Follow the gap13 minOpenAI wanted $30 billion at a $1.4 trillion valuation. After a tech selloff and a midterm backlash against data centers, it has closed $11 billion at $1.25 trillion and is courting wealthy individuals for more. Follow the money, and it leads closer to ordinary portfolios than most people think.Read now →

OpenAI's Chief Economist Expects Unemployment Under 5% in 2030. The Rest of His AI Forecast Is Stranger
He thinks AI will show up in the productivity numbers within a year or two, that unemployment can stay at 5% or lower, and that the real bottleneck to AGI is not the models but the org chart. Here is the full forecast from the economist with a rare view of how people actually use AI.
Contents
- The computer-age puzzle, again
- The job he wrote for someone else
- The next 18 months: three calls
- What is noise
- Who adopts next: lawyers and HR
- Cheaper startups, faster failures
- Where the world is surprising
- Inflation, capex and cheaper everything
- The 2031 economy and the care boom
- Agents that navigate bureaucracy
- The China-shock warning
- Why task lists mislead
- The numbers he is watching
- Why this matters
- Practical takeaways
- Counterpoints and risks
- FAQ
- The bottom line
In 1987, the economist Robert Solow wrote a line that became famous: "You can see the computer age everywhere but in the productivity statistics."
Offices were full of PCs. The national numbers barely moved. The productivity boom finally arrived in the late 1990s, almost a decade later.
Now the same question hangs over AI. Everyone can see it. Can the economy?
Few people are better placed to answer than Ronnie Chatterji, OpenAI's first chief economist. His team can see how millions of people actually use AI at work, data most statistical agencies would love to have. His answer is more optimistic on timing than many economists, more cautious on hype than many technologists, and full of surprises about who wins, who loses and what the bottleneck really is.
The computer-age puzzle, again
Chatterji's headline call is that AI will begin showing up in productivity statistics "in a more significant way" over the next 12 to 18 months, with the bigger payoff arriving within a year or two.
That would be a far shorter lag than the computer age delivered. But he is careful about what "showing up" means.
AI is already in GDP, he points out, through the investment channel. The flood of spending on chips, servers and data centers has added somewhere between 40 basis points and 1 percentage point to US GDP growth, by his reading. What economists really want to see is total factor productivity, the output gains that come from working smarter rather than spending more.
On that, he is cautious. Recent data are "consistent with" AI showing up, he says, but "I don't think we yet know for sure that it is AI."
He recalls a famous 1990s Federal Reserve debate. Alan Greenspan was convinced he saw a productivity surge in the data. Janet Yellen, then a governor, said she did not yet see it. The numbers were later revised sharply upward. Chatterji thinks we may be at a similar moment. Attributing the gains to AI, though, will only be possible in hindsight.
His bottom line borrows a phrase attributed to William Gibson: the future is already here, it is just unevenly distributed.
The job he wrote for someone else
Chatterji did not go looking for the role. He had just finished a government tour that included coordinating implementation of the CHIPS and Science Act in the Biden White House, after a career as a professor at Duke's Fuqua School of Business.
OpenAI first approached him about semiconductors. The conversations kept going. Eventually the company said it needed a chief economist. He told them he did not want the job, but offered to write a job description so they could find the right person.
They came back saying he was the right person and sent him a job description. It was the one he had written. OpenAI announced his appointment in October 2024.
The role, he says, has changed "like six times" since. Two parts stayed constant: publishing research that gives the public accurate information about AI's economic effects, and building a team that mixes PhD economists and data scientists with operators who know how to get things done inside an organization.
The next 18 months: three calls
Asked what will change between now and mid-2028, Chatterji gives three predictions.
1. The agentic revolution is just beginning
"We are in the early innings of an agentic revolution," he says. That may sound odd in San Francisco, but outside tech hubs the shift has barely started.
His evidence: a relatively small share of firms using AI have even switched on connectors, the features that link AI to the software and data a company already uses. Without that, AI is just "an ask and answer machine." He expects agentic AI to spread across the rest of enterprise over the next 12 to 18 months.
2. The job market changes underneath
The headline job market has not shifted as much as AI's capabilities might suggest. What has changed is the work beneath the job titles.
OpenAI's data show a large share of work now falls outside people's job descriptions. Finance professionals do marketing and HR tasks. Communications staff run the numbers. Chatterji expects that trend to accelerate as agents spread.
