I Put an AI to Work and Learned That Leadership Beats Prompting

When I first started using AI, I figured it was just one more tool in the box.

Developers used to write the code themselves. Now we ask AI to write it. That was the whole difference, as far as I could tell.

Which led me to this conclusion:

"From here on, the people who write good prompts will be the ones who get things done."

I spent months on that theory. If I could just polish my sentences well enough, the output would improve. Looking back, most of what I was chasing during that stretch was surface.

Then one night I had four or five work windows open, each running a different task, and I looked at what I was actually doing. It wasn't sentence polishing.

The skill of using AI well sits a lot closer to leadership than to prompt writing.

Delegating well is its own skill

Think about what a leader does at a company.

Good leaders don't do everything themselves.

They set a goal. They break down the work that gets you to that goal. They hand each piece to whoever will do it best. They check intermediate results, correct course when the direction is off, and reshuffle assignments when something breaks. At the end, they stitch everyone's output back into one thing.

Laid out in order:

Set the goal → break down the work → assign roles → execute → observe → review → readjust → integrate

At some point it hit me. That was exactly what I'd been doing with AI.

Using AI well isn't only about asking good questions

Say I'm building a new service.

I could just tell the AI:

"Build me this service."

Something will come out. But I'd rather split the work up first.

Somebody has to research the market and the users. Requirements need to be written down, the service structure designed, a data model defined. Implementation and testing follow, and there's still review from a security and operations angle.

One big problem becomes a pile of small ones.

Only then do I hand out roles. One window analyzes the market. Another writes up requirements. A third reviews the architecture, and sometimes I set up a separate one whose entire job is to tear the implementation apart from a different angle.

How many AIs you're running doesn't matter.

What matters is that a human decides who gets what, in what order things run, and where the output gets verified. That's where it splits.

At that point calling it prompt engineering feels a little thin. It's closer to running an organization.

Good leaders don't delegate everything

One thing worth pinning down here.

Reading leadership as nothing but "the ability to make other people do things" gets you in trouble.

Same with AI.

If a task takes me 10 minutes to do myself, and instead I spend 20 minutes explaining it to an AI plus another 10 verifying the result, there's no reason to delegate. It took me a few rounds of that before I backed off. I can still feel the mood of the day I spent two hours explaining a script that would've taken 30 minutes to write.

Flip it around — three hours of research, five hours of development, two hours of testing, sliced up properly and run in parallel — and it's a different story.

So an unexpected skill turns out to matter a lot when working with AI.

Allocation.

What do I do myself? What goes to the AI? What runs concurrently? Which results do I absolutely have to inspect myself?

People who use AI well aren't the ones who hand everything to AI. They're the ones who draw a good line between what they keep and what they pass along.

Same thing a good leader does in an organization.

The way I worked as a developer isn't the way I work now

I started as a developer.

I built frontends and backends, then moved through financial systems and APIs, platforms, MSA, cloud, and Kubernetes.

As the years piled up, the role kept shifting. Not just building, but looking at structure from an AA and SA perspective, dividing work from a PM seat, and — in ISP and consulting — having to define the problem itself.

The order of my thinking changed around then too.

I used to hit a problem and immediately think "how do I implement this?" Now it's usually "how do I split this up?"

Who should do it. How much to delegate. Which tasks can run in parallel. Where the dependencies are. What result I need before the next step can start.

Working with AI, that same mental habit transferred over intact.

Which might be why I treat AI less like a search tool or a coding tool and more like someone I'm working alongside.

You become a leader even working alone

As AI agents get better, this shift gets much bigger, because one person runs several agents at once.

One researches, one develops, one tests, and another one keeps chewing on the results.

The person moves out of the Worker role — doing every task by hand — and into a coordination seat.

Put in terms developers know: you move from Worker to Control Plane.

A Control Plane doesn't process workloads itself. It decides what runs where, watches the state, and reconciles when desired state and actual state drift apart.

Running AI agents starts to look a lot like that. Define the target state, place the work, look at the results, adjust when something fails.

After all my years around Kubernetes, that picture looks oddly familiar.

Which may mean leadership gets more valuable, not less

Some people say better AI means we'll need fewer leaders.

I see it the other way.

The cheaper execution gets, the more valuable it is to decide what to execute.

Say the cost of producing code drops from 100 to 10. The bottleneck isn't how fast you can produce code anymore. The bottleneck becomes deciding what to build.

Ten agents don't automatically make you 10x more productive either. Point them the wrong way and you get wrong results 10 times faster.

So there's still a human share.

Defining the problem. Dividing the work. Setting priorities. Allocating. Verifying results. Catching a wrong direction before it runs.

These are the things we've been calling leadership all along.

Organizing intelligence

The Industrial Revolution gave humans more force. Computers gave us faster calculation. The internet gave us nearly unlimited access to information.

AI gives us something different.

Intelligence you pull off the shelf when you need it.

If that's true, the next round of competition isn't decided by who has memorized more. It probably isn't who uses AI the most, either.

How well you can move several intelligences toward a single goal. I think that's where it gets decided.

Leadership in the AI era isn't a skill reserved for CEOs and executives. The moment a single developer runs several agents, they're leading a small organization. Same for PMs, designers, researchers.

Maybe AI isn't erasing human leadership so much as taking what used to be demanded of a handful of managers and demanding it from everyone. I'm not sure yet. But in front of my monitor, it's already going that way.