The share of graduate job postings in China requiring artificial-intelligence skills has risen to four in ten, while average employee training hours in Hong Kong have climbed to a 14-year high of 19.4 hours per year, as companies across Asia scramble to adapt their workforces. The Beijing-based recruitment portal Maimai reported that nearly 40 per cent of vacancies for new graduates in the first five months of 2026 were AI-related, up from roughly 30 per cent a year earlier. In Hong Kong, a survey by the Hong Kong Institute of Human Resource Management covering 127 firms and nearly 80,000 full-time staff found training hours rose 6.8 per cent from 2024, driven by a dual emphasis on AI and soft skills.

Yet the economic picture remains stubbornly opaque. In Washington, a bipartisan group of senators introduced a bill in June that would require the federal government to produce an annual report on AI’s effect on the workforce, after researchers documented the difficulty of measuring the technology’s spread in real time. The US Census Bureau has been asking businesses about AI use in a fortnightly survey since 2023, but estimates of adoption vary sharply depending on how questions are framed. A Goldman Sachs analysis projects that 9 per cent of the US workforce—roughly 15 million people—will be displaced, though the bank’s economists argue that history shows new positions reabsorb displaced workers. MIT’s Neil Thompson notes that capability is only the first step: an AI system also needs access to the right data and must be cheap enough to run, hurdles that slow adoption well below what the technology can technically do.

Across economies, the mismatch between AI’s potential and the readiness of institutions is widening. A survey of more than 12,000 executives, cited in Brazilian business coverage, found that 99 per cent expect AI to reduce headcount within two years, with the largest gains coming from redesigning work and developing competencies. In Stockholm, a Microsoft Sweden report shows over half of workers now perform tasks they could not do a year ago, yet only one in five AI users say their managers provide clear leadership on the technology. In Brazil, teachers are adopting AI to cope with heavy workloads without systemic support from school networks. In Paris, mastery of AI has become a hiring criterion for executive assistants, whose proximity to strategic decisions and confidential data makes the role particularly sensitive. In Moscow, recruitment platforms report a 24 per cent rise in IT CVs but 93 per cent of companies say candidates lack necessary skills, pushing employers towards practical assessments and skills-based hiring rather than relying on résumés.

The push for better measurement is now colliding with efforts to make AI itself more reliable. Laboratories including Anthropic, Google DeepMind and OpenAI are investing in confidence calibration, teaching models to estimate their own uncertainty and, when appropriate, to say they do not know rather than produce a convincing but incorrect answer. The US Senate bill, sponsored by Democrat Mark Kelly, marks the next concrete milestone in the data-collection effort; its progress will determine whether policymakers gain the tools they argue are essential to respond before the effects become entrenched.