AI Changed the Job. Why Haven’t You Changed Who Gets Promoted?

By Liat Ben-Zur

As AI makes strong output easier to produce, companies need to rethink whether traditional performance signals still identify who is ready to lead.

I’ve sat in enough promotion discussions at Qualcomm, Philips, and Microsoft to know how the conversation usually goes. Every promotion decision rests on an assumption about what predicts future leadership. For decades, companies have rewarded people who produce strong work and deliver consistent impact. Those habits used to be rare, and the work itself was a reliable proxy for the judgment behind it. That link is now weaker thanks to AI. The finished product reveals far less about the depth of thinking that produced it, and output alone has become a shaky basis for deciding who is ready to lead.

The answer is not to stop rewarding expertise or execution. It is to recognize that AI has changed what those signals tell us—and to put greater weight on judgment, decision-making, adaptability, and the ability to see what the technology misses.

Why did output become such a powerful signal of leadership?

Organizations were built around the idea that strong output reflects strong judgment. No CEO can sit in on every customer call, track every sale, or follow every decision being made across a company. Hierarchy became the way organizations filtered all that information and moved it upward. Millions of transactions became a few lines on a P&L. The org chart became an information pipeline. Data flowed up through a dozen human filters; strategy crawled back down the same pipe.

Our promotion criteria grew up inside that system. The people who accumulated the most knowledge and executed most reliably became the obvious candidates for greater responsibility. Seniority used to be the only way to get the full picture. That’s no longer true. An employee can examine patterns across the business that once required access to several layers of management.

AI unsettles a basic assumption behind the career ladder. It breaks the link between sophisticated output and the expertise previously required to produce it. An entry-level hire can generate senior-looking analysis in an afternoon, and the deliverable alone tells you far less about the depth of thinking behind it.

That doesn’t mean the analysis is necessarily good. It means the finished product is becoming a weaker proxy for the person who produced it.

What happens when AI gets the answer wrong?

The more revealing test is what happens when the analysis is wrong. Researchers from Harvard, Wharton, MIT Sloan, and BCG gave 758 BCG consultants a series of assignments, some with GPT-4 and some without it. On 18 common consulting tasks, AI helped people work faster and produce better answers. Then came a second case, built to sit outside the model’s capabilities: identifying which brand had the greatest growth potential using financial data and interviews. The AI often misread the evidence and pointed them toward the wrong conclusion. Consultants who used it were 19 percentage points less likely to get the answer right than those who worked without it.

Being wrong has never looked so persuasive.

A leader has to catch what the model missed before someone turns its answer into a decision. Can this person spot a bad assumption? See the gap in the data? Push back when everyone else is relieved to have an answer? These abilities have always mattered, but promotion systems often treated them as secondary because they were harder to measure.

The lesson from the research is not that AI makes people worse at their jobs. In many of the tested tasks, it did the opposite. The lesson is that AI’s usefulness depends on whether a task falls within its capabilities. That makes human judgment about the quality, limits, and context of an answer more important—not less.

What should promotion systems measure now?

Most of a manager’s day went to coordination. Tracking the work, reviewing it, pushing information up a layer and waiting for it to come back down. The leadership part was harder to schedule. Deciding where to go. Making a trade-off that costs somebody something. Growing people. Owning the outcome when things go wrong.

AI is absorbing the coordination, and what remains is suddenly visible. We can see who can actually lead, and who got promoted for being good at the machinery. Yet many promotion rubrics still reward the machinery.

Which means the promotion rubric has to evolve.

Expertise and reliable execution still matter. They are simply not enough to predict who will lead well. Pay attention to what people do when the stakes are high.

Look for people who:

• Challenge a flawed premise before it becomes an expensive mistake.
• See consequences others miss.
• Change course when the evidence changes.
• Resist doubling down simply to save face.
• Create an environment where junior colleagues feel safe disagreeing with them.
• Take ownership of outcomes rather than simply producing deliverables.

These moments are hard to capture on an HR scorecard. But as answers become cheap and abundant, they tell us far more about who is ready to lead.

How can companies put those qualities into practice?

The shift doesn’t require abandoning existing performance systems. It requires changing what leaders look for within them.

Instead of evaluating only the quality of the final deliverable, ask how the employee arrived at it. What assumptions did they make? What did they question? Where did they decide not to trust the available answer? How did they respond when new information challenged their original conclusion?

Promotion discussions can also make room for evidence that is less visible in traditional performance reviews:

• How does the employee handle ambiguity when there is no obvious answer?
• Do they identify risks before they become problems?
• Can they explain why a decision should be made, not simply what should be done?
• Do they help other people develop better judgment?
• When AI produces an answer, do they accept it, challenge it, or know when to investigate further?

The point is not to create another complicated scorecard. It is to recognize that the most valuable evidence of leadership may now appear in the moments when an employee decides what not to accept, what to question, and when to change direction.

That is especially important as AI makes polished work increasingly easy to produce. If everyone can produce the same kind of output, the differentiator moves upstream to the thinking that determines what should be produced in the first place.

What happens if promotion systems don’t change?

In a July Fast Company essay, I argued that the biggest mistake in AI transformation is automating work that should be eliminated. Companies repeat a similar mistake when they promote people to preserve an operating model they are trying to dismantle. When the way we work changes, what it takes to get ahead has to change too.

The risk is not simply that companies will promote the wrong people. It is that they will continue rewarding behaviors that belong to an operating model AI is already changing.

If AI handles more of the coordination, synthesis, drafting, and analysis that once demonstrated expertise, then organizations need to become better at identifying the capabilities that remain distinctly important: judgment, discernment, accountability, adaptability, and the ability to lead other people through uncertainty.

The career ladder doesn’t have to disappear. But the signals used to move people up it can no longer remain frozen in the world that existed before AI.

As answers become easier to generate, the harder and more valuable question is increasingly: Who knows what to do with them?

About the Author

Liat Ben-Zur is a technology executive, independent board director, and AI advisor with more than 30 years of experience at Microsoft, Royal Philips, and Qualcomm. She is the author of The Bias Advantage and founder of LBZ Advisory, and serves on the boards of Compass Group and Splashtop.