Why working with AI agents feels like leadership to me: setting direction, creating alignment, and reviewing execution without scripting every step.
In my last note, I wrote that working with AI feels more like leadership than coding. The Hacker News discussion pushed hard on one word: leadership. Several readers said I really meant management, supervision, or orchestration. That is a fair challenge, and it helped me clarify the distinction.
Supervision stays close to the task. It observes progress, compares the work with a standard, and corrects deviations. Management coordinates the system around the task: priorities, ownership, resources, dependencies, and time. Leadership sets the direction. It creates a mission, a vision, and enough shared understanding for others to act without waiting for instructions at every step.
All three matter. The difference is the level of autonomy they assume.
At my company, I try to practise servant leadership. My job is not to be the smartest person in the room or to prescribe every action. It is to provide context, remove obstacles, help people grow, and create the conditions in which they can make good decisions. That is close to intent-based leadership: make the goal and boundaries clear, then let decisions happen near the work.
Humans have agency. They bring judgment, ambition, emotion, responsibility, and their own point of view. AI agents do not have human agency or human needs. They are not colleagues, and servant leadership is not something we practise for a model.
What transfers is the operating pattern. An AI agent can choose actions inside a defined task instead of following a fixed flowchart. My role is to establish the outcome, provide the relevant context, define the boundaries, and explain what good looks like. That is much closer to setting intent than writing an algorithm.
The leadership task nearest to my heart is alignment. Many failures begin before execution, when the meaning in my head does not match the assumptions inside the team or the model. I often use Matt Pocock's Grill Me approach: ask questions until the hidden choices, dependencies, and trade-offs become explicit. With a large agent task, I may spend hours doing this before a group of sub-agents starts working.
Then comes execution. I review plans, intermediate decisions, tests, and final output. I compare the result with the mission and feed corrections back into the shared context. Current agent guidance calls these pieces instructions, orchestration, guardrails, and human intervention. The words are technical, but the responsibility feels familiar.
Supervision asks whether the work follows the expected standard. Management asks whether the work is organised. Leadership asks whether autonomous actors can move in the same direction without being told every step.
With people, that responsibility includes service, trust, and growth. With agents, it means context, constraints, and evaluation. These are not the same relationship. But both require me to make intent clear, create alignment, and remain accountable for what is ultimately delivered.