What it means for work

Language models and agents are already changing what work looks like. Not all at once, not everywhere, not for everyone — but the changes are real and ongoing.

Tasks that previously required specialized expertise can now be approximated by someone with a well-structured prompt. First drafts, code skeletons, research summaries, translation, data formatting — these are all faster now for people who know how to work with these tools.

That's a shift in leverage, not a replacement. The model can draft; it cannot decide what to draft or whether the draft is any good. The model can find patterns; it cannot tell you whether the patterns matter. The model can help you work faster; it cannot tell you what work to do.

The jobs most affected are not necessarily the ones people expected. Many knowledge-work tasks that felt cognitive-but-routine — transcribing, summarizing, first-drafting, basic research — turn out to be automatable. Tasks that require judgment, relationship, physical presence, or genuine creativity are proving more durable.

What's clear is that fluency with these tools matters. Not coding, necessarily. Not knowing how transformers work in detail. But understanding what these systems actually are — what they can do, what they can't, where they fail — is already a practical skill.