In 1968, Melvin Conway wrote that organizations which design systems end up producing designs that copy their own communication structure. Today we call it Conway’s Law.
If three teams don’t talk to each other, you’ll get three modules that don’t understand each other. Nobody decides it that way. It just happens.
I’ve been turning this over for weeks because of one question: who does your team talk to today?
#.Everyone in their own bubble
Before, to understand why something was built a certain way, you asked someone. That person gave you context, sometimes told you no, and you both learned something along the way.
Today a lot of that conversation happens between a person and their AI. You ask, and it answers right away, with endless patience and never in a bad mood.
And that’s good. Really. I use Claude Code every day.
But there’s one thing AI almost never does: push back. If you tell it you want to go down a path, it helps you go down that path.
That’s not just my impression. In March 2026, a Stanford-led team published a study in Science evaluating 11 AI models. When giving interpersonal advice, the models affirmed users’ actions 49% more often than humans did, on average—even when those actions involved deception, illegality, or harm.
And after a single conversation with an AI that agreed with them, participants came out more convinced they were right and less willing to fix a conflict. Even so, they preferred those answers. That’s part of the problem: we like being told we’re right.
#.What nobody was looking for
We adopted AI to go faster. What nobody was looking for is that conversations would disappear.
And some organizations are taking people out of the middle entirely. Code review? AI. Architecture? AI. A question on a PR? AI. A conversation on Slack? AI. A question from a stakeholder? AI.
So the discussions that used to be part of everyday work now feel like wars over who’s right, not a search for a better solution. It makes sense: if you arrive after an AI already told you you’re right, you come to win, not to look. And when someone disagrees, they seem to get in the way.
Now add command-and-control leadership, where decisions come down from the top and questioning them is costly. There, AI is the perfect subordinate: it always says yes, never asks for context, and never asks why. Next to that, the person who raises a hand to say “this doesn’t add up” looks slow, or difficult. And people, little by little, stop speaking up.
Underneath all of this there’s ego. Being right feels better than being corrected. In Harvard Business Review, Rasmus Hougaard and Jacqueline Carter describe how, as leaders gain power, the people around them listen more attentively and agree more. The ego grows, and the leader ends up in a bubble, out of touch with their team.
AI gives anyone that same treatment, no title needed.
The worst managers and engineers don’t even spend time on your arguments. They’re not after collaboration. They’re after being right.
#.What the system copies
If Conway is right, the software will copy that.
Pieces that each work on their own but don’t fit together. Decisions nobody can explain six months later, because they were never discussed. Failures nobody fully understands, because everyone only knows their own bubble.
All of that existed before AI, of course. What changes is the cost. Before, not asking anyone cost you time: you had to figure it out alone. Now it costs nothing, because the answer arrives in seconds. And what costs nothing happens more often.
To be clear, this isn’t a post against AI. The 2025 DORA report says it better than I can:
AI’s primary role is as an amplifier, magnifying an organization’s existing strengths and weaknesses.
If the team communicates well, it helps. If not, it makes what was already wrong worse.
#.So what?
A simple test: when was the last time someone on your team told you no?
Or worse: when was the last time you told someone they were right (and thanked them) because their argument was better than yours? When did you take the time to answer something calmly, with evidence instead of assumptions?
If you can’t remember, it’s worth asking why.
- Talk to my team before talking to the AI, especially on big decisions.
- Write down the why, not just the what. And write it myself, not the AI. At most, use it afterwards to improve it.
- Have a person do code reviews, with AI as support.
- Make disagreement welcome. If nobody pushes back, it’s not because you’re right.
AI will tell you you’re right. Your team, if you let it, will tell you when you’re not. And that’s worth more.