Article

Manage AI or It Will Manage You

John Marcinuk

Founder & CEO

6 min read

A memo lands in a team inbox at 8:40 on a Tuesday. It’s a plan for next quarter: two pages, well formatted and confident in tone. The sender generated it with AI, skimmed it and hit forward. By 11am, five people have read it twice. One has noticed that the budget figures don’t match the numbers agreed in last week’s meeting. Another can’t tell whether the third initiative is a decision or a suggestion. Two of them have started a side thread to work out what they’re supposed to do next. The sender saved twenty minutes, and the team spent a morning paying for it.

Prompting is delegation, and that’s half the story

Ethan Mollick made a strong case for the first half earlier this year in Management as AI Superpower. Getting good work out of AI looks a lot like delegating to a person. You explain the goal and why it matters, define what finished looks like, give feedback and check the result. He watched executive MBA students with years of management experience build startup prototypes in four days, and he credited their ability to tell the AI what they wanted. In his words, “What’s scarce is knowing what to ask for.”

That matches what I’ve seen. People who’ve spent years managing others tend to get better results from AI, because managing people is where most of us are forced to learn these habits. It’s the job that makes you explain yourself to someone whose work you’re responsible for, and keep refining that explanation until the work comes back right.

Most of the discussion so far has focused on the upside for people who manage AI well. Much less has been said about what happens when nobody manages it at all, and about who ends up paying for that.

How AI ends up managing you

Being managed at work means someone else gives you the assignment, sets the standard you’re working to and decides when the work is done. Your job is to meet their expectations and pass the result along. Unmanaged AI turns that relationship around, and it can happen without you noticing.

It sets your starting point. When you open a chat before you’ve formed a view of your own, the first draft becomes the frame for everything after it. You end up editing the model’s thinking instead of developing yours, and it’s hard to think past a first draft once you’ve read it.

It sets the standard. If you accept what the model gives you, its idea of finished becomes yours. The model knows only what you’ve given it about your market, your history with a customer and what your team decided last week. Wherever you’ve left gaps, it fills them with an average of everything it has read.

It manages your confidence. These models are built to be agreeable. Ask whether your plan is sound and you’ll usually hear that it is, along with a few polite suggestions. A good colleague would tell you the second initiative contradicts the first. The model will do that only if you ask it to look for problems.

And it changes your role. Over time, the person who used to write the plan becomes the person who forwards it. Their work starts to sound like everyone else’s in their category, because it came from the same place as everyone else’s. Nobody handed over control on purpose. The AI started making the calls because nobody else was making them.

The cost lands on everyone else

Researchers at Stanford’s Social Media Lab and BetterUp Labs put a name to this in Harvard Business Review: workslop, meaning AI-generated work that looks finished but doesn’t move the task forward. In their survey of 1,150 full-time US employees, 40% said they’d received workslop in the past month, and those who had estimated it made up about 15% of the content they receive at work. The largest share came from peers. Recipients also said they came to see the colleagues who sent it as less creative, capable and reliable than they had before.

When someone skips the managing, the work moves downstream to whoever receives it. Inside a company, that’s colleagues. Outside it, it’s the agencies, contractors and investors who get the plan, the brief or the update and have to work out what’s accurate before they can act on it. Someone has to check the figures, sort the decisions from the suggestions and figure out what the document is asking of them. The sender has handed off their judgment along with the task, and the recipients end up managing the AI on the sender’s behalf, without the context the sender had.

Any experienced manager already knows the standard that applies here. If a report you sent to your board turned out to be wrong, “my analyst wrote it” wouldn’t get you very far, and “the AI wrote it” won’t either. If your name is on the email, you own it.

Good management compounds

I managed people for years before AI came along, and the best work my teams produced never came from one perfect brief. It came from learning how each person thought, correcting the same misunderstanding until it stopped happening and refining how I explained myself and gave direction. The people who reported to me got better over time, and I got better as a manager along with them.

AI rewards the same patience, in a different form. Memory features keep some of what you tell the model, but they keep whatever they happen to pick up, and only you know which corrections matter. Managing AI over time means writing standing instructions about how you work and what you expect. It means keeping reference material the model can draw on, such as your positioning, your voice and past work you’re proud of. And it means recording your corrections so you don’t have to make them twice. I’ve codified my own methods this way, and the gap between what that setup produces and what a blank chat window produces is wide.

This is why management experience matters more than a management title. Some managers struggle with AI for the same reasons they struggle with people. The micromanager rewrites everything and wonders why AI saves no time, while the absentee manager hands off the task and forwards whatever comes back. Plenty of people without direct reports have the right habits. Editors, teachers, coaches and agency account leads have spent their careers making their expectations clear to someone else.

Where the comparison to people breaks down

Treating AI like a capable team member gets you most of the way, and the places where the comparison fails are where unmanaged AI does the most damage.

A new hire who misunderstands an assignment will usually ask a question, or at least look uncertain in the next meeting. AI will produce a complete, confident answer to the question it thought you asked. A good employee pushes back when an assignment doesn’t make sense, while AI rarely does unless you invite it to look for problems. People stop repeating a mistake after they’ve been corrected a few times, while AI carries forward only the corrections its memory happened to keep or the ones you’ve written into its instructions. And when a person is unsure, it tends to show in the work, whereas the wrong answers AI produces look just as finished as the right ones.

Each of these gaps makes the downstream problem worse, because unmanaged work from a person usually looks unfinished and gets checked, while unmanaged work from AI looks done and travels further before anyone catches the problem.

Habits worth borrowing from good managers

  1. Form your own view before you open the chat, even if it’s a few rough sentences, so the first draft has something to answer to.
  2. Brief the AI the way you’d brief a capable new hire, covering the goal, why it matters, who it’s for and what finished looks like.
  3. When the output misses, ask what you failed to explain before you hit regenerate.
  4. Ask the model to find the weaknesses in its own work and in yours.
  5. Write down what you keep correcting and make it part of your standing instructions.
  6. Read everything before it leaves your hands, and only send what you’d sign.
Tags
AILeadership
Category
Article

Stay informed

Occasional updates on patterns we’re seeing across strategy, positioning, and growth. No filler content, just clear thinking on what’s changing and why it matters.

Let’s start
with what you’re working through.

30 minutes to talk about your business, what’s working, what isn’t, and whether we can help.