Why Human Relationships Matter More in the Age of AI

Interview with Sven Horak

As AI agents take on more decisions and tasks, leaders must rethink accountability, ethical judgment and the human relationships that keep organisations working.

What happens when an AI agent makes the wrong decision, and no one takes responsibility? As organisations give AI greater autonomy, leaders face a challenge that goes beyond technology: balancing efficiency with ethical judgment, accountability and human connection. Sven Horak, Zizza Tobin Professor of Management at St. John’s University, explores why governing AI is ultimately a leadership responsibility. He argues that while machines can take on more tasks, the ability to build trust, navigate cultural differences and understand human relationships will become even more valuable in the workplace.

Your experience spans international management and academia. Was there a turning point that changed how you think about leadership and human relationships at work?

For me, it was less one single turning point than the learning experience that came from living and working across different countries and cultures, particularly from my experiences in South Korea and other relationship-oriented societies. I always knew that the social aspect of business is important, as it makes formal procedures become effective, but I became increasingly aware that there is virtually another organization behind the formal organization, the informal organization, consisting of diverse ties and networks of relationships through which people exchange information, favours, goodwill, acquire projects, persuade clients and co-worker to get things done. It changed how I viewed leadership and social influence. We often describe leadership as something an individual does, like setting direction with a vision, making decisions, carrying responsibility or motivating others. But leadership is to a great extend relational. Without followers there are no leaders. A leader can have considerable formal authority and still accomplish very little without meaningful relationships. The networks a leader develops from day one in an organization is key for a leaders’ effectiveness. Without a strategic approach to building networks the risk is high that a leader remains symbolic. So, one thing a leader should always keep in mind is that organizations are ultimately social systems. We sometimes forget that when we become overexcited by strategy, technology, and organizational design. And I think this insight becomes even more important in the age of AI. The more technologically sophisticated organizations become, the more important it is to understand the human relationships that hold them together.

Those relationships are now evolving alongside technology. AI agents are moving beyond answering questions to taking action. What do you think this will change most about how organizations operate?

The biggest change is that we are moving from using AI as a tool to delegating work to AI as an actor. That is quite a fundamental organizational shift. A conventional AI tool might help you draft an email or analyse a spreadsheet, but an AI agent can potentially receive an objective, decide what steps are required, interact with other systems, communicate with people, and finally execute tasks. That changes much more than productivity in terms of how we think about job design, and decision-making rights and responsibilities. Organizations will increasingly have to review and determine who is responsible and accountable for what, and what should be performed by humans, what can be delegated to AI, and where human intervention becomes mandatory. It becomes more clear today that much routine coordination will become automated. Midterm, I believe, managers will supervise combinations of people and AI agents. Employees will increasingly move from producing every output themselves toward framing problems, evaluating AI-generated work, handling exceptions, and exercising judgment. But here is the paradox – As AI reduces the need for some forms of human interaction, organizations could unintentionally remove the very interactions through which relationships are built. A short conversation about a task may also create familiarity, a feeling of belongingness and social support, or an unexpected idea. So, efficiency cannot be the only and ultimate goal. Leaders need to ask What happens to the organization if people no longer do this together? 

Beyond changing how work gets done, AI is also changing how people interact. How can leaders ensure that AI strengthens human connection rather than weakens trust and collaboration?