3. Productivity finally shows up
As above: AI is already visible in GDP through investment, and he expects productivity gains to become clearer within 12 to 18 months.
What is noise
Chatterji is just as clear about what he thinks people get wrong.
Wild growth numbers. People outside economics often multiply historic growth by five. But in a mature economy, he says, even one extra percentage point of US growth would be "transformational," and two points would be felt by everyone. Ten percent growth in the next couple of years is "probably something that we're not calibrated correctly on."
Overnight mass unemployment. If unemployment rose even half a point to a point, there would be serious national debate. He notes that it took COVID to triple unemployment in the spring of 2020. Predicting a similar jump from AI in the next 12 to 18 months implies "a different economic model than I would be" using.
Who adopts next: lawyers and HR
Engineering and research teams led adoption. Who follows? Chatterji's answer: legal and HR, two functions "people should not sleep on."
Inside OpenAI, he watched the legal team accelerate as fast as any group. And he describes a telling internal lag. By late 2025, nearly all research and engineering staff were using agentic AI for essentially all their work. The rest of the company, including Chatterji, who jokes that as an economist he now counts as "non-technical," caught up only around April or May 2026.
That is a four-to-five-month gap between technical and non-technical staff, at a company that builds the technology. He suggests watching fields with lots of structured data and process, like legal and people operations, for the next wave elsewhere.
Cheaper startups, faster failures
Chatterji began his academic career studying entrepreneurship, and he sees AI as "yet another accelerant" of new company formation.
His favorite insight is counterintuitive. The value is not just more successes; it is faster resolution. Most startups fail, and failing quickly helps founders learn, start the next venture, or realize they are better as early employees than founders.
He cites his own research finding that one of the most common failure points in starting a business is simply telling someone about the idea and getting feedback. AI removes that friction.
He also points to "solopreneurs," people like his mother, a piano teacher who is technically a one-person business. AI can handle marketing, billing and parent communication so she can focus on teaching. In regions where it is hard to find a co-founder or hire a marketing chief, people "can borrow an expert from AI." OpenAI has reported that about 4 million Americans use ChatGPT somewhere in the entrepreneurial process. A restaurant in Southeast Asia, he says, can now test menu and location decisions it could never afford a consultant for.
Where the world is surprising
Chatterji has visited 12 countries since joining OpenAI. Some patterns surprised him.
- East Asia (Japan and South Korea): strong optimism and heavy enterprise use, helped by past tech success and firms already benefiting from the chip supply chain.
- Indonesia and Brazil: much higher usage than income or education levels would predict, driven largely by young populations. AI use is heaviest among 18-to-25-year-olds worldwide. University students in Jakarta, he says, talked about AI the same way students in San Francisco do. "There was no lag."
- India: a surge of AI startups, plus incumbent conglomerates determined not to become "the dinosaur" in a business-school case study.
- Europe: less a negative surprise than an active debate about enterprise adoption and keeping control of key parts of the technology stack.
In many developing countries, he notes, AI is mostly a consumer story for now, which fits historical patterns of technology diffusion.
Inflation, capex and cheaper everything
The IMF has warned that the AI investment boom is propping up US demand before productivity arrives. Two forces collide: spending pushes inflation up now, while productivity should push it down later.
Chatterji's take: most inflation drivers today, such as geopolitics and groceries, have nothing to do with AI. The AI-related part comes from hardware bottlenecks. He flags memory prices in particular, noting how dramatically DRAM prices have climbed over the past year. That component should recede as supply expands.
Over time, he expects AI's deflationary effect to win, lowering the cost of many goods and services. Two years may be too soon, though. And cheaper often means more demand. "In San Francisco, they call it Jevons paradox," he says. "I just call it downward sloping demand curves."
The 2031 economy and the care boom
Assume AI keeps delivering. What does 2031 look like?
Chatterji points to research by Nicholas Bloom, Raffaella Sadun and John Van Reenen showing that US firms in the IT era adopted more technology than UK firms, paired it with better management practices, and captured more productivity. He expects a similar, possibly bigger, version with AI: early-adopting countries like the US pulling ahead, with growth potentially reaching 3 to 4% in a steady state.
Then his most distinctive prediction: more jobs in education and healthcare, not fewer.
His reasoning:
- Richer societies spend more on care and learning.
