Most Work Is Translation (aparnacd.substack.com)

The piece argues that most knowledge work is really translation—converting information from one form to another so someone else can act—and that large language models have become the first “universal translators” for those tasks. It points to hard numbers: only 39% of the workday is spent on core duties, knowledge workers spend ~57–60% of time coordinating, developers write new code just 10–16% of the week, and meetings consume about 14.8 hours/week (~37% of working hours). LLMs can automatically turn a 20‑page paper into a one‑page “so what” memo, a 60‑minute call into a four‑minute decision brief, spreadsheets into charts and headlines, whiteboard photos into roadmaps, and requirements into code scaffolds with tests—drastically lowering the unit cost of these translation tasks. Technically and organizationally, that collapse in translation cost implies a redesign of teams: the thick middle layers—managers, coordinators, analysts—exist largely to pay translation overhead, and when AI internalizes that work the middle can compress into AI infrastructure. Individuals shift from drafting to judging and decision‑making; teams shorten coordination loops and raise throughput; smaller, leaner firms can coordinate at scale. Humans remain essential for judgment, taste, and accountability, but the “blank‑page tax” is nearly gone, reshaping how work is produced, reviewed, and scaled.