Will AI help you do your job or replace you? The BBC’s analysis examines how big AI promises translate into real effects on work, wages and hiring. AI firms have signaled that their software can automate some tasks while augmenting others, and the owners of the world’s largest companies are steering large sums into these tools with the expectation they can cut headcount or reallocate labor. The phrase “Flat is the new up” is echoed in investor conversations as firms weigh whether to hire new staff or deploy armies of AI Agents—virtual workers designed for specific tasks, some of them highly skilled. If even partial claims prove accurate, the impact would touch many sectors and individual careers, possibly sooner than expected.
Economists and policy researchers have begun to quantify the potential effects. Nobel laureates warned that action is needed to ensure AI raises living standards rather than exacerbating displacement. London businesses recently told researchers that they are struggling to find the skills needed as AI disruption accelerates. While it is early days, patterns are emerging in the data. A chart used by researchers shows that three years ago large language models (LLMs) could only reliably complete tasks humans accomplished in seconds or minutes. Today, LLMs are increasingly capable of handling fairly complex tasks that may take an hour for a human to complete. In some cases they can identify problems in a cryptocurrency contract and even streamline the model itself, a process that would have required hours from a specialist previously. The newest generation of models could potentially begin self-development within the next year or so.
The trend is not limited to coding. The same pattern appears in fields such as financial analysis, early-stage legal work, and even some entry-level creative tasks. In the United States, four years of data on employment outcomes by age and occupation most exposed to AI (for example software developers and customer service reps) show a decline in employment for 22- to 25-year-olds and a more pronounced hit for workers in AI-exposed sectors like finance, software, and creative industries. Stanford’s wage and jobs analysis indicates a 2.7% employment hit for 22–25-year-olds since ChatGPT became widespread, rising to 12.8% in the most AI-exposed sectors. Not all economists agree that AI is the only driver, with other factors such as rising interest rates cited as possible explanations. Online job postings have also been affected since the advent of ChatGPT and other LLMs.
Many factors influence outcomes, including interest rates and tax policy. The OECD has highlighted that differences in how exposed sectors are measured can affect results, with the UK showing notable vulnerability to AI-driven job losses in some measures when rates were stable or falling and before National Insurance changes. The UK’s service sector has a structure that leaves it particularly susceptible to AI-related shifts in employment. AI usage is tracked in tokens, small text chunks AI systems use for processing language, with one token roughly three-quarters of an English word. 2026 has seen a surge in AI use, and token growth has outpaced declines in per-token costs.
Top firms have deployed token leadership dashboards to push productivity gains, but the resulting high and sometimes unpredictable consumption of AI resources has led many to ration usage. The broader takeaway is that there are limits to how much work can realistically be automated and that the economics of virtual workers versus human labor vary by task. A notable caveat is that many Western companies are shifting toward cheaper AI forms based on freely available Chinese models, complicating certainty about future labor needs and costs. The picture remains mixed: automation can replace certain tasks and augment many others, with the final balance still unsettled as adoption scales across industries.