- Adding technology to those sectors has historically added staff, not removed them.
- People prefer humans. Asked whether he would rather replace a kindergarten teacher with a robot or hire a second teacher to halve class sizes, he would hire the teacher, and he thinks most Americans would too.
Education and health already account for roughly one in seven US jobs. The catch is pay. If people move into care jobs that pay less than the jobs they left, the transition will hurt. Those jobs "need to improve and need to be rewarded." Funding could come from the "AI dividend," possibly requiring changes to how labor and capital are taxed. Entertainment and new creator-style jobs could also grow, though he wonders how much AI-made content people can really consume and how concentrated demand will be.
Agents that navigate bureaucracy
Chatterji sees a big, overlooked opportunity: AI moving from information to navigation.
The internet could tell you what goes on a financial aid form. An agent could actually help you fill it out. The same applies to a family navigating a school voucher program or a patient trying to get on the right transplant list.
The problem is obvious. If everyone's agent floods a system, as when every diner's agent calls a restaurant with only a phone, the system breaks. His answer: platforms on both sides, with agent-to-agent rules set democratically. He also sees AI as "the first expert" for people who never had access to one, such as someone hundreds of miles from a behavioral health specialist.
Always-on agents raise the same question. Inside OpenAI, he says, agents already work about three times as many hours as humans, and teams have had to set rules on how much AI-generated content can be pushed into a Slack channel and who is responsible for reviewing it. Sending someone an incomprehensible pile of AI output, he agrees, is almost an attack on them. He argues that parallel, persistent agents in fields like cybersecurity and finance are what justify today's enormous compute investments, something people outside AI may be "sleeping on."
The China-shock warning
Asked who loses, Chatterji reaches for the China shock: the wave of import competition that hit specific US manufacturing regions in the 2000s. The direct job losses were a small share of a huge economy, but the pain was concentrated, the benefits were diffuse, and the political consequences lasted a generation. He lives in North Carolina, which was hit hard.
AI could follow the same pattern. Benefits are spreading widely, including large consumer surplus, the value people get from free tools that never shows up in GDP. He puts it at roughly $300 billion a year and says the welfare gains may be largest in places like India and Brazil where other services are scarcer. Losses may cluster in particular industries and regions, feeding skepticism already visible in local fights over data centers and in the 2026 elections.
His prescription has three parts:
- Empathy, then evidence. Do not answer anxiety with a headline unemployment rate. Acknowledge that it is a hard time to look for a job, then show where AI is and is not hurting. He points to Stanford research finding weaker employment for early-career workers in AI-exposed jobs, as the clearest negative signal so far.
- Fix the boring safety net. Before inventing new programs, make sure "automatic stabilizers" like unemployment insurance and food assistance actually work, including fixing creaky state unemployment systems. These are "no regrets" moves.
- Keep humans in charge. People need to feel they are making choices, not having things done to them.
He adds that labs alone cannot steer AI toward augmenting workers rather than replacing them. Agreeing with Nobel laureate Daron Acemoglu that technology's direction is a choice, he says it must be made collectively, by labs and government. He praises CEOs focused on creating new value over those focused only on cost cutting.
Why task lists mislead
Much of the AI-jobs debate began with a simple method: list a job's tasks using the government's O*NET database, test which ones AI can do, and estimate exposure.
Chatterji calls the conclusion people drew from that "not how it works at all." His father, also an economist, got an office computer in 1984 that could do virtually every task in his job. He was not replaced. He ran bigger models and wrote more papers. Exposure can be good.
Jobs also are not fixed task lists. White-collar workers have fuzzy goals and constantly invent new tasks. That is why he values OpenAI's GDPval, a benchmark of real work products created by experienced professionals across 44 occupations, while stressing that jobs keep changing, maybe "faster even than AI." He also argues O*NET is "not fit to purpose" and has urged the Labor Department to build new measurement infrastructure, possibly internationally, since early warning signs may appear in Philippine call centers or Indian outsourcing firms before they appear in US data.
The numbers he is watching
In a rapid-fire round, Chatterji shared a few personal bets:
- The weekly number he watches: revenue per employee at a set of AI-exposed companies, as an early "canary" for productivity. Some industries he expected to move are not moving fast, he says, and some "sleepers" are.
