The Manager Squeeze: Why Middle Management Is Becoming an Impossible Job

The organisation keeps adding responsibilities to the manager’s job. Almost nobody asks what should be removed.
A curious thing is beginning to happen inside companies. Work is being completed, decisions are being prepared and customers are being contacted, but not always by an employee. As AI agents become part of the workforce, organisations must decide where digital labour adds value and where human judgement must remain firmly in control. Until recently, workplace AI was largely something people used. An employee opened a chatbot, typed a question and received an answer. The technology might help draft an email, summarise a report or organise a set of ideas, but the person remained firmly in charge of the process. AI agents take that relationship a step further. Give an agent an objective and, within the permissions it has been granted, it can work out the steps, use different systems and carry a task through to completion. It might identify a promising sales enquiry, prepare a response, update the CRM and arrange a follow-up meeting. Another agent might monitor inventory, flag a potential shortage and contact approved suppliers for quotations. In HR, agents could coordinate interviews, answer employee questions, prepare onboarding documents, identify unusual workforce patterns or remind managers about actions they have failed to complete. Individually, none of this sounds revolutionary. Much of it resembles the automation companies have used for years. The difference is that an agent can exercise a degree of initiative. It does not merely follow one fixed instruction. It can decide what the next step should be. That is where the conversation becomes more interesting. If an AI system can receive work, make limited decisions and produce an outcome, is it still only a tool? Or has it become something closer to a member of the workforce?

What are AI agents in the workplace?

Microsoft has popularised the expression ‘digital labour’ to describe this emerging category. Its 2025 Work Trend Index found that 82% of leaders expected to use digital labour to expand their workforce capacity over the following 12 to 18 months. The appeal is easy to understand. Most companies want higher productivity, but their employees already feel stretched. An agent can operate at any hour, process large volumes of information and take repetitive work away from overloaded teams. It does not become bored halfway through reconciling records or forget to send the third follow-up email. But ‘digital labour’ is a deceptively simple phrase. Human labour arrives with a structure around it. Employees have job descriptions, managers, performance expectations and limits to their authority. There are procedures for appointing them, evaluating them and, when necessary, removing them. AI agents need the same clarity, yet many organisations are introducing them without it. Who is responsible if an agent sends incorrect information to a customer? Who notices if it repeatedly disadvantages a particular group of candidates? Who decides which company records it can access? At what point must it stop acting and ask a human to intervene? Saying that ‘the AI made the decision’ cannot become an acceptable corporate explanation. Technology can execute a decision, but accountability must still belong to someone.

How AI agents are changing the manager’s job

The arrival of AI agents will affect managers before it transforms entire organisations. Managers have traditionally distributed work among people. Soon, they may have to decide whether a task should be handled by an employee, supported by an agent or delegated almost entirely to one. That requires more judgement than it may appear. A manager must understand what an agent does well, recognise where context is missing and question an answer that looks convincing but may be wrong. They must also prevent employees from becoming so dependent on AI that they lose the ability to do the work themselves. This creates an uncomfortable possibility. In some companies, junior employees may understand the technology better than their managers, while managers remain responsible for approving its output. Organisations cannot solve that problem with a two-hour course on prompt writing. Managers need to understand the work itself, the risks around it and the limits of the systems operating within their teams. They will also have to manage the human response. Employees are unlikely to object to an agent that removes an irritating administrative task. Their reaction will be different if they believe they are training a system that will eventually replace them. Leaders often speak about AI as a way to ‘free people for higher-value work,’ but rarely explain what that higher-value work will be, how employees will be prepared for it or whether there will be enough of it to go around. People notice that gap. Honest communication will matter more than polished reassurance. If jobs will change, employees deserve to know how. If certain roles may disappear, vague promises about empowerment will not protect trust. Most people can cope with difficult information. What they struggle with is the feeling that important decisions are being made around them while everyone pretends otherwise.

