The Deskilling Nobody is Talking about

The purpose of feedback is to teach, correct, guide, and mentor. When a manager outsources that to a chatbot, they lose more than accuracy. They lose their authentic voice as a leader, and over time, they lose the capacity to develop one.

The discomfort of giving feedback is precisely what builds character and experience. The friction of choosing your own words. The moment of sitting with uncertainty before you speak. The uncomfortable pauses. The second guessing afterward that sharpens you for the next conversation. Frankly, if you did not write it and you did not struggle with it, it is not feedback. It is a script. A manager reading a script is not leading authentically. They are performing.

HR and internal comms professionals have been "enabling" this problem for a while, and I want to be direct about that. I have written my share of scripts for leaders on what to say in difficult conversations. Most of us have. We told ourselves we were helping, and sometimes we were. But our scripts were just low-tech AI models, and that carries the same long-term cost: the leader never develops the muscle themselves. AI did not invent this failure mode. It scaled it.

I have deployed AI coaching platforms. The technology is genuinely impressive. But the consequence is that managers stopped struggling with the hard questions. Not because they had mastered them. Because the tool answered before they had to think. Thus, the risk is a generation of people leaders who are technically capable but humanly underdeveloped. The most alarming part is that they don't know what they're missing, because they have never had to develop it.

Bias laundering

Even in organizations without a sanctioned AI tool, managers are already using AI to assess real humans. Quietly. On personal accounts. Performance reviews have become one of the most common shadow AI use cases.

Here is the mechanism most people miss. A manager has a gut sense of an employee, fair or unfair, examined or not. They open ChatGPT and describe that employee in their own words. The prompt itself is the problem. If a manager writes "she struggles with executive presence" or "he isn't a team player," the AI does not push back. It does not ask what the manager has actually observed, or whether the framing is fair. It takes the input as fact and produces polished, confident, professional language built on top of it.

The bias is still there. It has just been laundered through a tool that sounds objective.

The manager now believes the assessment is more rigorous than it was. The employee, reading it, has no way to tell that what looks like measured feedback is a gut reaction with better grammar. HR sees a clean piece of writing that gives no signal anything is wrong, if anything the manager is praised for a well written assessment.

This is harder to detect than almost any other failure mode in performance management, and harder to challenge. The cost shows up later, in attrition, in engagement scores, in legal exposure, and in the slow erosion of trust in the performance system itself.

The honest counterargument

AI coaching tools are not all bad. For a junior manager with no mentor, working remotely, who has never been taught how to give difficult feedback, an AI coach is genuinely better than nothing.

What I am arguing against is not the existence of these tools. It is the quiet replacement of human development with them. There is a difference between AI as a scaffold for growing leaders and AI as a substitute for them, and that difference is the whole argument.

What companies can actually do

The fix is not more mentorship programs. That advice is thirty years old and it has not solved the problem yet. The work in front of companies is easier and more practical.

Raise the bar on what counts as feedback. Assisted writing is here. The argument over whether managers should use AI is already over. The argument that matters now is what good feedback looks like in a world where polish is free. The answer is examples. Lots of them. A performance conversation built on three or four specific moments, what the employee did, when, what the impact was, what the manager observed about how they handled it, cannot be generated by a chatbot working from a vague prompt. It can only come from a manager who has been paying attention. The presence of detail is the new signal that the thinking happened. This should be non-negotiable.

Audit written documents. Performance documents should be expected to carry specifics. Dates, behaviors, and direct observations. A review that says "she sometimes struggles with stakeholder management" should be sent back. A review that says "in the March product launch, she did X, which created Y impact, and I'd like to see her try Z next quarter" should be the standard. AI cannot fabricate detail the manager never observed. Detail is the antidote.

Replace scripts with role play. The fill-in-the-blank conversation templates, FAQ's and talking-point documents need to go. They were the low-tech version of the same problem we are now solving for. The replacement is role play, and this matters especially in remote and distributed companies, where leaders no longer absorb difficult conversations by watching a senior leader handle them in a meeting. A manager preparing for a hard conversation should practice it out loud, with another human, before they have it. Stumbling through it the first time in front of a peer or an HR business partner is uncomfortable. That is the point. The discomfort is the development. A polished AI script skips the part that actually grows the leader.

Make coaching the central pillar of the HRBP role. HR business partners have drifted toward process administration over the past decade, running cycles, owning systems, managing compliance. That drift is part of how we ended up here, and it is also why boards and CEOs increasingly ask what they are getting from the function. The HRBP role should be reset around real-time coaching of leaders. Sitting with a manager before a difficult conversation. Working through what they actually want to say. Challenging the framing. Role-playing the response. Not the polished version, the hard version. This is the work AI cannot do and should not do. It is also the work that develops leaders who can stand on their own, and the work that justifies the investment.

Hold the lines that cannot move. This is the BIG one. Performance improvement plans, terminations, anything touching mental health or personal crisis. These require a human in the room, preparing in their own words, accountable for the outcome. AI cannot draft the conversation. This is not a preference. It is a must.

The bottom line

The companies that get this right will not be the ones with the most sophisticated AI tools. They will be the ones whose leaders can still think, observe, and speak in their own voice, and whose HR function knows the difference between supporting that and replacing it.

We know AI can't replace human judgment. The companies that let it happen by default will pay for it later. In attrition. In culture. In a bench of leaders who cannot lead without a script.

Notes

A note on how this piece was written. I drafted it using AI as a writing assistant, over several rounds of editing. The research, the thesis, the examples, and the point of view are mine. The AI helped me organize the structure, tighten the prose, and pressure-test the argument.

I am including this note deliberately, as an example of what disclosure can look like in practice. The piece argues that the bar in a world of assisted writing should shift from "did you use AI" to "did you do the thinking." A short, honest note at the end of a written document is one way to demonstrate that standard. It separates what the tool contributed from what the human contributed, it normalize the use of these tools without hiding them, and it invites the reader to evaluate the work on its substance. Managers writing performance documents, board memos, or coaching plans can do the same. The form does not need to be formal. It just needs to be honest about where the thinking came from.

References

Leadership Development in the Age of Artificial Intelligence, Harvard Kennedy School (2024) https://www.hks.harvard.edu/sites/default/files/centers/mrcbg/Final_AWP_244.pdf

"Reimagine learning and development for the AI age," McKinsey & Company (March 2026) https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-organization-blog/reimagine-learning-and-development-for-the-ai-age

Cybernews coverage of shadow AI in the workplace (citing Microsoft/LinkedIn Work Trend Index) https://cybernews.com/ai-news/bring-your-own-ai-rise-shadow-ai-workplace/

Erica Pandey, "Meet the people using ChatGPT to write their performance reviews," Axios (February 2024) https://www.axios.com/2024/02/12/chatgpt-human-resources-performance-reviews

IBM, What Is Shadow AI?https://www.ibm.com/think/topics/shadow-ai

Palo Alto Networks, What Is Shadow AI? How It Happens and What to Do About Ithttps://www.paloaltonetworks.com/cyberpedia/what-is-shadow-ai

SHRM, "AI Coaches Will Be the Death of Annual Performance Reviews" (2025)https://www.shrm.org/topics-tools/news/hr-trends/ai-coaching