and work less well at scale, because one placement can displace someone else.
Raimondo's answer was two words: "Growth is the only answer." The hope has to be that artificial intelligence produces growth in the form of more jobs — "otherwise we're totally screwed."
The timing is the difficulty she keeps naming. She said "the layoffs will come before the new jobs," which is the gap the organization exists to cover.
The policy ask is unchanged: find a way to get companies to reinvest some of the productivity gains into new growth areas and new job creation.
The other half of her answer is company formation, and not the famous kind. She said she means small entrepreneurial ventures rather than the next Amazon or Meta, and that new business formation has fallen for 30 years.
The number she uses strips out the shell companies: "Except if you take away the ones that don't employ anyone, it's like 400,000."
This is the part of artificial intelligence she is most enthusiastic about, because in theory it is cheaper and easier than ever to start a company.
Deming cited recent research on AI-native firms that cuts both ways: there is a boom in them, they employ fewer people each, but they grow faster and there are more of them, so the net effect on employment is unresolved.
Raimondo's answer to that was disarming: "I have no idea is my honest answer to these things." Deming replied that none of us do. "But I know it'll be better if we try some things," she said.
Her precedent is her own first term. Rhode Island had the highest unemployment rate in the country when she took over as governor and everyone wanted the single answer; there wasn't one, so she worked tourism jobs, building-trades jobs and manufacturing jobs separately.
9. Jobs of the Future
Deming asked the question she has deliberately avoided answering: which jobs are safe.
Her first answer is to reject the framing. "First of all, I would say don't try to find things that are future proof" — design yourself, your skills, your way of learning and your tolerance for volatility instead.
The line she gives her own children, aged 19 and 22: "And, my little mantra to them is face the future with courage and optimism." To do that, she said, you have to know how to think, have a strong network, be entrepreneurial and not be resistant to change.
She then named sectors anyway, with a hedge. "Advanced manufacturing of some form I think is a good bet," on her view that making more things in America is a long-term trend, which points at mechanical and electrical engineering and design work.
Health care she described as up and to the right forever, and she added science: "Yeah, science, healthcare, advanced manufacturing."
Her example is Eli Lilly, a member of the RAISE US coalition, whose use of artificial intelligence to speed up research and discovery she said will create more jobs rather than fewer.
10. The Verifiability Problem
After the interview, Deming recorded a second segment doing the homework Raimondo assigned. He started by dismantling the premise.
Wages and formal training are already fully tax-deductible, he said, so on paper the tax treatment of labor is not obviously worse than that of capital.
The gap is in the training the tax code cannot see. Most job training is informal and on the job — a new office worker learning to run a project end to end, a medical resident learning to triage and to communicate uncertainty to a family, a junior engineer learning to fix other people's code and trade speed against reliability.
General training is underprovided for two reasons. The first is that broadly useful skills leave with the worker, so the employer who paid for them often does not recoup the cost. The narrow, company-specific training that is fully deductible is, he said, probably overprovided.
The second is information. "Hiring someone into a new role they've never done before is a shot in the dark." His own illustration is that he could not credibly prove to an accounting firm that he would make a good accountant, however true it might be.
The research he builds on is Amanda Pallais's 2014 paper, Inefficient Hiring in Entry-Level Labor Markets. Her argument is that a first employer generates information about a worker that benefits the worker far more than the employer: "They create a public record of what you can do, and other employers get to free ride off that."
The experiment behind it hired workers on the freelance platform then called oDesk, now Upwork, and randomly gave some of them a detailed public evaluation. Those workers did better afterward — but the result Deming emphasizes is that the evaluations raised total hiring and total hours on the platform. "They're actually making markets."
The pattern repeats across studies of verifiable work certificates, references and letters of recommendation, which have outsized effects on entry-level workers.
His conclusion is that entry-level hiring is inefficiently low because nobody is paid to fix the information problem — employers would rather let a rival take the first chance and poach the good ones later.
The corollary is why junior pay is low. Medical residents, law clerks and interns accept low wages because part of the compensation is training that pays for itself later.
The timing argument: the unemployment rate for new college graduates was 5.7% in August, against 4.1% for all workers. Deming said the downturn in graduate hiring began before artificial intelligence and may be a consequence of the pandemic shift to remote work, but that AI has probably made it worse by taking over the entry-level grunt work that used to buy a new hire their training.
11. A $20B Hiring Credit
The proposal itself is a refundable federal tax credit for entry-level hiring, with entry-level defined by occupation rather than age, so it covers career switchers as well as new graduates.
The illustrative schedule: 20% of the first $20,000 of wages in year one, a maximum of $4,000, then 10% of the first $10,000 in year two, another $1,000 — $5,000 maximum per hire.
