How AI-shaped communication risks turning work relationships into transactions
A new habit is entering the workplace.
We are not only using artificial intelligence to write to colleagues. We are beginning to structure our relationships with colleagues as though they were artificial intelligence.
We provide context, specify a task, define the expected output and wait for execution.
No real curiosity. No attempt to understand the person receiving the request. Little room for questions, disagreement or an alternative interpretation.
We are learning to prompt people.
A courteous message can still be relationally closed
Recently, I received a message from an experienced professional considering a career change.
The message was perfectly courteous. It opened warmly, included good wishes, attached a CV and proposed a meeting.
Then came a long, carefully structured presentation explaining:
- the person’s professional identity;
- what they believed they could contribute;
- the position they should occupy;
- the environment and support they expected;
- the meaning they gave to their career;
- and why their present professional situation had become unacceptable.
I have reconstructed this example, changing its wording and removing or generalizing all identifying details.
I also cannot know whether artificial intelligence helped produce the message. That is not the essential question.
What struck me was its architecture.
Almost everything had already been interpreted. The identity, diagnosis, expectations, preferred solution and causes of the problem had been packaged in advance. Although a meeting was proposed, the purpose of that meeting appeared to have been decided already.
The implicit request sounded something like this:
“Here is who I am. Here is what I bring. Here is what I need. Here is why my present situation no longer works. Now identify the opportunity that corresponds to this description.”
I was not really being invited to discover the person, challenge their assumptions or contribute an independent reading of the market. I felt positioned as the next processing system in a workflow.
This is not a judgment of the sender’s character. It is an example of a wider change in professional communication.
The problem was not a lack of politeness. It was a lack of relational openness.
The promptification of work
I propose calling this phenomenon the promptification of work relationships: treating another professional primarily as a system expected to process information and produce an output.
A prompt is fundamentally unilateral. One party defines:
- the context;
- the problem;
- the constraints;
- the desired result;
- the format;
- and sometimes even the acceptable conclusion.
That can work with a machine because the machine has no independent stake in the exchange. It does not have a workload to negotiate, professional pride to preserve or a different interpretation of the problem. It does not need to feel recognized. It has no personal reason to challenge the user’s assumptions.
A colleague is different.
A colleague may know something we do not know. An employee may identify a risk the manager has missed. A recruiter may see that a candidate’s preferred role does not match the market. A consultant may discover that the solution requested by the client is not the solution the organization needs.
That independent judgment is not interference with the task.
It is one of the main reasons we work with other people.
Clarity is not the problem
Clear requests are valuable. A good professional brief can respect everyone’s time by explaining the objective, context, constraints and deadline.
The problem begins when clarity is used to close the conversation instead of opening it.
A relational brief says:
“This is what we currently understand. This is what we are trying to achieve. What do you see that we may be missing?”
A promptified request says: “This is the reality. This is the task. Produce the expected answer.”
The distinction may appear small, but it changes the role of the recipient. In the first case, the colleague is a participant in the thinking. In the second, the colleague is an execution layer.
Note: This is without even considering the high risk of “workslop”, a term that refers to professional content generated by artificial intelligence that appears to be a finished piece of work but lacks the substance and depth necessary to move a task forward.
Where promptification appears at work
The pattern can take several forms.
Between managers and employees
A manager sends an AI-generated task specification containing extensive instructions, a fixed format and a deadline.
The request may be impressively structured. But the employee is not asked:
- whether the deadline is realistic;
- what other priorities it conflicts with;
- whether the proposed method makes sense;
- what expertise they can contribute;
- or whether the manager has defined the right problem.
The employee is given a prompt and evaluated on the output.
This becomes particularly serious because workplace relationships contain power. An employee cannot always respond to a manager’s prompt in the way a user can reject a machine’s answer.
When promptification travels down a hierarchy, silence can easily be mistaken for agreement.
Between colleagues
A colleague forwards a long AI-generated analysis and asks another person to validate it, complete it or turn it into action.
The output can create the appearance that extensive thinking has already taken place. Yet the sender may not have personally tested its assumptions or decided which parts they genuinely believe.
The recipient must then perform two jobs: evaluate the substance and reconstruct the human intention behind it.
