How does one get the most out of an agent? By leading and managing it well.
How do you lead well?
-
My friend Rich described it well: Leaders call people towards a vision which is an attractive picture of an attainable reality. What do we want? Where are we going? Why? And then they set out first with their person and effort toward that vision asking the team to come with them.
-
I have told my teams: “We have a clear vision. My job now is to empower you to lead your teams to achieve that vision. I will give you the information, tools, resources you need to accomplish it, and check in with you at the right cadence for feedback, prioritization, and course adjustment as no plan survives contact with the battlefield.”
How do you manage well?
-
Managers define the work, help it match to the vision, break it into the right phases and chunks, iterate with the team, refine with team as the facts on the ground come in
How might you manage a team of agents in a sculpting project?
Let’s consider an example. Suppose you have a lump of marble and you want to lead and manage your team of agents or people to create a beautiful statue out of this rock.
A Bad Leader doesn’t understand where the project is going, doesn’t do any work themselves. They give a vague prompt like “Turn that rock into a statue”
A Bad Manager likewise isn’t able to iterate with the team as its making progress. They are just running the schedule and managing the reporting structure. “How’s it coming? I need to give status.”
When working with an agent, this looks like a vague short prompt that you one-shot hoping the agent will save your butt and just make a beautiful status leap out of the rock.
And sometimes it does.
A good leader has a vision of where to go. Which means they must have a deep understanding of what is possible, what they want, and they must describe it in sufficient detail.
A good manager must understand the plan, see the changes, iterate with the person being managed, guide toward the vision, update the vision based on what the work shows, and iterate with the team.
When working with agents, this looks like at least a detailed prompt, but more likely a detailed plan document in markdown where you iterate with the agent for a while honing and refining the vision before you even begin. The more time spent in that plan, the clearer the vision, and the better result. The more time spent learning and thinking about where you are going, even as you go there, the better.
The richer, relevant context that you can provide to the agent about your vision and the more you can together shape the plan, the better job it will do.
Agents poorly prompted with limited context and limited planning can do surprising things. Agents well led with rich context, especially context that you have built by iterating with the agent, bring the best results.
Conclusion
This is what it looks like to lead and manage AI agents well. Leading well is hard work that requires investment and continual learning by the leader. I have found my own thinking and understanding deepening as I attempt to lead and manage agents well in every area of my work.
We are building Nimbalyst, https://nimbalyst.com as a platform for you to lead and manage agents (right now Claude Code) well by iterating together on your context, managing the work and the sessions. Check it out!
