AI, judgement and what organisations should really be assessing

1. The conversation about AI has changed

Over the past year, we have found ourselves talking about AI with clients increasingly often. Not because they are asking whether their Employees should be allowed to use it. That question has, in many respects, already been answered.

Instead, the conversations are increasingly about how Employees are using it, and some of the unintended consequences we are beginning to see in practice.

Critical thinking, sound judgement and curiosity have always been important workplace skills. AI has not changed that. What AI has changed is the relationship between thinking and producing work. Increasingly, it is possible to produce polished, well-written and convincing work without necessarily reflecting the same depth of understanding behind it.

That means managers, interviewers and Employers can no longer judge the quality of someone’s thinking simply by looking at the quality of the output.

2. Using AI without losing the ability to think

Before we go any further, a brief disclosure: I used AI to help me write this article. That is precisely the point.

Ironically, it probably took me longer to write than many of the articles I have written without AI. Not because AI made the writing more difficult, but because it gave me the space to think more deeply about the ideas, challenge my own assumptions and keep refining them.

I did not ask AI to write an article about AI in the workplace. The observations and concerns came first, based on what we are increasingly seeing in our work. AI then became a tool to explore and develop those thoughts.

It helped me structure ideas, test arguments and improve the writing, but it also challenged my thinking. I questioned its suggestions, disagreed with some of its conclusions, added context, asked it to reconsider certain points and identified things I thought were missing. That back-and-forth often prompted new thoughts or helped develop existing ones further.

At the same time, AI took care of much of the time-consuming work of drafting, restructuring, refining the language and correcting grammar.

Used this way, AI should not reduce the amount of thinking we do. Instead, it should create more space for it.

If AI can save us time on drafting, grammar, formatting, summarising and other more mechanical aspects of our work, that time can instead be spent interrogating the substance: questioning assumptions, testing ideas, checking accuracy and applying our own knowledge, experience and judgement.

The efficiency should come from reducing the time spent producing the work, not from reducing the time spent thinking about the work.

Good AI use is therefore not simply about knowing how to write a good prompt. It is about engaging critically with what comes back: providing sufficient context, asking better follow-up questions, challenging assumptions, testing different perspectives and recognising when something does not make sense. It also means knowing enough about the subject matter to question the answer, rather than assuming that a confident and well-written response must be correct.

Ultimately, we need to remain responsible for the thinking. AI does not carry the consequences of the work or decisions we make with its assistance.

We do.

The aim should therefore not be to discourage Employees from using AI, but to equip people to use it in a way that strengthens critical thinking rather than replacing it.

3. AI only knows what you tell it

This has become particularly noticeable in Employee Relations matters.

Employees facing disciplinary action, performance concerns or workplace conflict are increasingly turning to AI for advice. There is nothing wrong with that in principle. The concern is that the advice they receive depends heavily on the information and context they provide.

Imagine an Employee asking AI: “My manager constantly criticises my work, excludes me from meetings and speaks to me in an intimidating manner. I feel like I am being targeted. Is this bullying?”

AI may very well respond that the conduct could constitute bullying or harassment and explain the Employee’s rights and possible recourse. But what was not included in the question? For example:

• The “criticism” may relate to legitimate and ongoing performance concerns.
• The meetings from which the Employee was excluded may have had nothing to do with their role.
• What was described as intimidating conduct may, in context, have been a firm but legitimate performance discussion.
• There may be a lengthy history to the matter that was not mentioned.
• Emails, witnesses or other information may paint a very different picture.

None of that context was included in the question. AI does not know what it does not know. And yet, because the response is usually articulate, confident and well-structured, it can carry a great deal of authority.

The Employee may then arrive at work armed with terminology such as “harassment”, “bullying”, “victimisation”, “retaliation” and “constructive dismissal”, without necessarily understanding what those concepts mean or the seriousness of making such allegations.

Employees who genuinely feel unfairly treated should, of course, be able to raise their concerns. However, serious allegations should not be made simply because an AI tool suggested a label based on one version of events. Depending on the circumstances, making reckless or knowingly false allegations can itself have serious disciplinary consequences.

4. This concern is not merely theoretical

Recent reports from the United Kingdom suggest that employment tribunals are already experiencing some of the unintended consequences of AI-assisted claims. Judges and employment lawyers have raised concerns about AI-assisted claims that are excessively lengthy, poorly focused or rely on arguments that the claimant does not necessarily understand. Unfair discrimination claims are among the more complex cases contributing to the growing demands on the tribunal system.

In one reported case, an Employee submitted 67 grievances across 282 pages with the assistance of AI but reportedly could not identify which of them he actually intended to pursue. In another, a claimant who had used ChatGPT left the judge with the “strong feeling” that she was “pursuing a claim she does not understand and cannot personally justify when asked.” [1]

There is, of course, another side to this. AI can improve access to information and help people articulate legitimate concerns that they may previously have struggled to express.

The problem arises when the ability to produce a convincing argument outpaces the user’s ability to evaluate whether that argument is sound.

That brings us back to the central issue: AI can provide the language, structure and confidence to articulate a case, but it cannot compensate for missing facts, poor judgement or a lack of understanding of the principles being relied upon.

5. When impressive work tells us less than it used to

The impact is not limited to Employee Relations. We are seeing it in many different aspects of the workplace.

