EVERY MONTH IN 2026, I’ve received an email from Anthropic – the leading developer of frontier artificial intelligence (AI) – about their newest model and what it’s capable of. In July, I was informed about Claude Opus 5, which boasts major gains in coding and knowledge work: ‘It’s a clear step up on the work your team does most: higher-quality code, agents that run reliably for hours, and complex professional work like financial modeling and legal analysis with sharper reasoning across long documents.’ As someone who has built my livelihood and identity on my ability to reason sharply across long documents, I appreciate these monthly missives about my slow march towards obsolescence!

But even as the spectre of forced retirement hangs over us knowledge workers like the shadow of a guillotine, the promoters of AI insist there’s nothing to worry about. It will be wonderful, they say, to be freed from the yoke of menial drudgery – freed for higher-order tasks, freed to exercise judgement, to focus on work that is ‘uniquely human’! AI labs have been peddling this line to corporations with stupendous success. Anthropic’s annual run-rate revenue increased from US$1 billion in early 2025 to