Most of the analysis being published about AI and employment is answering the wrong question. Supply has eaten the conversation: which roles are most exposed, which tasks can be automated, which professions sit on safer ground. All of it rests on an assumption that the amount of work to be done stays roughly constant as the cost of doing it falls.

Culture comes first. Technology follows. There is no other sequence that works. Most organisations approach AI readiness backwards. They invest in infrastructure first, data platforms second, and culture last, if at all.

The boardroom conversation around artificial intelligence has, until now, fixated on model capabilities and upfront licence fees. This narrow lens is proving dangerously inadequate. Across FTSE 350 companies and mid-market PLCs, a silent cost escalation is underway, one driven not by the AI models themselves but, by the context they consume. Every document ingested, every customer interaction logged, every agentic workflow executed: each carries a chain of token consumption,