Recruitment provides perhaps the clearest example. AI is making it easier than ever to produce work that looks impressive. It is not making it easier to know whether the thinking behind that work is sound. Candidates may be asked to prepare a presentation as part of an interview process and arrive with an exceptionally polished piece of work. There is nothing necessarily wrong with using AI to assist with the presentation, but Employers need to reconsider what they are assessing. A beautifully prepared presentation may tell us very little about the candidate’s own thinking.

Rather than trying to catch candidates using AI, Employers may need to change the assessment: Give the candidate a more complex scenario, introduce new information during the interview and ask them to explain or defend their reasoning. The question becomes less about what did you produce? and more about: do you understand it, can you think critically about it, and can you explain and defend your reasoning without relying on the presentation in front of you? These questions are much harder to outsource to AI.

6. AI as a thinking tool, not an answer generator

The same applies to everyday work. An Employee may use AI to research a topic, analyse information, draft a report, develop a proposal, solve an operational problem or prepare a presentation. AI can make all of these tasks significantly quicker. The opportunity, however, should be to use some of that saved time for more critical thinking, not less.

Good AI use is therefore rarely just asking one question and accepting the answer. It should involve a genuine back-and-forth: challenging the response, adding relevant context or new information, asking what may have been overlooked and testing its reasoning against our own knowledge, experience and judgement.

That is where AI becomes a useful thinking tool rather than simply an answer generator. For example, a manager considering a solution to an operational problem might receive an initial recommendation from AI but know immediately that one part will not work in their environment. They explain why; AI revises the recommendation; the manager challenges another assumption, adds a budget constraint and asks what risks they may still be overlooking. The final outcome may look very different from the first response.

That is where AI can be brilliant.

The real value is often not in the first answer, but in the conversation that follows.

It also illustrates why knowledge and experience still matter. The manager in the example above was able to challenge the recommendation because they understood the context, recognised its limitations and knew which questions to ask. That is true across professions. The better we understand a subject, the better equipped we are to use AI well. We are more likely to recognise what is missing, question assumptions and identify when an answer simply does not make sense.

Perhaps a useful test is this: If your name is on the work, could you explain, defend and take responsibility for it without AI sitting next to you? If the answer is no, AI may be doing more than assisting you with the work. It may be doing too much of the thinking for you.

7. Are we losing the human element?

Another consequence of AI that we have noticed is that we are starting to miss people’s mistakes. The slightly clumsy email, the sentence that is a little too long or simply an email where you can hear the person’s voice while reading it. These imperfections are part of what makes communication human.

Increasingly, workplace communication is becoming technically perfect and, at times, strangely impersonal.

There is nothing wrong with asking AI to help improve an email. It can organise thoughts, improve grammar and help ensure that a message comes across as intended. But there is a difference between asking AI to help you communicate your thoughts and asking AI to think and communicate on your behalf.

Equally important is recognising that AI is not always the right tool. Sometimes information needs to be independently verified. Sometimes professional advice is required. And sometimes what is needed is not a perfectly drafted message at all, but simply a conversation with another human being.

Particularly in people management, that difference matters. If you are recognising someone’s contribution, addressing poor performance, apologising, responding to a concern or having a difficult conversation, the person on the other side should still be able to hear you.

8. Employers need to rethink AI in the workplace

This is not simply an Employee issue. Employers and managers can misuse AI just as easily.

A manager should not, for example, insert an Employee’s version of events into an AI tool, accept its analysis and send an AI-generated response without properly applying their own mind to the matter.

Similarly, organisations should not assume that an AI policy dealing only with confidentiality, privacy and the uploading of Company information has adequately addressed the risks associated with AI in the workplace. Those issues are critically important, but they are only part of the picture.

Organisations also need to consider broader questions around appropriate use, accountability, verification and disclosure of AI-assisted work, while ensuring that Employees and managers are equipped to use AI critically rather than simply accepting what it produces.

Perhaps the biggest shift is that organisations may need to rethink how they assess work. Managers can no longer assume that the quality of someone’s thinking is reflected in the quality of the output alone. As AI becomes increasingly capable of producing polished work, Employers need better ways to evaluate the judgement, reasoning, and understanding behind it.

The discussion therefore needs to extend beyond whether Employees may use AI to how they should use it, how managers should assess AI-assisted work and, ultimately, how organisations adapt to a workplace where the quality of the output is no longer a reliable indicator of the quality of the thinking behind it.

9. In conclusion

As AI becomes increasingly embedded in the workplace, organisations need to ensure that their policies, practices and people evolve alongside it.

At FvT HR Consulting, this is an area we are actively developing in response to what we are seeing in practice. Our approach focuses on three key areas: helping organisations establish appropriate AI governance through practical workplace policies, equipping Employees to use AI responsibly and critically, and assisting managers in adapting to how AI is changing recruitment, performance management, and the assessment of work.

Ultimately, our aim is not simply to help organisations introduce AI into the workplace, but to ensure that it is used in a way that strengthens good judgement, critical thinking and better decision-making.

Perhaps that is the real challenge AI presents. Not whether AI can produce impressive work, but whether we continue to use it without losing the ability to think.

For more information on our AI workplace services, please contact Suzaan de Stadler at suzaan@fvtconsulting.co.za.

Written by Suzaan de Stadler (HR/Labour Consultant)

[1] The Economist, “The tragedy of the commons, AI edition”, 6 August 2026