- The consensus he thinks is wrong: "The path to AGI runs through the organization." Organizations, which he calls humanity's greatest invention, decide how fast technology is adopted.
- US unemployment at the end of 2030: he hopes for 5% or lower.
- Daily AI use by US workers in 2030: up from about 40% today, citing survey work by David Deming and co-authors, to around 80%, possibly higher once AI is invisible inside other software.
- What the economy can do in 2036 that it cannot today: run agentic organizations, possibly smaller than the grandest visions, with compute cost the main constraint.
His biggest surprise since joining OpenAI is not economic. It is how fast capabilities improved. When he started, AI could not reliably do his daughter's math homework. Two years later, he says, AI systems are helping solve problems that stumped humans for a century. Many people's view of AI is 18 months out of date, he argues, so it has to be "reintroduced" constantly.
Why this matters
Chatterji's view is a useful middle path between two loud camps.
- Against the doomers, he argues mass unemployment within 18 months would require an economic model very different from historical evidence, and that care, education and entirely new categories of work can absorb people.
- Against the boosters, he warns that even small, concentrated job losses can reshape politics for decades, and that productivity gains will not be automatic. They depend on organizations adapting.
He also has a structural point that matters to every business owner: the bottleneck is less the model than the company using it. Connectors, workflows, rules about agent output and management practices decide who captures value.
Practical takeaways
- Connect your AI. If you only chat with AI, you are leaving most of the value behind. Linking it to your files, calendar, CRM or codebase is the step most firms have not taken.
- Expect your role to stretch. Work outside your job description is becoming the norm. Learn the adjacent skills AI makes accessible, from data analysis to marketing.
- Use AI to test ideas early. Getting feedback on a business idea is a classic failure point. Ask AI to critique it, then validate with real customers.
- Set rules for agent output. Decide who reviews AI-generated work and how much can be pushed onto colleagues.
- Watch care and education. If Chatterji is right, these sectors grow, and AI tools that make care workers more productive could be a major market.
Counterpoints and risks
- He works for OpenAI. Chatterji acknowledges the credibility challenge and tries to address it through peer-reviewed publishing, outside collaborators and funding independent research teams. Readers should still weigh his optimism accordingly.
- The early-career signal is real. Stanford's research finds employment for workers aged 22 to 25 in highly AI-exposed occupations has fallen well behind comparable peers, mainly through slower hiring. That pain is concentrated and could grow.
- Care jobs may stay underpaid. His optimistic scenario depends on policy choices, including taxes, that may never happen.
- Consumer surplus is hard to measure. Big estimates of free value are uncertain and do not pay rent.
- Predictions get revised. He chose to focus his team on present-day data precisely because long-range forecasts keep slipping.
FAQ
Who is OpenAI's chief economist?
Ronnie Chatterji, a Duke University professor who previously served as the White House's CHIPS coordinator. OpenAI named him its first chief economist in October 2024.
When does he think AI will show up in productivity statistics?
He expects a more significant signal within the next 12 to 18 months, while noting that AI is already contributing to GDP through infrastructure investment.
Does he think AI will cause mass unemployment?
Not in the near term. He hopes US unemployment will be 5% or lower at the end of 2030, and argues policy can soften transitions if disruption is larger.
Which jobs does he think will grow?
Education and healthcare in particular, plus new entertainment and creator roles, because richer societies spend more on care and people prefer human teachers and caregivers.
What is GDPval?
An OpenAI benchmark built from real work tasks created and graded by experienced professionals across 44 occupations, designed to measure AI on economically valuable work rather than exams.
The bottom line
Chatterji's forecast is not a utopia and not a collapse. It is an economy where AI shows up first in investment, then in productivity, while the real fights happen in specific towns, industries and job levels.
His most important idea may be his simplest: the path to AGI runs through the organization. The models are racing ahead. Whether that becomes prosperity depends on how quickly companies, governments and safety nets learn to keep up.
Sources
- OpenAI, Dr. Ronnie Chatterji named OpenAI's first Chief Economist
- Reuters, OpenAI hires ex-White House official as chief economist
- Duke Fuqua School of Business, Aaron Chatterji faculty profile
- OpenAI, GDPval: Measuring the performance of our models on real-world tasks
- Stanford Digital Economy Lab, Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI
- Stanford Digital Economy Lab, The AI employment gap for young workers has widened to 19%





