Managing AI Agents Will Become a Core Leadership Skill

For more than two decades, my career has revolved around managing people. I’ve led individual contributors, managers, distributed international teams, technical experts, software engineers, technical writers, and program managers spanning different functions and experience levels. Over time, I learned that effective management is rarely about handing down identical instructions to everyone; it's about understanding what a person needs from you in a specific context. Sometimes that requires directing or coaching, while other times it means supporting someone who needs confidence and air cover, or simply delegating the task entirely and staying out of the way. This philosophy, rooted in situational leadership, has guided how I evaluate the person, the task, their confidence, and the consequences of getting it wrong. Lately, I find myself applying these exact same management instincts in a vastly different setting.

As I build Cert Buddy™, an AI-powered certification-preparation product, I'm not managing a traditional team of employees. Instead, I work daily with a constellation of AI coding agents, research tools, content generation systems, and analytical models that help me interpret user feedback, inspect product flows, draft implementation plans, and ship new features. This shift made me reflect on whether managing AI is an entirely new discipline or merely an extension of the management frameworks many of us already practice. My conclusion so far is that it's fundamentally both.

AI Still Requires Real Management

It's tempting to view AI as a simple tool that waits for a prompt and returns a clean answer. In practice, building a complex product alongside AI feels less like operating a calculator and far more like managing a blisteringly fast, highly capable, and occasionally overconfident contributor. An AI agent can produce an astounding volume of work in a short window, but it can also misinterpret primary goals, optimize for the wrong metrics, introduce flawed assumptions, or deliver a technically sound feature that completely misses the mark for user experience.

Because of these tendencies, tasks cannot simply be handed off and forgotten. I still have to define the underlying problem, establish architectural context, set explicit boundaries, review outputs, provide targeted feedback, and decide if the code is actually ready to ship. I still have to manage. The key difference is that AI demands a distinct adaptation of familiar management styles.

Adapting Situational Leadership to AI

The traditional situational leadership model relies on four core postures: directing, coaching, supporting, and delegating. I have found all four to be remarkably useful when guiding AI systems, even if they manifest differently than they do with human teams.

• Directing: When a task is high risk, precise, or bound by existing product designs, ambiguity becomes exceptionally expensive. I must explicitly define what should change, what must remain untouched, what success looks like, and how the output will be validated...down to specific screens, edge cases, and acceptance criteria. Unlike a human contributor who might pause to ask a clarifying question when hit with uncertainty, an AI model will often confidently fill in the blanks with its own assumptions.

• Coaching: When I know the target outcome but need to explore the best path to get there, I engage the model in an iterative dialogue. I ask it to analyze current flows, evaluate trade-offs, challenge my assumptions, and propose alternative approaches. For instance, while redesigning Cert Buddy’s onboarding, I leveraged AI to map user journeys, synthesize feedback, and highlight friction points. I retained full decision-making authority, but the AI helped me move rapidly from raw evidence to actionable strategy.

• Supporting: Supporting human employees often centers on active listening, building confidence, and removing institutional roadblocks. AI doesn't need emotional reassurance, but its equivalent is system context. Providing a model with broader perspective - such as product goals, user personas, past test results, established terminology, and brand tone - enables it to make far better local decisions and prevents it from solving an isolated problem while breaking the broader user experience.

• Delegating: Contained tasks like analyzing a dataset, generating test cases, refactoring modular code, or drafting initial features can be delegated for rapid execution once constraints are established. However, delegating to AI is distinct from trusting a human peer. With experienced humans, delegation is rooted in demonstrated judgment and shared accountability. AI handles execution; human leaders own accountability. And Responsible AI demands that final outcomes remain entirely on us.

  
  

The Judgment and Empathy Gap

The most profound difference between managing people and managing AI comes down to human judgment. High-performing employees understand institutional history, team dynamics, customer sensitivities, and the strategic risks of a technically viable option. Crucially, humans experience the consequences of being wrong...they care about user trust, organizational reputation, and team friction. AI can simulate analytical reasoning, but it carries no accountability, worries about no mistakes, and has no reputation to protect.

The human leader must permanently act as the judgment layer, deciding what is actually appropriate for the product and its users.

Similarly, emotional intelligence remains indispensable because the end product ultimately serves real people. When directing AI to write tutor explanations, design study plans, or frame feedback on missed practice questions, I'm shaping how a learner feels in a moment of frustration. The AI generates the copy and interactions, but the empathy must be specified by the person directing the technology. Emotional intelligence hasn't lost its utility; it has simply become part of the software specification.

Speed Demands Greater Leadership Discipline

AI-assisted development drastically compresses the timeline between an initial idea and a live product update. During a recent iteration of Cert Buddy’s first-run onboarding, I moved from analyzing user feedback to prototyping a revised flow and re-testing it with users in under 24 hours. In a traditional corporate environment, that process might take weeks of stakeholder reviews, alignment meetings, design handoffs, and prioritization queues.

While collapsing this distance removes organizational friction, speed brings its own operational risks. A flawed assumption or bad product decision can be propagated across an entire codebase in a matter of hours. The ability to move fast makes clear direction, rigorous code review, thorough testing, and steady judgment more vital than ever. AI removes execution friction, but it increases the necessity of strong leadership.

The Future of Blended Workforce Leadership

Managing AI will not replace managing people. Organizations will always depend on human leaders to establish long-term vision, build trust, resolve interpersonal conflicts, navigate ambiguity, and make hard calls with incomplete data. However, leaders will increasingly oversee a blended workforce made up of human teammates, AI agents, automated workflows, and domain-specific models.

The most effective future leaders may not necessarily be the best prompt engineers. Instead, they will be the ones who excel at deciding what tasks to delegate, what processes require oversight, where human judgment is non-negotiable, and when an AI-generated answer must be challenged. They will assign work thoughtfully, supply rich context, inspect results relentlessly, and hold themselves accountable for the final output.

As I navigate this shift in my own work, I continue to rely on the core leadership principles developed over the last twenty years: clear expectations, deep context, continuous feedback, and sound judgment remain paramount. AI is not an employee, and treating it like one is a mistake. It does not require empathy, but the people using its outputs do. It can execute rapidly, but it cannot own outcomes. Ultimately, the emerging leadership challenge is learning to apply our established management skills to a medium that operates fundamentally differently, constantly asking ourselves: What does this technology need from me to perform at its best, and what must remain entirely my responsibility?