A senior operator’s read on their own company is mostly built from a specific habit: reading the raw stuff themselves. The customer email that came in on Saturday, or the Slack thread from the engineering team arguing about a rollback. The shape of the company someone runs in their head gets updated in those moments rather than in the review meetings.
Reading the first pass of something is different from reading a summary, even when the summary is good. The first pass has weight the summary can’t carry: the raw phrasing of the customer’s complaint, and the hedge in the engineer’s sentence that says more about how the team is doing than the summary ever will. Those details don’t make it into the summary because the summary is doing the right job, which is to compress. The compression is what lets the executive move fast, and it’s also what makes the executive’s mental model of the company a compressed model.
This wasn’t a problem before, because the executive was still doing the first pass themselves on most streams. They’d read the raw customer inbox on a Sunday, or skim the engineering channel after the standup. Those passes kept the internal read updated, without a process or a report mediating between the executive and the material.
AI changes this. The model writes the first pass now. The executive reads the summary, which is clear and accurate, and moves through their Sunday more efficiently than before. The raw material doesn’t touch the executive’s attention. A week of raw signal gets compressed to a page of insights, and a quarter of raw signal gets compressed to a dashboard.
The internal read stops updating. The executive still has a read, built from the months before AI started doing the first pass, and that read keeps working for a while. It stays accurate for things that haven’t moved, but goes stale on things that have. The stale bits don’t announce themselves, so the executive starts saying things in meetings that are accurate as of about six months ago, and the people in the room who are closer to the material notice.
This is different from not knowing something. A CEO who hasn’t heard about a customer escalation can learn about it when someone brings it up. A CEO whose internal model of the customer base is six months stale will hear about the escalation, absorb it into a model that doesn’t have the updated context, and draw the wrong conclusion without knowing they’ve drawn it. What changed is the frame the information gets processed against. The information is still arriving, but the frame it’s landing in is the one from six months ago.
The atrophy moves in months rather than years. Three months in, the executive’s read is slightly off in the places the company is changing fastest. By six to twelve months, the drift is large enough that the executive is getting corrected by people one level down, and is the last person in the room to understand what the company has become.
Keep doing part of the first pass yourself. Not all of it. AI summarization is a real productivity gain, and worth keeping. The part worth keeping for yourself is the raw signal from the parts of the company that are moving, which is usually customer signal and the team signal that hasn’t yet been aggregated into a metric. Something like twenty customer emails a week, in whatever format they came in, is usually enough. Small enough to be tractable, and large enough to keep the read updated.
The authority a senior operator has, inside their own company, comes from being able to feel the company under their decisions. That feel is what lets them override a report when the report is wrong, or trust a junior person’s read when the data doesn’t support it yet. Formal authority continues to function once the internal read goes stale, but it works for fewer people and carries less weight in the room.
There’s a cost to the speed AI provides, and this is one of the invisible ones. A leader who can’t feel their own company loses the specific authority that came from being able to. The leaders who’ll come through this best are the ones who’ve kept a hand on the raw signal while they let the model do the rest.
Pillar II
**Operating Systems for Executives**
> *Five memos on the parts of a senior role that have to be built as > systems rather than carried as personality.*