AI agents and the future of entry-level work

The impact on entry-level work may prove to be one of the most important, and least anticipated, consequences of workplace AI. Junior employees have always been given the tasks senior colleagues no longer want to do: research, first drafts, meeting notes, scheduling, basic analysis and report preparation. Much of that work is repetitive, but it is also how people learn. A young analyst does not develop judgement only by attending training programmes. Judgement is built through small assignments, corrections, awkward client conversations and exposure to how experienced colleagues reach decisions. AI agents can now perform many of those starting tasks. That may make a department more efficient this quarter, but it creates a longer-term question: if the machine does the junior work, where does the junior employee learn? Early signs deserve attention. Anthropic’s 2026 labour-market research did not find higher unemployment across the occupations most exposed to AI. It did, however, find tentative evidence of slower hiring among people aged 22 to 25 in those professions. That does not prove AI is eliminating the first rung of the career ladder. It does suggest that leaders should look more carefully before celebrating every reduction in junior hiring as an efficiency gain. A company can remove entry-level work faster than it can replace the experience that work used to provide. Five or ten years later, it may discover that it has fewer people capable of making the senior decisions that AI still cannot be trusted to make. The answer is not to preserve meaningless work for tradition’s sake. It is to make development more deliberate. Junior employees may need earlier exposure to customers, structured rotations, simulated decisions and closer coaching. They should learn how to direct AI agents, but also how to question them. Someone who can produce a sophisticated analysis in seconds but cannot explain where it came from is not yet more capable. They have simply acquired a faster way to appear capable.

Workforce planning in the age of digital labour

For decades, workforce planning has begun with headcount. How many employees do we have? How many do we need? Which positions can we afford? That logic becomes less useful when work can also be performed by contractors, outsourcing partners, automation and AI agents. The better starting point is the work itself. What does the organisation need to accomplish? Which activities genuinely create value? Which require empathy, negotiation, accountability or an understanding of circumstances that may never appear in the data? Which can be automated? Which should be supported by AI but remain under human control? Only after answering those questions should leaders decide how many employees or agents are required. This sounds obvious, but it demands a level of honesty that many transformation programmes avoid. Some processes should not be automated. They should be stopped. Giving an agent responsibility for an unnecessary report does not make the organisation more productive. It simply allows the organisation to produce unnecessary reports more quickly. There is also a danger that AI will increase the amount of work rather than reduce it. An agent can generate dozens of presentations, summaries, messages and recommendations in the time it once took an employee to produce one. Other employees must still read, evaluate and respond to that output. A workplace already overwhelmed by information does not automatically benefit from producing more of it. Productivity should therefore be measured through useful outcomes, not volume. Did the customer receive a better answer? Was a problem resolved faster? Did the decision improve? How much human correction did the agent require? What did employees do with the time that was supposedly saved? Without those questions, companies may confuse increased activity with increased value.

Why HR must help govern AI agents

Technology teams will naturally lead the selection, security and integration of AI systems. But the moment an agent begins to influence a job, assess an employee or make a decision, it enters HR territory. HR should be involved in deciding how roles change, which decisions require human oversight and what new capabilities managers need. It should examine whether AI affects particular groups of employees more heavily than others and whether career paths remain viable after routine work is removed. Companies may also need something that sounds unusual today but could soon become standard: a register of their AI agents. Leaders should know which agents are active, what each one does, which systems it can access, who is responsible for it and when its performance was last reviewed. Organisations would not allow an unidentified employee to move across departments, access sensitive information and communicate with customers without supervision. There is no good reason to apply a lower standard to a digital worker simply because it is less visible. Yet HR must avoid becoming only the department that writes policies after the technology has already arrived. Its more valuable role is to help the business design a workforce in which people and agents contribute differently, but deliberately.

Building a human-centred AI workforce

The International Labour Organization estimates that one in four jobs worldwide is potentially exposed to generative AI. It also concludes that transformation, rather than outright replacement, is the more likely result. That is encouraging, but transformation is not automatically positive. A job can be transformed into something more meaningful, or into a fragmented role where an employee spends the day correcting machines and carrying responsibility for decisions they did not really make. AI agents could remove dull administrative work, help small teams perform at a much larger scale and give people more time for creativity, relationships and judgement. They could also weaken career pathways, blur accountability and create workplaces in which no one fully understands how decisions are reached. Both futures are plausible. The difference will not be determined by the sophistication of the technology alone. It will depend on the quality of the decisions organisations make around it. The companies that benefit most will probably not be those that deploy the greatest number of agents. They will be the ones that know where an agent adds value, where a person must remain in control and when the most intelligent decision is not to automate at all. Your newest colleague may not be human. But somebody human must still decide what kind of colleague it will be.