Three design problems from Anthropic’s workforce data

Anthropic’s report on its own workforce is the clearest data so far on what happens to an organisation when AI does most of the execution.

Agents now log six times more working hours than employees. The work left for people is judgment, coordination and decisions, and the report shows three design problems that follow.

First, the pipeline. Junior technical hiring fell from 27% to 9% of hires in a year. Entry-level work was where people learned the judgment that senior roles now depend on. Remove it and the question becomes where your future leaders build that judgment.

Second, redeployment. Employees were 50% more likely to move departments this year, and whole teams were moved into new work. Anthropic built an early warning system that tracks role-level productivity against AI capability, so people move before their role erodes. In most organisations it’s unclear whether that sits with HR, finance or the business.

Third, visibility. AI works well only when decisions and context are written down and shared. Decisions made in private conversations and unlogged meetings are invisible to it. One researcher said this forces leaders to give feedback in the open. That could be a trade-off against candour and psychological safety, and it’s the leaders’ decision.

One gap the report admits: Anthropic looks after employees whose jobs change by moving them into new roles. That does nothing for the junior people it no longer hires.