The Economist and the Machine
The IMF spent two years forecasting what AI would do to everyone else's jobs. With the launch of StatGPT, the machine is now inside the building.
MDB jobs disappear fast. Are you tracking the latest openings?
Subscribe to the Premium Plan and never miss an opportunity:
Get Staff roles on Mondays
Get Consultant roles on Fridays
Grab a free trial here to see what you’ve been missing (no payment required).
For two years, the IMF has been the world's most prominent forecaster of what artificial intelligence will do to jobs. In June 2026, the machine arrived inside the Fund's own headquarters. The launch of StatGPT, an AI platform for the world's official statistics, is a small event with a large meaning: the institutions that produce the global economy's numbers are now automating parts of that work, and the career ladder into them is being rebuilt rung by rung.
The announcement came in the routine flow of an IMF press briefing on 4 June 2026, sandwiched between travel schedules. Julie Kozack, the Fund’s communications director, informed reporters that Deputy Managing Director Bo Li would deliver opening remarks the following week at the launch of StatGPT, “the Fund’s AI-powered platform for official statistics.” It was the kind of line that slides past a room of journalists waiting for news about Argentina. It should not have.
The man chosen to open the launch tells you something about its weight. Bo Li is one of the more unusual figures in international finance: a former practicing attorney at Davis Polk in New York, a Stanford economics PhD with a Harvard law degree, a former deputy governor of the People’s Bank of China, and, in between, vice mayor of Chongqing, a municipality of more than thirty million people. Since 2021 he has been one of the IMF’s deputy managing directors, and he has spent the last two years narrating the Fund’s institutional bet on data. “What do we need to inform those actions?” he asked an IMF statistical forum in late 2024. “Data! Reliable, timely, accurate, actionable data.” It was at that same forum that he first described StatGPT publicly: an AI assistant “that lets users talk to data.”
The problem StatGPT exists to solve is one of the quieter scandals of the AI boom. Large language models, the engines behind ChatGPT and its rivals, are spectacularly bad at official statistics. The IMF’s own technical paper on the project, published in March 2026, put it with unusual bluntness: tested against real queries, general-purpose AI applications “frequently provide dangerously ‘reasonable’ but incorrect figures.” Bert Kroese, the Fund’s chief statistician, a Dutch mathematician who spent twenty-five years at Statistics Netherlands before taking over the IMF’s statistics department in 2022, described the failure mode precisely. Leading models, he wrote, consistently fail to reproduce accurate growth rates from the Fund’s own World Economic Outlook even when handed the source. “Most of the numbers are close but incorrect, which is arguably more dangerous than being wildly wrong: Plausible errors are harder to detect and more likely to mislead.”
StatGPT’s answer is a deliberate act of humility. The system does not let the AI generate numbers at all. Instead, the language model translates a plain-English question into a structured query against the official databases of the world’s statistical agencies, using SDMX, the data exchange standard sponsored by the IMF, the World Bank, the OECD, Eurostat, the European Central Bank, the Bank for International Settlements and the UN. The user gets the exact published figure, every time, with the conversational convenience of a chatbot. The Fund built it with EPAM Systems, a US-listed engineering firm whose Reliable AI Lab assembled the platform on its open-source AI DIAL framework; an earlier version was tested by more than a hundred representatives of the world’s major statistical institutions before this month’s launch.
What makes the launch a story for this newsletter is the institution it happened inside, and the research that institution has been publishing about everyone else. No organisation has done more to quantify AI’s threat to employment than the IMF. Managing Director Kristalina Georgieva’s January 2024 warning, built on the Fund’s staff research, became the global reference point: AI will affect almost forty percent of jobs worldwide, and about sixty percent in advanced economies. By this year her language had sharpened. At the World Governments Summit in Dubai in February 2026 she called the technology “like a tsunami hitting the labor market,” noting that one in ten job postings in advanced economies now requires at least one new skill. And at Davos in January she delivered the sentence that should be pinned above the desk of every young person reading this: “Tasks that are eliminated are usually what entry-level jobs present, so young people searching for jobs find it harder to get to a good placement.”
The Fund is now living inside its own forecast. So are its peers. The World Bank has built an internal AI tool called mAI, a ChatGPT-like system trained on its own institutional knowledge that staff use to search decades of documents and draft routine outputs. As Harry Daniel Lersch, an AI strategist at the Bank, told Devex: “The vision that we have for AI is that we use AI to accelerate, deepen, and sustain development impact.” Across the multilateral system, as Devex reported this year, dedicated AI teams have appeared in agency after agency while existing IT units pivot wholesale to the technology. And in February 2026, when the heads of the multilateral development banks held their first meeting under the Asian Development Bank’s chairmanship, with the IMF participating, they named their joint priorities for the year: jobs, artificial intelligence, water, critical minerals and nature. Masato Kanda, the ADB president and former top currency diplomat at Japan’s finance ministry who chairs the group this year, framed the agenda as bringing the institutions’ “combined institutional weight” to bear on shared problems. Jobs and AI, side by side, at the top of the list.
Walk the implications through the buildings. The work StatGPT automates, finding, retrieving, formatting and explaining official numbers, is precisely the work that has long occupied junior statisticians, research assistants and entry-level economists, the apprenticeship through which generations learned how the data actually fits together. The Fund’s newest staff research, published in January 2026, found that the diffusion of AI skills is associated with lower employment in highly exposed occupations where the technology substitutes for people rather than complementing them, and said plainly that this poses challenges for the young. The institutions are not hiding the shift. They are hiring for it.
Look at what is actually being recruited. This spring the IMF advertised for a section chief for data science in its technology department, a role spanning econometric modelling, machine learning and high-performance computing, working “closely with economists.” Its statistics department has been hiring data scientists for data management. The Fund runs a standing Tech Talent Pool to pull technologists into an institution built by macroeconomists. At the World Bank, the outcomes department has recruited data scientists to build applications using large language models for measuring development results, the development data group hires for geospatial and data management roles, and DIME, the impact evaluation unit, ran a 2026 recruitment round for research assistants, field coordinators and data science analysts. The roles rising in value share a shape: they pair economic literacy with the engineering skills to govern data rather than merely retrieve it. Metadata curation, API and data governance work, the unglamorous plumbing that makes official statistics what the Fund calls AI-ready, has gone from back-office afterthought to strategic priority. Kroese has said it directly: statistical organisations must compete for talent and invest in technology, and their data must carry well-structured metadata and programmatic access.
For candidates, the strategy writes itself. The old apprenticeship of pulling numbers and formatting tables is dissolving, so do not present yourself as someone who does what the machine now does. Present yourself as someone who makes the machine trustworthy. An economics degree plus demonstrated data engineering, a public repository, experience with statistical APIs or SDMX, a project where you validated or governed a dataset rather than just analysed it: that combination now clears a bar most traditionally trained applicants cannot reach. And the joint MDB agenda means this is portable across the system, from Manila to Abidjan to 19th Street.
Georgieva likes to end her AI speeches with a hedge that is also a challenge. “AI looks unstoppable,” she said in Dubai. “But whether or not countries can successfully capitalize on AI’s enormous promise is yet to be determined.” Inside the institutions she and her peers run, the people who will determine it are being hired right now. The machine has the numbers. The career belongs to whoever the institutions trust to stand between the numbers and the world.
Make sure you subscribe to MDB Jobs to get the latest vacancies delivered straight to your inbox each Monday, and consultant positions each Friday.