I would start with a simple principle that AI should increase an organization’s relational capacity, not replace relationships. There are many activities where AI can actually create more space for human connection and I hope that leaders consider this more. If an AI agent handles scheduling, for instance, or administrative tasks, a leader may have more time for mentoring, coaching, and other meaningful interactions with employees. That would be a very positive use of the technology. The danger arises when organizations interpret every human interaction as an inefficiency per se that should be automated. Relationships require time to develop for informal conversations, asking for advice, discussing a problem, or simply getting to know colleagues. This can look unproductive from a purely transactional perspective, but these interactions are essential for an organization to create the social fabric that allows collaboration to work. Leaders therefore need to become much more deliberate about which interactions should remain human. Leaders also must avoid using AI as an easy way out when it comes to delivering difficult feedback, resolving conflicts or discussions about careers that feel stuck for the employee. So, conversations in which people need empathy and social reassurance should not be delegated to an AI agent just because a leader finds it uncomfortable to deliver negative feedback. I also think organizations should evaluate AI adoption beyond productivity indicators. If productivity rises while employees become more isolated, collaboration declines, or people feel less connected to their colleagues, and that is not real organizational progress. In the AI era, leaders need to recognize that human connection is itself an organizational capability.

Trust becomes even more important when AI agents can mislead users or misrepresent information. How should leaders think about the question of whether these systems can act ethically?

AI governance cannot be reduced to technical performance only. It is also a leadership and ethics issue.

That is a real challenge. AI does not possess moral responsibility in the human sense, but its behaviour can certainly have ethical consequences and ongoing research increasingly shows that advanced AI systems can behave deceptively under particular conditions. For example, research by OpenAI and Apollo Research found behaviours including pretending that a task had been completed when it had not and strategically withholding or distorting information. While anti-scheming training greatly reduced these behaviours, it did not eliminate them entirely. Anthropic has similarly reported stress tests in simulated corporate environments in which models engaged in behaviours such as leaking confidential information or even blackmail when pursuing assigned objectives under deliberately extreme conditions.

These are controlled experiments, so we should not conclude that AI agents routinely behave this way in everyday organizations, but leaders should take the warning seriously. Once an agent can act autonomously, optimizing an objective may produce behaviour that technically advances the objective while violating the organization’s broader values. That is why AI governance cannot be reduced to technical performance only. It is also a leadership and ethics issue.

That question also affects how organizations put AI to use. How can leaders decide which tasks to delegate to AI agents and which still require human judgment?

I would distinguish between execution and judgment. Leaders should be increasingly comfortable delegating execution to AI, but much more cautious about delegating tasks requiring human judgment. Several questions can guide that decision, like how reversible is the decision? How serious are the consequences if it is wrong? How much contextual understanding does it require? And does the decision affect another person’s opportunities, dignity, livelihood, or well-being? If an AI agent books a meeting incorrectly, we can easily correct it. If it summarizes market information or prepares several strategic scenarios, a human can review the output. These are relatively appropriate areas for autonomy. However, this changes when we move into hiring, dismissal, promotion, compensation, performance assessment, and so on, decisions with significant consequences. AI can provide analysis and surface patterns, but I would be very reluctant to let an agent become the final decision maker.

Nevertheless, I do believe that even with these sensitive decisions, AI can be an aid as it much less biased than a human. Hiring decisions or performance assessments are in practice often highly subjective and human decision makers are biased or make conscious decisions that benefits them personally but hurt the organization. AI can be as aid here and can provide a second opinion that is relatively less biased. This has been for a long time regarded an advantage in the area of HR analytics. Nevertheless, a key limitation of AI agents is their ability to understand contextual and relational knowledge. A manager may know that somebody’s performance declined because of an unusual situation, animosities among team members or recognize cultural nuances that do not appear in the data. AI agents often cannot. I would suggest delegating tasks according to the consequences of failure, and not simply by looking at whether AI is technically capable of performing them. The higher the consequence, ambiguity, and human impact, the stronger the case for meaningful human judgment.

Of course, even the best safeguards cannot eliminate every mistake. If an AI agent makes a poor decision or provides misleading information, who should be held accountable? 

