She had the dashboard open on her laptop when I walked into the session. Engagement scores. Productivity heat maps. A neat little recommendation glowing at the top: put Sarah on a performance plan. “I’ve known Sarah for six years,” she said. “Something is wrong with this picture and I can’t name what.”
Most leaders who have inherited an AI-augmented performance system in the last eighteen months have had some version of this moment. The data is technically clean. The recommendation is confident. And something inside them knows the answer is incomplete, but the language to push back doesn’t arrive fast enough.
That tension is not a personal failing. And it is not, contrary to where most of the conversation has landed, a problem with the AI. The AI is doing what it was built to do.
The problem is the model the AI was bolted onto. And until we name that, no tool we add will rescue us.
The performance review was already broken. AI just exposed it.
Most performance management systems still in use in 2026 were architected during the Second Industrial Revolution. Frederick Taylor’s Scientific Management was published in 1911. The annual review, the rating scale, the forced ranking curve, the productivity score, these are not modern tools. They are factory-floor instruments designed to optimize the output of interchangeable workers on an assembly line. They assume that performance is a measurement problem. That the manager’s job is to rate the past. That the worker is best understood as a unit of production.
That model has been quietly failing for decades. Adobe killed annual reviews in 2012 and reported a 30% drop in voluntary turnover. Microsoft walked away from stack ranking and unlocked the cultural shift that catalyzed its AI era. GE retired the vitality curve it championed for thirty years. The companies that moved earliest discovered the same thing: frequent developmental conversations beat annual judgment by a wide margin.
And yet most organizations still run on Taylor’s blueprint. The AI tools landing on top of those systems right now are not fixing them. They are accelerating them. A 1911 conversation, run by a manager who has never been developed to do anything else, just happens faster when an algorithm pre-loads the recommendation.
Faster broken is not better. Faster broken is liability.
The frontline managers can feel it before anyone else
DDI’s Global Leadership Forecast 2025 surfaced a finding that should have been treated as a five-alarm signal: frontline managers are 3X more concerned about AI than the executives rolling it out. The default reading was that the front line is behind on the technology. After 20 years of coaching senior leaders, I think the conventional read has it backwards.
Frontline managers are not behind. They are ahead. They are the closest people in the organization to the moment AI either supports human judgment or replaces it. They feel the cost first, because they are the ones who actually have to walk into the conversation that the algorithm has prepared for them. And they understand, in a way the C-suite has not yet metabolized, that the tool is being asked to optimize something that was fundamentally not an optimization problem to begin with.
The conversation a leader has with a person about that person’s growth is not a productivity transaction. It is the most important developmental moment a manager has all quarter. And we are layering 2026 algorithms on top of 1911 conversations and wondering why our best leaders are quietly burning out.
The shift: from rating workers to growing leaders
The performance management research and the motivation research have been converging on the same conclusion for the better part of a decade. People perform best when work feels meaningful, supported, and self-directed. Intrinsic motivators, autonomy, mastery, purpose, psychological safety, are more durable than extrinsic incentives for any kind of complex, knowledge-intensive work. And the manager is the central lever, because the quality of the relationship and the texture of the conversation shape whether those conditions exist or not.
That is a fundamentally different model than the one the appraisal was built for. It is not a system for rating the past. It is a system for engineering the conditions under which the next best performance can emerge.
The phrase that has been showing up across the leading practice work of the last two years says it cleanly: enable the next performance, do not rate the last one.
This is the shift most organizations are not yet making. They are buying the AI tool. They are not redesigning the underlying conversation. And without that redesign, the tool just calcifies the old model in faster, prettier form.
Three things AI cannot do, and the human capabilities your leaders need instead
Sarah’s manager was being asked to do something her training had not prepared her for. She had been promoted into leadership on her ability to deliver inside the old model: hit your numbers, run the cycle, complete the form. Nothing in her development had built the muscles the new model actually requires.
Three of those muscles, in particular, are the work of growing leaders for the era we are now in:
- Reading what is missing from the data
Real performance lives inside a story. Sarah’s output had dropped 18% in the last quarter, the dashboard was correct about the number. The system could not see was that Sarah’s mother had been hospitalized in March, and that two of her best people had quietly reassigned out from under her in February, and that she had been carrying a cross-functional project for four months that nobody had ever formally given her credit for. The data was the last chapter. The story was the whole book. A leader who has been developed for the new model knows how to read the story underneath the metric.
-
Holding presence in a hard moment
AI can transcribe a conversation. It cannot feel the room. When a person’s body language shifts, when the silence stretches a beat too long, when the language softens in a way that signals something is happening underneath the words, a steady leader notices and adjusts. Presence is the single most important variable in whether a performance conversation lands as developmental or punitive. And presence is something that has to be developed in human beings. It cannot be downloaded.
-
Holding complexity without collapsing it
Algorithms are built to give you an answer. Leadership, in a performance conversation, is sometimes built to refuse to give one. To sit inside the unresolved. To say, “The answer is not on this dashboard, and the question I want to ask you is going to take more than this meeting.” That is the exact move AI cannot make. And it is the exact move that, in 2026, separates the leaders whose teams trust them from the leaders whose teams quietly stop bringing them anything difficult.
Where the investment actually pays off
Most organizations are over-investing in AI fluency and under-investing in the human capability that AI cannot reproduce. The reallocation that consistently produces returns over a 12 to 18 month horizon is straightforward: invest in growing the leaders, not in upgrading the dashboard. Coaching for the front line. Vertical development for senior managers. Conversation infrastructure where leaders rehearse difficult conversations on real cases until the muscle memory is real.
That is the work the AI tool is now demanding our managers do. And it is the work most organizations are still treating as optional.
Sarah is not on a performance plan. The conversation we prepared for went somewhere very different, a real one about what was actually going on, a redistribution of work that wasn’t Sarah’s fault to begin with, and a quiet reckoning about how a project nobody had owned ended up consuming the time of someone the organization could not afford to lose. That conversation didn’t require a better dashboard. It required a manager who had been developed for it.
The AI was never going to save her. Her own growth as a leader did.
