Why I think Leadership Matters More Now, Not Less

Most AI conversations I participate in are about the tools. Which model, which license, which platform, …

It’s mostly technical topics. Almost nobody asks who decides what the thing should do.

Photo by Donald Janssen on Unsplash

And that’s inherent to AI. AI makes things cheaper. But it doesn’t make decisions cheaper: Yes, ideas have become cheap, but execution isn’t. You still must decide which one of the dozens of options seems most reasonable and how to actually execute and operate it.

Not too long ago things like getting a decent draft, a first analysis, or an MVP of a script took real effort. Today it often just takes a prompt to kick things off. The number of ideas in many teams goes up.

But without execution an idea doesn’t change much. Someone must connect a solution to existing systems (trust me that’s often the most complicated thing) convince (i) the people who will provide a budget and (ii) the people who will use the solution. That hasn’t gotten faster with AI. The bottleneck is now different: It used to be “can we build it?” Now it’s “which one of these 20+ things should we actually finish?”

That is a leadership question.

What AI doesn’t take off your desk

Three things stay with people (at least in my experience):

• Direction: Someone has to say why we’re doing this (the bigger picture tied to where an organization is now, how the environment looks now and where it wants to go). An agent will optimize whatever you hand over (including the wrong thing).
• Judgment: AI answers sound plausible. That can lead you in the wrong direction if you don’t have people who know the domain well enough to find out when something is off.
• Accountability: When an automated process gets it wrong, the customer doesn’t call the agent. My management board doesn’t either. So, it is still a person who has to be accountable.

The risk I’d watch first: shadow AI

Someone in finance pastes data into a free online tool because it saves two hours.

Someone in the production department builds an automation nobody knows about. Nobody is acting in bad faith. On the contrary: (most) people do want to do a good job and improve. And as AI often is extensively used in private, ideas are transported to the workplace. And they do it on their own, because the official path often is slow or unclear.

The results are experiments. They might work, but either violate underlying governance or are not very stable and break easily. And if you try to ban them, people will find another way.

In my experience there are only two things that can help:

• set clear rules (and communicate them well)
• give people a path if ideas come up and most importantly somebody, they can ask/talk to who gives them an answer within days (not months)

What I do differently

I try to treat an agent like a new team member on their first day.

A new colleague doesn’t know which system is the real source of truth (if there is contradicting information), who to ask for what, or which rules are written down and what other rules do exist (but aren’t written down). Therefore, we have a mature onboarding process in my organization. I do something similar with my agents:

• I give them documentation (not just for their task, but also about the organization)
• I write a short explanation on how the department actually works
• I formulate what a good result actually looks like for me

It sounds simple, but it really improved how the agents operate.

What didn’t work: preparing workshops or a bigger meetings. I tried it. The agent gave me a clear agenda and good methods, but it was still off. It lacked originality and it didn’t know that … e.g. two others have history on this topic, or what the room is really looking for. This often is unwritten, implicit knowledge about the organization, culture and people themselves.

AI is great for groundwork: collecting material and drafting options. But the actual preparation stays with me. That’s not a tooling gap the next model release will close. It is the part of the job that is leadership.

So

Clear goals, clear decision rights and fast feedback loops. That’s nothing new. Good leaders have always needed it.

What’s new is the price of not having it. With agents on the team, unclear direction doesn’t just slow you down, it speeds you up in the wrong direction (what’s actually much worse).

One final question: when your AI agent gets it wrong, who owns the decision?

---

If you found this one interesting, also check out this article about what human skills still matter in the age of AI.