The company you run has already been reorganised. The org chart has not changed. Titles are unchanged, reporting lines look the same, and the seating plan in the Monday leadership meeting has not moved. Something else has shifted underneath. A different set of people are doing different kinds of work than they were nine months ago, and a different set of questions are moving through the system on different paths. If you run the place, some of this will feel familiar. Most of it you will not yet have seen.
Every company has two org charts. One is the formal one, drawn in a slide and updated once a year when someone gets promoted or a team is moved under a new VP. The other is the working chart, which is the map of who actually asks whom for what, whose review matters, and whose sign-off unblocks the work. The formal chart lags the working chart by months in a healthy company and by years in a tired one. The working chart is where the company really is.
AI redraws the working chart first. An FP&A analyst who used to stop at her own model now runs a variance check her manager would have run, and sends it up one level. An engineer who used to schedule a call with the principal to unblock a difficult refactor finishes it alone in an afternoon and never raises the ticket. An account manager who would have looped in a solutions consultant to answer a configuration question answers it herself and closes the thread. Each of these looks like a small productivity gain in isolation. Taken together, they are a reorganisation in substance: a set of handoffs, reviews, and teaching loops that used to exist and no longer do. The call was never made. The review was never requested. The work stayed at the edge of the company instead of becoming a conversation inside it.
On the engineering side, the shift is easiest to see in PR review traffic. Senior engineers were load-bearing because they saw everything. That was the implicit contract: you commit code, you get read by someone who has been there long enough to spot the shape of a future outage. Once AI is doing a first pass on pull requests, catching the obvious mistakes and answering the junior engineer’s clarifying question before she asks it, the traffic to senior engineers thins. The senior engineers still have the same job description. They are no longer the default node in the diagram. This does not usually register as a problem until something breaks that a senior engineer would have caught on the way past.
In finance it shows up differently. A director of FP&A runs a small team of analysts. The old rhythm was that analysts built the first pass and the director rebuilt it, teaching on the way. With AI handling the mechanical reconciliation, the analysts are arriving with a cleaner first pass, sometimes a better one. The director’s role quietly mutates. Some of what looked like management was actually quality control that has now moved upstream. The director is still the director on the chart. On the working chart, she has become the teacher at the point where she thought she was being a manager. The part of her job that was about catching errors early is gone. The part that calls for judgment on what the numbers mean is untouched. Whether she notices the difference depends on how much of her week she was spending on the catching.
In customer-facing work, the redraw runs the other direction. The VP of sales has historically been pulled into escalations because mid-market reps could not resolve complex deals alone. Increasingly, the reps resolve them. They draft their own response to the awkward customer question, check it against how the company has handled similar ones before, and send it back the same afternoon. The escalation queue in the VP’s inbox gets shorter. What looks like a better-trained team is often a team that has stopped asking. The VP’s judgment is still available. It is no longer being sought.
The thing that used to travel through these loops, besides answers, was context. The senior engineer did not just catch the bug. She knew, because of the three hundred bugs she had caught before, which parts of the codebase were getting fragile. The FP&A director did not just correct the model. She knew which analyst’s estimates she trusted and which she double-checked. The VP of sales did not just close the escalation. She saw which accounts were getting nervous three weeks before they churned. When the loops thin, the answers still arrive. The context does not. The confidence read does not. The early warning does not. The training signal that made someone senior in the first place is no longer being generated at the same rate.
The public framing of this is usually that AI replaces middle management, which is too crude to describe what is actually happening. It is a story about job titles when the real story is about information flow. Middle managers are not being replaced. The routing function they were providing, often without knowing they were providing it, is being done by something else. The middle manager still turns up to the Monday meeting. Her job description has not changed. The questions that used to arrive at her desk now arrive somewhere else, or nowhere.
The executive consequence is more uncomfortable than it looks, because the information you personally receive has not changed much. The weekly report still lands. The board deck still gets built. The customer calls still come in. What has changed is everything upstream of those artifacts. The deck is cleaner because it had fewer hands on it. The customer calls are shorter because more questions got resolved before the call. The weekly report is tidier because fewer people were involved in producing it. You are still getting the output of the company. You are no longer getting the friction. The friction was where you learned things.
If you are an operator trying to read this, the question is not which roles to automate. It is which loops have thinned. Where did three people used to touch a thing and now one does. Whose calendar used to fill up with clarifying questions and is now quiet. Which function is still producing output at the same quality but with half the hands, and whether anyone is teaching the remaining hands what the missing people used to teach. The absences are the data now. The calls that did not get made, the reviews that did not get requested, the escalations that did not reach you. That is where the reorganisation is.
None of this is an argument against using AI. It is an argument for knowing what it has already done. The formal org chart will catch up eventually, because someone will notice that a team of twelve is no longer working through the same handoffs it used to and will write a title for the new shape. That is always the slowest step. The reorganisation is already under way. The only question is whether leadership notices before it starts managing from a chart that no longer describes the company it is trying to run.