As artificial intelligence spreads through workplaces, a recurring and uncomfortable pattern has emerged: in some roles, the people using or training AI systems are effectively helping to automate parts of their own jobs. A series of worker accounts gathered by The Guardian put a human face on that dynamic, describing feelings of being “devalued” and warning of a decline in the quality of work that automation was meant to improve.
The scale of the shift
The backdrop is a widely cited International Monetary Fund analysis estimating that AI could affect roughly 40% of jobs worldwide — and closer to 60% in advanced economies, where more roles involve the kind of cognitive tasks AI can perform. The IMF’s managing director, Kristalina Georgieva, has described the change as “like a tsunami hitting the labor market,” and has warned that it is likely to worsen inequality unless policymakers act, with entry-level positions and parts of the middle class especially exposed.
Training a replacement
Several workers described being asked to train AI tools that their employers hoped would eventually do the same work more cheaply. One content editor recounted earning less while spending long hours correcting the mistakes of AI-generated copy, and described feeling trapped in the arrangement by financial necessity even as colleagues left. The experience captures a tension at the centre of these accounts: the same people best placed to teach a system are the ones whose roles it is designed to absorb.
Where AI helped — and where it struggled
Not every account was negative. A palliative care consultant and professor at an NHS trust in Cardiff described willingly contributing to a pilot chatbot meant to help patients navigate the complexities of metastatic cancer and palliative care, recording guidance for hours and supplying agreed patient-information materials. That example points to genuine, carefully supervised benefits in sensitive settings — alongside practical limits, such as the system struggling with patients’ pronunciation.
A translator with years of experience offered a more sceptical view, saying that engines intended to replace human translators remained unreliable even after prolonged refinement. In his account, output tended toward formulaic results and still required word-by-word review, so the tools did not reliably save time and, in his judgement, lowered overall quality. The broad theme across these stories is that AI can handle the rough or routine version of a task while leaving humans to catch the errors that matter.
Limitations and what to watch
These are individual testimonies rather than a representative survey, so they illustrate experiences rather than measure how common each one is. The IMF’s 40% figure refers to jobs that could be affected — which includes roles that AI may augment rather than eliminate — and is a projection, not a count of jobs lost; estimates of AI’s labour-market impact vary widely between studies. The accounts also skew toward roles already exposed to automation, such as editing and translation, and may not generalise to every sector. What they do capture reliably is a real and growing source of friction: the gap between how AI is marketed and how it currently performs in everyday work.
The original interviews were reported by The Guardian, and the labour-market projections come from the IMF’s analysis of AI and employment. For related reading on this site, see coverage of who benefits from AI adoption and how AI tools keep changing.
