AI does not make expertise worthless, but it weakens the old status bargain in which leaders mattered because only they could produce or bless the answer. This edition names the ordinary human grief behind that shift and argues that leaders can release answer-based authority without shrinking, if they move their credibility toward judgment, risk ownership, context, and helping others orient.
There is a particular kind of silence that enters a room when a younger team brings in a strategy memo that is… good.
Not perfect. _Not ready to ship._ Not wise by itself.
But good enough that the old choreography breaks.
The experienced leader can no longer lean back, find the obvious gap, and become necessary by supplying the missing answer. The memo already has a workable structure. It has found the comparison set. It has produced the first-pass logic. It has even committed the small courtesy of sounding confident in a way that makes everyone briefly forget how confidence is now available in bulk.
So the leader reaches for the red pen.
Maybe the team needs to share the prompt history. Maybe the argument needs another layer of analysis. Maybe the language should be tightened into the dialect of the senior meeting, where simple sentences go to acquire a badge and a parking pass. Maybe there should be a mandatory "AI sanity check," which lasts 47 minutes and mostly proves that someone still owns the red pen.
Sometimes that review protects quality.
Sometimes it protects the expert.
This is the part of AI adoption most operating advice skips. It talks about workflows, tools, governance, and productivity. All necessary. All easier to discuss than the quieter loss underneath.
When answers stop conferring status, the expert has something real to grieve.
The loss is not expertise
The previous edition, _The Ghost of Expertise_, argued that answers have become a lagging indicator. A leader can still know a great deal and still arrive too late, solve yesterday’s problem, or defend a frame reality has already left behind.
This edition sits one layer deeper.
The painful part is not that expertise suddenly has no value. That is too crude, and more importantly, it is false.
The loss is more specific: the loss is being needed because only you could produce the answer.
That distinction matters because many experienced leaders are not defending a spreadsheet cell, a paragraph, or a model choice. They are defending a familiar source of mattering.
For years, the expert’s social contract was clear. Know more. See faster. Correct others. Catch the hidden flaw. Translate ambiguity into the answer that lets the room move. The pedestal was not always vanity. Often it was earned through scar tissue, late nights, failed launches, uncomfortable customer calls, and the kind of pattern recognition no course can download into a person.
Then AI makes plausible first drafts abundant. It lets people explore domains they used to wait in line to access. In a field experiment with Procter & Gamble professionals, researchers found that generative AI helped individuals match some of the performance benefits of teams without AI and produce more balanced technical and commercial ideas in that setting.
That does not mean teams are obsolete.
It does mean the boundary around specialist contribution gets less tidy.
Stanford HAI’s 2026 AI Index describes AI adoption as broad enough to be part of ordinary organizational life, not a side project for the innovation corner. McKinsey’s 2025 AI survey work points in the same direction from another angle: the organizations seeing more value are not merely adding tools to old habits, but redesigning workflows around them.
Use those findings carefully. They do not prove that every expert is replaceable. They do support the shift leaders can feel in their bones:
Stored expertise is becoming less sufficient as a basis for authority.
That is not a small change. It touches identity.
Defensive expertise has a costume
Threatened expertise rarely announces itself as threatened expertise.
It arrives dressed as quality control.
It asks for another review cycle. It adds terminology nobody outside the room uses. It widens the stakeholder list until accountability dissolves into calendar fog. It demands one more analysis, not because the decision is under-evidenced, but because a decision made without the expert’s fingerprints feels vaguely unsafe.
There is a legitimate version of all of this.
Some work should be reviewed carefully. AI-generated output can be wrong, biased, incomplete, or too plausible for its own good. NIST’s generative AI risk guidance is a useful reminder that responsible AI management should be tied to context, risk, legal and regulatory obligations, and organizational priorities. In McKinsey’s survey data, organizations vary widely in how much generated content gets reviewed before use. That variation is not automatically a problem. Different risks deserve different controls.
The problem is review without a named risk.
If the review exists to protect customers, safety, legal exposure, financial integrity, reputation, or strategic alignment, name the risk and own the decision. If the review exists because the senior person feels oddly absent from the work, call that something else.
Some rituals protect quality.
