Within two days of each other in late May, two pieces of research described the same professional from opposite ends. On 19 May, the Strada Institute for the Future of Work reported what nearly 1,500 executives now expect of an entry-level hire. On 21 May, researchers at Northeastern University reported where that worker's data goes once they are inside the building. Read together, the two findings sketch a single figure: a person being asked to bring more to work than ever, in an environment that increasingly treats their working behaviour as raw material.
Start with what is being asked. The Strada survey found that artificial intelligence is not hollowing out the entry-level job so much as redrawing it. Across industries, 42 per cent of employers said AI had increased the analytical and judgement-based responsibilities given to junior staff, while 41 per cent said it had reduced the foundational, skill-building tasks those roles used to hold. The report's lead author put it plainly. Entry-level roles are becoming more like mid-level roles.
This is a quiet but significant shift in what professional value means. The tasks that AI now performs well, the drafting, the formatting, the first-pass research, were never only output. They were the means by which judgement was acquired. When they are removed, the work that remains is the higher-order work: deciding what matters, reading context, knowing when a confident output is wrong. Employers say as much. Asked to rank the skills they want in graduates, they placed critical thinking and communication at the top and AI literacy at the bottom.
We have argued before that the structural conditions of a working environment shape the cognitive capacity a person can bring to it. The Strada data is that argument stated by the market itself. The capability that survives automation is judgement, and judgement is not evenly available. It is built, through exposure, through low-stakes practice, through the very tasks now being engineered out of the junior role. Raise the bar and remove the ladder at the same time, and you do not get better professionals. You get a narrower gate, open mainly to those who arrived with the experience already.
Now turn to the environment those professionals enter. The Northeastern team did what surveillance research has rarely done: it measured the destination of the data. Examining nine common employee-monitoring applications, used by employers from CVS Pharmacy to Dunkin', it found that all nine sent workers' names, email addresses and employer details to third parties including Microsoft, Google and Facebook. Records of online activity went to more than 145 domains. A third of the tools could track a worker's precise location even when running in the background.
The significance is not that workers are watched. It is that the watching does not stay in the room. A monitoring tool installed in the name of productivity becomes a pipe through which the texture of a person's working day, how they move, what they read, how they think through a task, is sent onward to companies the worker cannot name and never agreed to. Consent to be managed is not consent to be sold.
Now set both findings beside a third development. Through 2025 and 2026, all four of the largest accountancy and consulting firms have deployed agentic AI platforms, Deloitte's Zora AI, EY's agentic platform, PwC's agent OS and KPMG Workbench, built to carry out the analytical work their own people have always done. EY has reported that a model trained on the firm's tax work reached 86 per cent accuracy, well beyond a generic system. In January 2026, Deloitte said it would retire the traditional analyst-to-manager ladder for its 181,500 US employees, a change it tied directly to the arrival of an agentic workforce.
This is the step beyond monitoring. The Northeastern study shows the data trail of a working day leaving the building. The professional-services rollouts show the working method itself, the way an experienced person reasons through a problem, being digested into a system designed to reproduce it. What has barely been discussed is what staff give up in that exchange, and on what terms, when their ways of working become the training material for the thing that may later stand in for them.
A pattern emerges that no single development shows alone. The same professional whose value now rests on judgement, on the human capacity to interpret and decide, is working inside systems built to capture, model and commercialise exactly that. The market prizes the cognition. The infrastructure absorbs it.
Our position is that these are not separate stories but one. Each describes a quiet transfer of cost and risk onto the individual, made without a change in contract or a line in a job description. The entry-level hire is asked to arrive already formed, because the formation has been automated away. The experienced worker is asked to pour their craft into a model, and to accept that the record of their thinking belongs to others, because no rule says otherwise.
So the question is not whether judgement matters more now. The evidence says it plainly does. It is who is responsible for the conditions in which judgement can be formed and held, and what it means that the same moment asking more of human cognition is the one busy digesting it.
The contents of this article are for informational purposes only and do not constitute professional, legal, or financial advice.
