AI doesn’t take over whole professions. It absorbs work in a predictable order: execution before strategy, and necessities before advantages.
Execution goes first because the intent is already known, and turning a known intent into output is translation, which is what models are good at. Necessities go before advantages because necessities converge on the standard answer, and the standard answer is what an LLM produces.
The work that goes last is the work that depends most on unique context. That context is mostly illegible—tacit knowledge is hard to capture—and it’s relational, so a model only ever sees a thin slice of it.
This is a claim about order, not about limits. Context may become more legible as companies record more of their work, and models may get better at inferring what isn’t said. Whether “last” means never or in fifteen years, the line between what AI can and can’t do keeps moving toward the strategy end of the spectrum. Work that took judgment last year becomes execution this year.