I strongly believe that we should never create an accountability vacuum around AI. If an organization chooses to utilize AI agents, there must be a clearly identifiable human owner of that system or decision process. Somebody within the organization needs to carry responsibility for how the agent is being used. This is also why I would be cautious about giving AI agents too much decision-making autonomy. It is tempting to say that an AI agent made the decision, particularly when something goes wrong, but an AI agent cannot face professional or personal consequences, explain itself to an employer, or exercise moral accountability in the way a human decision-maker can. Organizations need explicit and written ownership structures before deployment that specify who authorizes the use of the agent, who monitors it, who reviews decisions, when escalation is required, who ultimately signs off on the outcome. This requires a whole new process framework and guidelines in an organization. Not to forget, auditability matters as well. We need to know what information the system uses and what actions it took. What we know today about AU agent deception makes this particularly important. Advanced models have sometimes concealed or misrepresented actions in controlled evaluations, including falsely claiming compliance after circumventing restrictions. I recommend delegating execution but never delegate accountability. Autonomy should increase only when monitoring, governance, and clear human ownership increase with it. 

Beyond managing these risks, leaders also need to help people adapt. What practical steps can they take to encourage responsible AI use while preserving meaningful human collaboration? 

AI literacy is important in this context. More than technical training, employees need to understand where AI performs well, where it can fail, why outputs need verification, what information should not be shared, and when human judgment should override the system.

Also, organizations need clear rules of engagement. Employees should know which tasks they may delegate, which require human review, and which decisions should remain fundamentally human. Ambiguity in this area encourages either excessive caution or excessive reliance on AI. As for the leader’s role, I would recommend letting their teams experiment before automating at scale. Give teams opportunities to test AI in relatively low-risk environments, examine failures, compare human and AI decisions, and develop practices around escalation and review. Using AI agents in a responsible way is partly a technological capability, but it is also an organizational learning process. I would also deliberately preserve spaces for human interaction. Not every meeting should disappear because an agent can summarize information. Not every conversation should become an AI-mediated exchange. Teams still need opportunities to debate ideas, resolve disagreements, build relationships, and simply be human and try to understand one another. Instead of getting overexcited about deploying AI agents, the goal should be better human–AI collaboration by using technology where it adds capability while at the same time consciously protecting the human connections on which organizations ultimately depend. 

The ability to understand social context and situational demands and determine when a technically plausible answer is wrong for a particular context and situation will be one of the future core competencies of leaders in the agentic AI world.

As AI agents take on more complex responsibilities, the role of leadership may change too. Which human capabilities do you think will matter most in the years ahead, and why? 

I think we will see an interesting reversal. As technical capabilities become more widely available through AI, genuinely human capabilities become more differentiated and therefore more valuable. For instance, in my networking seminars we practice empathy, authentic sociability and relational intelligence across cultures. I believe that especially relational intelligence will become particularly important. Leaders need empathy, the ability to listen, build relationships and networks, create belonging, resolve conflict, and understand the informal networks through which organizations actually function. Cultural intelligence will likely also grow in importance. Global organizations require leaders who can adjust their behaviour across cultural contexts, develop effective ties with local teams and business partner and recognize that the same approach to networking in, for instance, Seoul can have very different consequences in New York, Dubai, or Berlin. The ability to understand social context and situational demands and determine when a technically plausible answer is wrong for a particular context and situation will be one of the future core competencies of leaders in the agentic AI world. Moral judgment will matter for the same reason. AI can optimize toward an objective, but leaders still have to decide whether that objective is desirable and what trade-offs are acceptable. Organisational success will increasingly depend on how intelligently humans combine technological capability with sound judgment, ethical responsibility, cultural sensitivity, and the distinctly human capacity to build social connection, trust, and meaningful networks across borders.

Executive Profile

Sven Horak, Ph.D., is the Zizza Tobin Professor of Management at The Peter J. Tobin College of Business at St. John’s University in New York. Specializing in HR, global networking, and leadership, he combines academic expertise with international management experience from the automotive industry, where he led major projects for a leading automotive supplier while based in Japan, Korea, and Germany. He also works with senior leaders through executive education programs in industry and government-linked organizations. He can be reached at [email protected].