What the research actually says.
Short, sourced reads on the latest reports and the science behind them. What each one found, what it means, and the practical takeaways, tips, and risks.
New data from Apollo, built on observed AI use across 321 occupations, finds AI-exposed jobs lost 6.7 percent of real wage growth since 2023 with no measurable job losses. The productivity is real. It is showing up as smaller raises, not layoffs, and it lands hardest on lower-paid work.
MIT's State of AI in Business 2025 found that despite 30 to 40 billion dollars spent, 95 percent of organizations see no measurable return on generative AI. The tools that fail do not fail on model quality. They fail because they do not learn, adapt, or fit the real workflow.
Gartner predicts over 40 percent of agentic AI projects will be cancelled by the end of 2027, on cost, unclear value, and weak controls. Most of what is sold as an 'agent' today is a chatbot with a new label. The projects that survive are narrow, measured, and rebuilt around the work.
Google's AI and Economy Atlas finds AI usage across 68 percent of occupations, covering 90 percent of employment, yet inside a job it touches only about a fifth of tasks and rarely automates anything. Adoption is a mile wide and an inch deep, and most of it happens outside work.
Anthropic's Economic Index, drawn from real usage, shows AI moving from helping people do tasks toward doing tasks for them. Consumer use crossed from mostly help-me to mostly hand-it-off for the first time, and enterprise use is already three-quarters automation. Adoption is also wildly uneven by place and income.
Decades of research say the same thing. Skill learned away from the job mostly evaporates unless it is practised on real work, spaced over time, and reinforced. The popular 'only 10 percent transfers' line is a myth, but the underlying problem is real.
McKinsey's research finds roughly a third of organizational transformations succeed. The ones that fail rarely fail on strategy. They fail on people, behavior, and the unglamorous work of building capability. Done rigorously, the success rate more than doubles.
Two randomized studies, one on about 5,000 support agents, one on professional writing, found the same thing. Generative AI raised output and, crucially, helped novices far more than experts, compressing the gap between weak and strong performers.
The field's most comprehensive annual read found 78 percent of organizations using AI in 2024, up from 55 percent a year earlier, and 252 billion dollars of corporate investment. Capability jumped on every benchmark. Trust, regulation, and workforce readiness did not keep pace.
McKinsey estimates generative AI could add 2.6 to 4.4 trillion dollars a year to the global economy, and that about three quarters of it sits in four functions: customer operations, marketing and sales, software, and R&D. The prize is not exotic. It is in the ordinary work people already do.
The IMF estimates almost 40 percent of jobs worldwide are exposed to AI, rising to about 60 percent in advanced economies. Exposure cuts both ways: roughly half could be helped by AI, half could see tasks displaced. The outcome depends on preparation, not fate.
A nationally representative US survey found that within about two years of ChatGPT's launch, 39.4% of adults aged 18 to 64 were already using generative AI, a faster take up than either the personal computer or the internet at the same point after release.
A global index from the UN's labour body finds about a quarter of the world's jobs sit in the exposure zone for generative AI. The headline is transformation of tasks, not wholesale replacement, and in high income countries women's jobs are far more exposed than men's.
PwC's analysis of close to a billion job ads and company results finds productivity growth in AI exposed industries has roughly quadrupled, workers with AI skills command a 56% wage premium, and employment is still rising even in the roles most open to automation.
Microsoft and LinkedIn's global survey found 75% of knowledge workers were already using AI at work, a share that had roughly doubled in six months, and that most of them were bringing their own tools rather than waiting for a company rollout.