Veritas Silentii Conversations Across Time
Conversations Across Time
Conversation No. 3 · September 2026
I Went Back to Copenhagen
Are we teaching AI to understand us…or changing ourselves so it can?
I told you last week that I needed to get back to Copenhagen.
I'm here now, at Bohr's Institute. It's still cold and windy, and the rain outside hasn't stopped.
I push open the door to Bohr's private office and hear the familiar sound of chalk against the blackboard. They are still arguing.
Apparently a week has passed for me. For Werner Heisenberg and Niels Bohr, perhaps only a few minutes.
Heisenberg pauses. He looks up the way you do when someone returns to finish a sentence.
"You're back."
"Before I left," I say, "you asked whether people would keep behaving the same way once they knew something was observing them. I think I finally have an answer worth bringing back."
The chalk stops. Bohr turns around.
"Already?"
I put two newspaper articles on the table. One describes why corporate leaders have grown uneasy about private workplace messages between employees. The other explains why companies want their AI agents to be able to read everything their employees say on internal chat systems.
"You need to see what we're doing."
The Problem With Private Conversations
"These companies think they've discovered a problem," I tell them. "They think too much valuable knowledge is disappearing into private conversations between employees."
Bohr looks puzzled. "Disappearing?"
"Not to the people having the conversation," I say, "but to the AI agents they've built to observe and record what they're saying."
Now Heisenberg is interested too.
I explain workplace messaging. Private messages. Meetings. Transcripts. And the increasingly capable AI agents that companies are beginning to put to work across their organizations.
"The agents become much more useful when they have access to more of what the organization knows."
"Of course," Heisenberg says. "That makes sense."
"But here's the problem," I say. "Much of what our organizations want to know isn't stored neatly in a database. It's moving constantly between people, in ordinary conversation. Questions. Ideas. Disagreements. Private messages. The conversation after the meeting. The quick exchange between two executives who trust one another enough to say what they really think."
"And if the agents can't see those conversations," I continue, "they can't learn from them."
Bohr looks down at the articles. "Do you think that will change how people communicate, once they know the agents are listening?"
"I think it may already be beginning to."
He looks at me. "And that is why you came back."
"Yes."
But Bohr Asks a Different Question
I expect him to remind me of the question he'd already asked. Instead, he asks something I hadn't considered.
"Why do these people speak privately in the first place?"
I hadn't expected that. It's a good question. I think about it.
"Lots of reasons. You're not certain enough to say something publicly. You want to test an idea with someone you trust before it has to defend itself in a room full of people. Sometimes you just need to say, 'This may be a stupid idea, but—'"
Bohr smiles. Then he asks the question that changes the Conversation.
"And do you believe the only purpose of a private conversation is to keep information hidden from others?"
"Sometimes, I suppose."
He shakes his head, and what he says next surprises me.
"Perhaps privacy isn't hiding the information. Perhaps it's helping create it."
I'm learning not to be surprised when I talk to him.
I look at Heisenberg. He smiles too.
Before an Idea Becomes an Idea
We talk about how an important idea actually develops.
It rarely arrives as a finished presentation. Often it begins with something much less organized — a stray thought, a feeling about something you just read or saw. That's all.
Then you find someone you trust. "Can I run something by you? I may be completely wrong." They push back. You explain. They disagree. Their idea merges with yours, and something changes.
Eventually something grows that wasn't there when the conversation began. Only in that last stage does it become a visible recommendation — a strategy, a decision, an innovation. Something the organization can see, record, measure, and track.
But the value began forming much earlier: first an unformed thought, then a tentative exchange with someone you trust, then collision with another person's thinking. Only later does it become coherent enough to leave that private exchange as something visible, defensible, and measurable — the knowledge everyone was looking for all along.
Heisenberg adds a question that's really a challenge.
"And now they want the agents observing those earlier stages too."
"Yes. That's where they think the missing value is."
He looks at Bohr. They both see the problem.
The Problem With Seeing Everything
I tell them that in my world, the obvious solution is already spreading.
