Summer of agentic work: a labour market reckoning

The mid-year employment data published this week carried the first sustained evidence that agentic AI is restructuring the white-collar labour market, and the evidence has the texture of a reckoning arriving quietly. Hiring in the clerical and junior professional tiers is down for the fourth consecutive quarter; the hours of the workers who remain are up; and the productivity gains that the technology promised are showing up in the accounts of the firms that deployed it, while the firms that have only read about it are showing up in the same statistics as everyone else. The summer of agentic work has not produced a wave of dramatic headlines. It has produced a quiet restructuring, which is how labour-market revolutions usually arrive.

The pattern in the data is worth reading carefully, because it is not the pattern that either side of the debate predicted. The jobs being shed are not the routine manufacturing roles that the previous automation waves claimed; they are the entry-level professional positions, the research assistants, the junior analysts, the first-tier support staff, the jobs that used to be the training ground for the middle of the career ladder. The firms report that the agents handle the first pass of the work that these positions used to do, and that the remaining humans are being asked to do the parts the agents cannot, which is a description of a labour market that is thinning at its base and widening at its top. The result is a generation of workers entering a profession with fewer rungs on the ladder to climb.

The geography of the effect is as important as the numbers. The deployment has been concentrated in the sectors with the thinnest margins and the most standardised workflows, the back offices, the claims processing, the contract review, the data entry that feeds the machine-learning pipelines themselves, and the firms that moved early have posted margins that their competitors read with alarm. The laggards are not immune; they are merely later, and the data suggests that the delay is costing them market share rather than protecting their staff. The summer’s lesson for the C-suite is that the technology is no longer a question of whether, and the lesson for the workers is that the transition has no agreed schedule and no negotiated protections.

The policy response so far has been the usual mixture of the inadequate and the theatrical: training funds announced for the skills that the agents have already absorbed, tax credits for the hiring that the companies had already planned, and a regulatory conversation that has focused on the models themselves rather than on the labour market that the models are reshaping. The parts of the system that could actually cushion the transition, the portability of benefits, the wage insurance, the retraining pathways that lead somewhere real, remain underfunded and unfashionable, and the gap between the rhetoric and the funding is the most reliable predictor of how the next two years will feel.

The reckoning that the data describes is not a catastrophe, and the numbers are worth keeping in proportion: the white-collar workforce is still growing overall, and the agents are still doing the work that no human enjoyed. But the summer’s statistics have settled a question that the industry had been debating for three years, which is whether the agents would remain tools or become colleagues, and the answer, on the evidence, is that they have become a layer of the workforce that does not appear on the payroll. The firms that understand this are redefining their human roles around the judgement that the agents lack; the firms that do not are redefining their margins instead. The labour market’s reckoning with agentic work has begun, and it is being conducted, as all such reckonings are, in the small print of the quarterly reports rather than in the headlines.