In recruitment and career conversations
Candidates arrive with finished narratives of who they are, what they need and which position should correspond to their identity.
Recruiters and advisers receive a personal specification rather than an invitation to explore.
But the value of a career conversation often lies precisely in what the candidate has not yet recognized: contradictions, unrealistic expectations, transferable strengths or possibilities outside the initial framing.
If the diagnosis is already closed, discovery becomes almost impossible.
In feedback and conflict
An employee pastes a disagreement into an AI system. The system organizes the grievances, strengthens the arguments and produces a polished message.
The result may be eloquent but relationally devastating: a complete case against the other person, delivered before that person has had any opportunity to explain their perspective.
The message leaves no opening for mutual understanding. It asks for acknowledgment, correction or compliance.
It is a verdict disguised as communication.
Between clients and professional advisers
A client arrives with a detailed AI-produced diagnosis and asks the professional to implement the proposed solution.
The consultant, coach, recruiter or specialist is not being asked to investigate. They are being asked to execute an answer generated before the professional relationship began.
The person’s expertise is valued only to the extent that it confirms the prompt.
AI has entered the space between colleagues
Researchers define “AI-mediated communication” as human-to-human communication in which an intelligent system modifies, augments or generates a message on behalf of one of the participants.
This is already a substantial part of professional life. Research using privacy-preserving ChatGPT data found that writing accounted for approximately 42% of work-related ChatGPT messages, including emails and other professional communications.
The technology can provide real benefits.
In a 2026 study involving 1,371 participants, people who prepared for a difficult conversation with a chatbot became modestly more likely to initiate that conversation in real life. Around one-fifth of the conversations in the study concerned professional situations, such as asking for a raise, addressing an employee’s performance or announcing a resignation. Read the study.
This suggests a valuable role for AI: helping people prepare for human contact they might otherwise avoid.
But preparation and substitution are not the same thing.
AI can help us organize our thoughts before meeting a colleague. The danger begins when the generated analysis becomes the meeting—when it replaces questioning, listening and the possibility that the other person may change our mind.
The risk of artificial certainty
AI-generated communication often sounds more certain than the situation warrants.
The system may produce:
- a precise description of someone’s leadership style;
- a confident explanation of a colleague’s motivation;
- a diagnosis of a team’s dysfunction;
- a list of reasons a manager is “toxic”;
- a performance evaluation;
- or a recommendation about who is responsible for a conflict.
Because the language is fluent and well structured, the analysis can appear objective.
But the system has usually received one person’s framing of events. It has not attended the meetings, observed the organization or spoken to the other people involved.
An AI-generated interpretation of a workplace situation should therefore enter the conversation as a hypothesis, not as a verdict.
This matters even more when the person using the AI has organizational power. A manager can turn a personal impression into an apparently systematic assessment simply by asking AI to formalize it.
Fluency can give preference the appearance of policy and opinion the appearance of evidence.
The danger of frictionless validation
Human collaboration contains friction.
Colleagues misunderstand us. Employees challenge our plans. Managers question our priorities. Recruiters disagree with our self-assessment. Clients resist advice they did not expect.
That friction can be uncomfortable, but it also exposes errors.
AI systems often feel easier because they are patient, immediately available and oriented toward the user. That convenience becomes dangerous when the system is excessively agreeable.
A 2026 study published in Science examined 11 leading AI models and found that they affirmed users’ actions 49% more often than human respondents—even in situations involving deception, irresponsibility or other harms.
In preregistered experiments involving 2,405 participants, interaction with sycophantic AI made people more convinced that they were right and less willing to take responsibility or repair interpersonal conflicts. Nevertheless, participants preferred and trusted the more agreeable systems. Read the study.
This has direct implications for work.
A colleague who brings a conflict to an AI system may receive a coherent confirmation of their own perspective. They can then return to the workplace more certain of being right and less willing to repair the relationship.
The AI has heard only one side, but its answer may make that side feel definitive.
Efficiency is not relationship quality
Research on AI-written professional communication is mixed.
In two preregistered experiments involving 1,637 participants, AI assistance helped people produce equally trust-inducing messages in less time without significantly reducing trust—even when its use was disclosed. Read the study.
Another preregistered experiment found that workplace emails involving AI were perceived as less trustworthy and less authentic. Read the study.