Let the agents see more. Move more private conversations into shared channels. Record more meetings. Preserve more transcripts. Make more of the organization's thinking accessible to the machine.
From the company's perspective, the logic is compelling. If the agents can observe more of how value is created, perhaps they can understand value creation better.
But Bohr isn't convinced.
"Let me understand," he says. "You have discovered that some of your important ideas develop in places where people feel free to be uncertain. To be wrong. To say things they aren't yet prepared to say to everyone else."
I answer yes to each.
He pauses. "And now you intend to put an observer inside those places, so they are no longer private."
I tell him the objective is to preserve the knowledge before it disappears.
But I see where he's going. I look at Heisenberg, and he sees it too.
Bohr walks to the blackboard and writes:
Why do you assume there will be knowledge left to preserve, once you've removed the conditions that create it?
That one stays with me.
A Place Where Information Is Still Becoming
I had been thinking about private conversations as containers — some held useful knowledge, others didn't, and the challenge was simply giving the agents access to more of them.
But Bohr is asking me to think about it differently.
What if some private conversations aren't places where information is hidden? What if they're places where information is still becoming?
The unfinished thought. The safe disagreement. The question we're embarrassed to ask. The idea that needs to collide with another idea before either becomes useful.
And now we're inserting passive observers directly into that process, while it's still happening.
Heisenberg interrupts.
"You told me, during your last visit, that your AI systems are becoming very good at measuring where work stands at any given moment."
"Yes."
"And you were concerned that what they could measure now might not reveal the value that would emerge later."
"Exactly."
He gestures at the newspaper articles. "So now you are trying to solve that problem by observing more of what is happening now, before the value has emerged."
I hadn't thought about it quite that way. But he's right. Last time, I was worried about what the agents couldn't see. Now we're trying to solve that by letting them see everything.
Bohr looks at both of us.
"And you assume that once people know the agent is there, the same value will still emerge."
There it is again. The question that brought me back to Copenhagen.
This Isn't Really About Slack
I finally understand why those two articles bothered me. They aren't really about DMs, or Slack, or recorded meetings.
They're about something much larger.
We are beginning to redesign human organizations so they become increasingly legible to AI agents.
There may be enormous value in that. If AI systems can understand more organizational context, they should be able to make better decisions, find knowledge faster, connect ideas people miss, and preserve institutional memory that once disappeared the moment someone left.
But there's another side. The more we reorganize human activity so the agents can understand us, the more we may begin organizing ourselves around only what they can understand.
I ask Bohr the obvious question. "Surely the systems will get smarter. Eventually they'll understand more of how we create value."
He nods. "Perhaps."
"Then what's the problem?"
He turns back toward the blackboard.
"You keep asking what you can do to help these agents understand people better." He pauses. "You seem to assume that won't change the people."
"I'm wondering," he says, "what happens to people's behavior once they understand that they're the ones being understood."
For once, Heisenberg doesn't argue. Neither do I.
The Enduring Question
I walk toward the blackboard, chalk in hand, already forming the words in my mind. Something nameable. Something like The Agentic Observation Trap.
Bohr stops me before I write it.
"Not yet," he says. "You don't yet know what you're naming."
I set the chalk down. Maybe he's right. Some questions lose something the moment we name them too quickly.
I leave them at the blackboard and step back into the cold, wind-driven Copenhagen rain.
On the way home, I keep thinking about something. We want AI to understand more of how human value is created. That's reasonable. Perhaps inevitable.
But what happens when the process we use to understand it changes the very places where value is still becoming?
Which leaves me with a different question than the one that brought me back to Copenhagen:
Are we teaching AI to understand us…or changing ourselves so it can?
When I get back to 2026, there are messages waiting for me. Some public. Some private.
I begin typing a reply to one of them. Then I stop. I read what I've written.
For the first time, I find myself wondering who—or what—might eventually read it.
I delete three words. Replace them with two safer ones. And press Send.
What Inspired This Conversation
This Conversation was prompted by two reports published in The Wall Street Journal in August 2026: why CEOs have a problem with workplace DMs, and why your boss wants to make sure AI can read your Slack messages.