Grisha Pavlotsky, Chief Transformation Officer at Miro.
Somewhere in your company, someone is having the best idea of the quarter. They are having it in a chat window, alone with an AI, and no one else will ever see it.
Silos have been and always will be the enemy of speed. AI was supposed to help (and it still can), but in the meantime, it quietly built a new silo for every individual in your company.
Understanding That AI Fluency Is Not Transformation
Let’s think about what actually happens when someone does good work with AI.
The real work likely happened during a private brainstorm session where ideas got tested, trade-offs got argued, wrong turns got taken and dropped, and the reasoning took shape. What ships is a conclusion which is really just the tip of an iceberg with tons of context that no one will get to see or experience. And all of the stuff that came before the artifact is where a teammate could have built on or pushed back on. It’s an opportunity for healthy friction that has been missed.
Individual adopters of AI tools may report something like five times the output, while according to IBM, only 29% of companies can confidently measure any ROI.
I first heard the term “silos of one” from Matt LeMay, who studies how product teams communicate. His warning is that if we skip the slow work of actually talking to each other, “We’re going to 10x our misalignments, not our outcomes. We’re going to wind up with silos of one, shipping disconnected things that don’t align into anything meaningful for our business or our customers.”
This idea perfectly illustrates the difference between adopting AI and transforming your company.
Why Your People Are Faster And Your Company Isn’t
The old silos were at least easy to see. A silo was a wall (sometimes literally) between two departments. You had a handful of them, and they were big enough that you knew exactly where they stood. But a problem you can see is at least a problem you can navigate.
AI took your handful of large, visible silos and shattered them into thousands of tiny ones, a separate silo for every person now thinking privately with a machine. Then it added a kind you’ve never had before: agent-to-agent, where software hands work to other software beneath the surface.
So now you have three types of silos:
1. Human-to-human
2. Human-to-agent
3. Agent-to-agent
Add remote, asynchronous work on top of all this and it’s no wonder so many executives are wondering why all their people are faster but their company isn’t.
The Exact Moment Where Speed Is Lost
A silo is a problem specifically because people inside it drift apart from people outside it. People might hear the same set of instructions, quietly read them three different ways and not realize it until later when they finally compare notes.
AI doesn’t make people drift further necessarily, but faster. Ask AI to dig into a plan and it says the plan is sharp. Ask it to benchmark your strategy and it says the strategy is world-class. Models are very good at reflecting your own language back at you, and hearing your own words repeated approvingly feels exactly like agreement, even when feedback from others would improve both alignment and outcomes.
Worse, people fall in love with their outputs, and that is the stickiest misalignment problem. When the team comes together, you are not solving for whose idea is better but who will give up their idea first.
AI is a lot better at repeating what we say than asking what we meant. So your people sprint off in their separate silos, all of them productive and all told by their own machine that they’re on track, and then they meet. Three smart people then discover they just spent a week building three different things. Nobody was lazy and nobody was wrong, but the work was disconnected.
AI Adoption: Only The First Phase
Every one of those silos got built by someone doing exactly what you asked them to do. You said adopt AI and they did. They’re faster and they love it, and you’re wondering why the company is busier than ever but not faster or more profitable than it was before.
The gains from individual AI productivity level off fast, and many companies have already hit that ceiling. But the real ROI from AI implementation has a global maximum that far exceeds where you likely are today.
Getting there starts with a shift in what you ask of your leaders. Instead of asking how many people are using AI, start asking how long work waits between each person. In most companies, the idle time between teams outweighs the time anyone spends doing the work, and no individual tool can remove that. Adding AI on top of old habits only speeds up the past.
Put your department heads in charge of rebuilding AI-native workflows that make it easier for teams to use AI together. That’s how you’ll get past the local maximum.
That best idea of the quarter is still sitting in a chat window somewhere in your company right now. The only question now is whether you’ll build something that lets it out.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?