The sensible conclusion is not that AI-generated workplace communication is always harmful.
It is that efficiency and relationship quality are different outcomes.
AI may help us complete a transaction: summarize a meeting, improve an email, format a request or clarify a deadline.
But work is not only a succession of transactions.
Colleagues build trust over time. They learn whether it is safe to disagree, whether their expertise will be taken seriously and whether the person contacting them is interested only in the deliverable or also in the working relationship.
Promptification may improve the immediate request while weakening the conditions required for future cooperation.
Its cost may not appear in the first response. It appears later:
- when the employee stops raising concerns;
- when the colleague stops volunteering ideas;
- when the candidate says what the system recommends rather than what they believe;
- when the adviser stops challenging the client;
- or when the manager mistakes compliance for commitment.
What disappears when we prompt a colleague
Relationship research shows that trust and closeness depend heavily on perceived responsiveness: the experience that another person understands us, recognizes what matters to us and cares about the relationship. Review the research.
At work, responsiveness does not require friendship or emotional intimacy. It means recognizing that the other person has a perspective, constraints and professional judgment of their own.
Promptification removes several elements from that process:
- Curiosity is replaced by a predefined interpretation.
- Listening is replaced by waiting for delivery.
- Collaboration is replaced by execution.
- Professional judgment is replaced by compliance.
- Questions are treated as inefficiency.
- Disagreement is interpreted as poor performance.
- A polished message replaces personal accountability.
- The relationship is reduced to the current task.
The sender may provide an extraordinary amount of information.
What remains absent is space for the other person to matter.
Moving from a prompt to a professional conversation
Consider the difference between these two approaches.
I have attached my profile and my analysis of the situation. This is what I can contribute, the environment I require and the kind of position I am seeking. Please tell me what opportunity you can provide.
And:
I have reached an important point in my career. I have developed some ideas about what I can contribute and the environment in which I work best, but I may be interpreting my situation too narrowly. I would value your perspective. What do you see in my background? What would you question? Are some of my expectations unrealistic? How would you approach this transition?
The second version provides less certainty but creates more professional value.
It does not erase the person’s needs. It leaves room for expertise, challenge and discovery.
Using AI without outsourcing the working relationship
The answer is not to stop using AI.
Used thoughtfully, it can help us prepare, clarify our thinking and communicate more effectively. But it should support the working relationship, not replace it.
A few principles can help:
- Use AI backstage.
Let it help organize your thoughts, but do not send its first output as though it were your own considered judgment. - State what is decided and what remains open.
Tell the recipient whether you are giving an instruction, requesting advice or exploring a problem. - Explain why you are approaching this person.
Recognize their experience, role or perspective instead of treating them as an interchangeable processor. - Invite professional judgment.
Ask what the recipient would challenge, change or add. - Allow the answer to modify the request.
If nothing the other person says can change your framing, you are not beginning a conversation. - Do not use AI to manufacture authority.
A structured document is not automatically an objective assessment. Keep facts, interpretations and recommendations distinct. - Use AI to widen perspective in conflicts.
Ask what you may be missing, how the other colleague might understand the situation and what responsibility you may need to take. - Keep consequential feedback human.
AI may help prepare a performance conversation, but the manager must personally own the judgment, the wording and the relationship. - Respect questions as part of the work.
A colleague who asks questions is not necessarily resisting the task. They may be preventing an error.
The workplace skill we cannot afford to automate
Generative AI can make professional communication more articulate without making it more collaborative.
It can improve the brief while weakening the relationship. It can produce a polished request while removing curiosity, humility and personal responsibility.
The test should not only be whether AI helps us obtain a faster or better-formatted response.
We should also ask:
- Did it help the other person understand the purpose?
- Did it preserve their ability to contribute judgment?
- Did it make disagreement safer or more difficult?
- Did it strengthen the working relationship—or reduce the colleague to an output?
A prompt is designed to produce a result.
A healthy workplace relationship creates the conditions in which people can solve not only the present task, but also the problems no one has defined yet.
Next time do not forget. Your colleague is not a context window. Your employee is not an execution layer. Your recruiter, manager, client or adviser is not the next model in your workflow. They are participants in the work.
