For the last 20 years, the smart way to run a company was bottom-up. You let smart people do what they do best. You pushed authority to the edges. You built culture, not hierarchy. And it worked.
Then AI happened.
Or more precisely, AI started changing things fast enough that a lot of the old assumptions started breaking. Bottom-up still works in many places. But if you’re building AI or adapting your company to it, you’ve probably started noticing something: the companies that are winning aren’t the ones with the flattest orgs. They’re the ones with the clearest direction.
Not necessarily the biggest or the richest. Just the ones where someone is saying, "This is where we’re going," and everyone else moves.
That used to be called top-down management. It’s coming back. But not the way people remember it.
The real reason
AI didn’t just introduce a new technology. It introduced a new tempo.
The field moves fast. Research changes monthly. APIs update weekly. Interfaces shift, models get cheaper, the ecosystem refactors itself in real time. You don’t just need to move fast — you need to choose quickly, and often.
In that kind of environment, decision-making becomes more valuable than consensus. Not because collaboration is bad, but because hesitation costs more.
Direction is not control
When people hear "top-down," they think bureaucracy or command-and-control. That’s not what’s returning.
What’s returning is direction.
The best leaders right now are not managing every detail. They’re doing something harder. They’re setting a direction and saying no to everything that doesn’t fit. They’re making bets. They’re repeating the mission over and over until people get bored of hearing it — and then repeating it again.
They don’t tell everyone what to do. They tell them why they’re doing it. That’s enough.
Coherence vs. chaos
In 1999, Garry Kasparov played a game of chess against the internet. Thousands of people voted on every move. They had access to engines, commentary, and experts. He still won.
Why? Because the crowd had no plan. They argued at every step. Kasparov had a strategy.
That’s what happens inside companies too. A hundred smart people pulling in slightly different directions doesn’t give you innovation. It gives you noise. Great products feel like they came from one mind. Because they did.
That mind might listen to others. It might even be wrong. But it moves.
Not forever
Top-down isn’t a permanent solution. It’s a tool.
You use it when the terrain is changing fast and you need everyone aligned. You don’t stay in that mode forever. Once the shape of the problem becomes clearer, you can open things up again.
The best founders I know switch between modes. When the team knows where to go, they delegate. When the team is confused, they zoom in. They don’t get stuck in one style. They manage based on the situation, not the fashion.
What’s actually happening
People assume flat orgs are fast. Sometimes they are. But many of the companies that look flat on the outside are sharp on the inside.
Nvidia looks like a normal public company, but it isn’t. Jensen Huang decides what matters. And everyone else orients around that.
OpenAI isn’t run by consensus either. Sam Altman sets direction. He’s not always right. But his team doesn’t spend six months in debate. They ship, then adjust.
Even Google merged its AI teams under DeepMind to cut the politics and accelerate launches.
The pattern is the same. Not more meetings. Fewer decisions, made faster, with more focus.
The hard part
The hard part of top-down isn’t making decisions. It’s being accountable for them.
Consensus feels safe. No one owns the outcome. Everyone shares the blame. That’s why people like it. But it also means no one drives. The car just coasts.
Top-down means picking a direction when others aren't sure. It means making calls that might be wrong. But if you don’t, the opportunity passes you by while you’re still aligning calendars.
What I’m seeing
At Rohlik, we’re living this. We’re not a legacy company in the classic sense. But we’ve been around long enough to build legacy behaviors. And when the vision isn’t sharp, those behaviors take over. Committees form. Energy dilutes. Momentum stalls.
Changing that has been hard. Harder than I thought. And if it’s this hard here, I can only imagine what it’s like in older, slower organizations.
But what I’m seeing is this: if you want to move fast, someone has to say where you’re going. Someone has to own the hill you’re trying to take. If no one does, the hill stays untaken.
So what?
If you're building in AI, or trying to adapt to it, you probably need to be more top-down than you’re comfortable with.
Pick three things that matter. Say them out loud. Kill the rest. Assign owners. Get out of their way. But stay close enough to steer.
And when it feels awkward to take the lead, remember: the only thing worse than being wrong is being directionless.
The title is great for getting people in, but is not in telling what you're doing afterwards :)
This situation is not new though; every company gets more coordination headwind as they grow: in hierarchical ones it's bottlenecked managers (who need to decide everything), in decentralized orgs it's meetings and committees. (this is a good presentation that explains and concludes pretty much what you say: give direction https://komoroske.com/slime-mold/)
AI speeds up execution, but does not help with coordination (ie. OpenAI is still chaotic, just look at their product direction and model naming mess). AI helps you to have the same output (in the digital world / shipping 'features') with fewer people: you can return to a smaller org.
As the company grows, sooner or later you won't have time to explain your vision to everyone, you'll need people to do that.