The anti-gaming condition is a third year with no subsidy. A firm that does not retain the worker for that third year has to pay the credit back.
The administration would run through payroll, as a quarterly credit verified at hiring, available only for hires who have not previously held that occupation — which he noted the IRS can check, because taxpayers already write their occupation on their returns.
The cost arithmetic, which he repeatedly flagged as rough: the Bureau of Labor Statistics recorded about 59 million private-sector hires in 2025 and projects 17.5 million annual openings over the next decade, of which 9.5 million are transfers between occupations. He takes 15 million as a working estimate, cuts 25% for government, self-employed and otherwise ineligible workers, and applies a 35% three-year retention rate from BLS duration data — just under 4 million eligible hires, or about $20 billion a year.
The comparisons are the argument. The Work Opportunity Tax Credit, which has a similar structure at smaller scale, costs $1.5 billion a year; WIOA state workforce grants about $3 billion; Job Corps about $2 billion. Against that, the Earned Income Tax Credit costs about $70 billion, and the 100% bonus depreciation provision costs taxpayers about $36 billion a year.
That last number is the point of the whole exercise: $36 billion a year of incentive for capital, and his proposal would put $20 billion back on the side of labor.
The outside evidence is Swedish. A 2019 paper by Emmanuel Saez, Benjamin Schoefer and David Seim found that a payroll tax cut for hiring young workers raised youth employment and caused firms that hire a lot of young people to expand. Deming prefers his own version because it targets career starters and career switchers rather than the young.
He named the pilot state himself. Maryland, already a RAISE US partner, has a job creation tax credit with a similar structure but no entry-level focus. "Call me, Secretary."
He was explicit about the limits of his own numbers: "Again, I made these numbers up, so you could easily scale it up or down to be less generous."
Bonus Insights
Every frontier AI lab funds RAISE US except one. Asked about the corporate list, Raimondo said the backers are all of them "minus Google," which she said would not participate.
The labs are giving data as well as money. She said RAISE US has data arrangements with them covering how and for what people are using artificial intelligence, and that they help the organization understand where the technology is heading — the rate of improvement every six months being, in her word, bananas.
Deming's introduction supplied the biography behind the mission. Raimondo's father worked at the Bulova watch factory in Providence until the company closed it to chase cheaper labor overseas, leaving him unemployed in his mid-fifties. She went to Harvard, became a Rhodes Scholar, took a law degree at Yale, founded the venture firm Point Judith Ventures, became Rhode Island's treasurer in 2011 and its governor in 2014, and then Commerce Secretary in 2021.
He also sized the thing she is best known for: the CHIPS Act authorized $280 billion in federal funding for an American semiconductor industry, and she ran the distribution, including money for workforce development and job training.
The exchange closed on the value of a general education. Raimondo said not a day goes by without gratitude for what Harvard taught her about how to think, and that it gave her a framework she applies to every problem despite never becoming an economist. Deming called it the best advertisement possible for a liberal arts and sciences education.
Raimondo's bottom line is that the artificial-intelligence labor shock has not shown up in the aggregate data yet and will, that the American system for moving people between jobs is in no state to absorb it, and that the fix is not an obligation imposed on companies but an incentive that makes hiring and retraining as attractive to them as buying equipment already is.
Products, Companies & Tools Mentioned
RAISE US (The organization she launched in June with Indiana Governor Eric Holcomb; more than $500M raised toward $1B, piloting wage insurance, job sharing, AI career coaching and paid years of service in four states)
Amazon, Microsoft, Bank of America, GM, IBM and UPS (Corporate backers; she described a long meeting with the executive who runs Amazon's retail business about implementing AI without mass layoffs)
Anthropic and OpenAI (Frontier labs funding the organization and sharing usage data with it)
Google (The one frontier lab that declined to participate)
Eli Lilly (A coalition member she cited as using AI to accelerate research and discovery, which she expects to create jobs rather than cut them)
Upwork (The freelance platform, then called oDesk, where the entry-level hiring experiment Deming cites was run)
Work Opportunity Tax Credit (The closest existing analogue to Deming's proposal, at $1.5B a year against his estimated $20B)
Books & Resources Mentioned
Inefficient Hiring in Entry-Level Labor Markets – Amanda Pallais (The 2014 paper behind the argument that a first employer creates information other employers free-ride on)
Payroll Taxes, Firm Behavior, and Rent Sharing – Emmanuel Saez, Benjamin Schoefer and David Seim (The 2019 Swedish evidence that a payroll tax cut for hiring young workers raised youth employment)
Watch the full episode:
If this was worth your time, send it to someone who has to have a view on this.
Get the latest market chatter as it happens: