Much of the public debate about AI focuses on models, laws and markets. Inside companies, the impact is more immediate and less abstract, it shows up in management.

Managers decide how work is assigned, how performance is judged, how exceptions are handled, how employees are promoted, and how conflict is resolved. As AI systems enter everyday business processes, those decisions will increasingly be shaped by automated recommendations, rankings, summaries, alerts and predictions. The manager’s role will become more politically important.

Training managers to use AI tools will not be enough. The deeper change is that managers will become the human interface between algorithmic systems and the people affected by them. They will need to interpret machine outputs, protect procedural fairness, challenge unreliable recommendations, explain decisions, and ensure that efficiency does not quietly override rights, dignity and accountability.

This shift deserves more attention in technology policy debates. AI governance is often framed at the level of legislation, corporate policy, model development or procurement but governance also happens at the point where a manager chooses whether to trust an AI-generated recommendation, whether to override a system, whether to disclose how a decision was influenced, and whether to give an employee a meaningful route to challenge the outcome.

This makes management part of AI governance, whether companies acknowledge it or not.

The workplace is where AI power becomes concrete

AI policy discussions often focus on platforms, elections, public-sector decision-making, or frontier model safety. These are important. But for many people, the most direct encounter with AI will happen at work.

In the workplace, AI is already moving into hiring, performance reviews, scheduling, productivity monitoring and risk triage. Some tools are harmless assistants. Others can influence pay, workload, reputation and career progression. Some systems will be simple assistants. Others will shape decisions with real consequences for pay, workload, reputation and career progression.

That matters because workplace AI does not enter a neutral environment. It enters a hierarchy. Employers hold information, authority and economic leverage. AI can increase that imbalance if workers are assessed by systems they cannot see, understand or contest.

The European Union’s AI Act recognizes this risk by treating certain AI systems used in employment, worker management and access to self-employment as high-risk. The logic is straightforward: decisions about work are not just administrative decisions. They affect income, mobility, reputation and a person’s ability to participate in economic life.

The same issue exists beyond Europe. Companies operating globally may face different legal regimes, but the ethical and managerial problem is similar. If AI influences decisions about people, the organisation must be able to explain how those decisions are made and who remains accountable.

Managers now have to judge the system too

Traditional management was already difficult. Managers had to coordinate people, allocate resources, resolve conflict, interpret strategy, and deliver outcomes under pressure. AI adds another layer: managers must now understand the behavior of decision-support systems well enough to avoid becoming passive conduits for them. Most managers do not need to build models, they do need to understand where the system gets its information, what kind of output it produces, what the output is suitable for, and where its limits sit.

If managers cannot question AI recommendations, probabilistic outputs can harden into managerial orders if they cannot explain how a system influenced a decision, trust weakens. And if AI is used to speed up performance management without context, judgment is replaced by administrative convenience.

The managerial skill set is therefore changing. Managers will need basic AI literacy and more importantly, they will need to understand accountability, evidence and escalation.

That includes the ability to ask whether the system is being used for its intended purpose, whether the data is appropriate, whether affected employees know how AI is being used, whether human review is genuine, and whether there is a route to correct errors.

Human oversight cannot be symbolic

Many AI policies rely on the phrase “human in the loop.” In practice, this can mean very different things. Sometimes it means a person genuinely reviews the output and has authority to reject it. Sometimes it means a person clicks approve because the system has already framed the decision. Sometimes it means human oversight exists on paper but not in the tempo of real work. Managers need to watch for behavioral distortion as carefully as they watch for productivity gains. AI systems can make some work more visible than other work, reward activity that is easy to measure, and make automated judgments feel harder to challenge. When that happens, the tool is no longer simply supporting management.

Human oversight is meaningful only when the human has time, authority, competence and organizational permission to disagree with the system. Companies can easily design AI workflows where the manager is formally responsible but practically constrained, the dashboard becomes the default truth, the recommendation becomes the safe option. The human review becomes a liability shield rather than a real decision point.

A serious AI operating model should avoid this. It should define when managers are expected to follow, question, escalate or override AI outputs. It should also protect managers from being punished for slowing down decisions when there is a legitimate concern about accuracy, fairness or context.

AI will change how accountability works inside firms

AI also complicates accountability, in a traditional organisation, a poor decision may be traced to a manager, a policy, a process or a lack of information. With AI, responsibility can become diffuse. The vendor designed the tool, a data team prepared the inputs, legal approved the procurement, IT integrated the platform, the business unit deployed it, and a manager acted on the recommendation. By the time harm appears, responsibility has been distributed across the chain.

When everyone is partly involved, no one may feel fully responsible.

Organizations need to resist the temptation to let AI create accountability fog. For every AI-supported management process, there should be a clear answer to three questions: who owns the use case, who owns the decision, and who is responsible for harm if the process fails.

This does not mean blaming individual managers for every system failure. Quite the opposite. Managers should not become scapegoats for poorly governed AI deployments. If senior leaders procure opaque systems, underinvest in training, ignore worker consultation or fail to define appeal routes, the accountability problem sits above the individual manager, but managers cannot be treated as passive users either. They are the point where abstract governance becomes lived experience. Their behavior will determine whether AI feels like a tool for better coordination or a mechanism of surveillance and control.

Worker voice must be part of AI adoption

One of the biggest mistakes companies can make is to deploy workplace AI as a purely managerial instrument. If AI is introduced only to increase visibility for management, automate supervision or extract more productivity, it will deepen mistrust.

Worker voice should be part of AI adoption from the beginning. Employees should understand what the system does, what it does not do, what data it uses, how outputs are reviewed, and how errors can be challenged. In higher-risk contexts, consultation should not be cosmetic. Workers often know where data is misleading, where workflows differ from formal process, and where metrics will distort behavior.

This is not just a labor relations issue. It is a quality issue. AI systems used inside organizations often fail because the official process and the real process are not the same. Workers can identify the difference. Ignoring that knowledge produces systems that look rational from the center and fail at the edge.

Managers need to watch for behavioral distortion as carefully as they watch for productivity gains. AI systems can make some work more visible than other work, reward activity that is easy to measure, and make automated judgments feel harder to challenge. When that happens, the tool is no longer simply supporting management it’s changing the terms on which people are managed.

What managers should prepare for

The practical implications for managers are already visible.

More decisions will need to leave a record. Managers will be expected to know when AI can inform judgment and when it cannot replace it. Performance management will become more contested, especially where workers believe systems are measuring activity rather than contribution. Managers will also need a working understanding of data provenance: not technical mastery, but enough knowledge to ask whether an output is based on reliable, current and relevant information.

The public interest question inside the firm

The debate about AI and democracy should not stop at elections or public platforms. Democratic values also depend on how power is exercised in economic life and workplaces are among the most important places where people experience rules, voice, fairness and accountability. AI makes management more transparent, consistent and evidence-based, it could improve organizational decision-making. It becomes a tool for opaque surveillance, automated pressure and unchallengeable evaluation, it will erode trust and concentrate power further.

The difference will come from choices made around the technology: whether workers have voice, whether managers can challenge systems, whether evidence is preserved, and whether accountability remains visible.

Companies preparing for AI should therefore stop treating managers as the final users of AI tools and start treating them as part of the governance system. They need training, authority, clear escalation routes and responsibility for preserving human judgment where it matters.

In complex organisations, the danger is rarely that a manager blindly trusts AI on day one. The more common risk is slower: the system becomes part of the routine, and questioning it starts to feel like creating friction.

AI will not eliminate management. It will test whether management can remain legitimate when decisions are increasingly mediated by machines.