Some rituals protect the expert from embarrassment.
Organizational researchers have long described defensive routines as patterns that avoid embarrassment or threat while preventing people from identifying hidden issues and creating new knowledge. That is the useful lens here. Not diagnosis. Not a clinical label. A behavioral pattern.
In AI-enabled work, defensive expertise often looks like this:
- turning judgment into blanket approval;
- replacing clear language with insider vocabulary;
- treating prompt inspection as authority performance;
- requiring extra analysis after the decision-relevant uncertainty is already clear;
- withholding context so others must keep returning to the expert.
The tragedy is that these behaviors usually come from people who still have real value to offer. Their judgment may be exactly what the team needs. Their domain memory may prevent a costly mistake. Their scar tissue may catch the subtle risk the model smoothed over.
But the value gets trapped behind a demand to be needed in _the old way_.
Grief is information
Call it grief in the ordinary human sense: a response to losing something that gave life shape, status, certainty, or a sense of self.
No diagnosis is required. No stages. No therapeutic theater in the conference room, please. The Q3 operating review has suffered enough.
But the word is useful because it refuses to trivialize the loss.
If you built a career on being the person with the answer, it can hurt when the answer becomes cheap. If your authority came from catching errors others could not see, it can feel destabilizing when a capable junior person arrives with an AI-assisted draft that is messy, useful, and no longer obviously junior. If your reputation rested on final review, it can feel like disappearance when the better leadership move is to define the risk, assign responsibility, and get out of the way.
That discomfort is not proof you are obsolete.
It is data about how you measured your worth.
The question is whether you can read it without letting it drive.
Because grief mishandled becomes nostalgia with decision rights.
It turns into the old expert insisting that every AI-assisted document pass through the same bottleneck, even when the work now needs faster feedback, clearer standards, and more explicit ownership. It turns into performative humility too, which is only the mirror image of the pedestal: "Oh, I know nothing now," said in the tone of someone waiting to be contradicted.
Neither posture helps.
The task is not to cling to the pedestal or theatrically fall off it. The task is to step down without shrinking.
Authority after the pedestal
Once the expert no longer has to prove value by owning the answer, a more demanding kind of authority becomes available.
It is quieter.
It is less theatrical.
It is harder to fake.
It looks like asking, before the review begins: What risk are we controlling?
It looks like saying: This draft is directionally strong, but the customer claim is too broad, the market assumption needs a check, and nobody owns the decision if the recommendation is wrong.
It looks like distinguishing what must be true from what merely sounds polished.
It looks like helping the team see the field, not reminding them who used to own the map.
This is where domain expertise still matters. In some ways, it matters more. When AI can generate plausible material quickly, the leader’s job shifts toward judging what matters, what is missing, what could hurt someone, what the organization is now responsible for, and which decision deserves commitment.
That is not answer ownership.
That is orientation.
And orientation requires expertise that has become mature enough to stop demanding constant tribute.
A better review ritual
If the old review ritual is starting to smell like status preservation, try replacing it with a simple test.
Before touching the work, ask four questions:
- What consequence could this create?
- What must a human be accountable for here?
- What uncertainty would change the decision?
- What part of my expertise helps the team orient, rather than merely proves I was present?
Those questions do not remove review. They make it sharper.
They also expose which part of the expert’s identity is being asked to change. Maybe the leader is no longer the only person who can produce the first draft. Fine. Then become the person who can tell which draft deserves to become reality.
Maybe the team no longer needs permission for every paragraph. Good. Then define the boundaries within which they can move without waiting.
Maybe the old red pen no longer proves value. Excellent. Use it only where the mark changes consequence, not where it preserves rank.
There is grief in that.
There is also relief.
A pedestal is flattering until you realize it is a very narrow place to stand. It gives visibility, but not much range of motion. It keeps you above the room, which means it also keeps you from seeing parts of the room.
The leader who steps down does not become irrelevant. The leader becomes more available for the work that was always larger than the answer.
To frame.
To judge.
To take responsibility.
To help others move without pretending certainty has returned.
The question, once the pedestal is gone, is not whether you still matter.
The question is what wider field you can finally see.
