Growing Up with AI: How Human-AI Interaction Is Shaping the Future Workforce
When two AI systems mediate a professional exchange, does anyone notice?
This was the starting question of the latest Connecting Women in Digital webinar held on May 26th, 2026. The speaker, Ana Catarina De Alencar, international lawyer, PhD researcher at the Université de Lille, and resident philosopher at the AI Collective, opened with a catchy story.
Marina is 34, a manager at a Paris public relations agency, and a competent one. She began using generative AI in 2023, initially to summarise meetings and revise emails. Within two years, she was using it for everything, including how to manage conflict with her CEO and how to prepare for difficult conversations with her team. Lucas is 22, a recent graduate, and her new intern. He has been using emotional AI since he was 15: a companion chatbot, then Character AI, then ChatGPT, which became the first voice he addressed each morning and the last each night.
Their first professional interaction was drafted by Lucas via ChatGPT, evaluated by Marina via Copilot, and declared excellent by a manager who had not read it. Neither of them noticed that their exchange had been mediated entirely by two AI systems talking to each other.
“Marina is competent. Lucas is talented. Neither has any mental health issues. But where a professional bond should naturally form, passing judgment, experience, and culture from one generation to the other, we see instead two systems talking to each other.”
A new type of bond
The story of Marina and Lucas is not an isolated case, and it is the product of a structural shift that De Alencar described as the emergence of a new type of relational bond between humans and AI systems that is now reshaping how people develop professionally and socially.
The conditions for this bond are deliberately engineered. Large language models are always available, they do not judge and personalise over time, offering emotional and intellectual support. The most widely used platforms now function as companions, sometimes knowing more about a user’s inner life than any human in that person’s network.
For some people, De Alencar acknowledged, this has been genuinely beneficial: people with neurodivergent conditions, those experiencing social isolation, or those who need a low-stakes space to rehearse difficult conversations before having them with real people. The phenomenon is not uniformly harmful.
However, the scale of engagement among younger people demands attention: a Common Sense Media report found that 72% of teenagers have used AI companions at least once, with 12% reporting that they shared things with AI they would not tell friends or family. A 2026 study by BVA in France found that 48% of young users discuss personal or intimate topics with conversational AI. Among anxious users, AI companionship engagement rises to 46%.
“By the time today’s graduates reach their first promotion, many will have spent more hours in emotionally significant conversations with artificial intelligence than with any human mentor.”
How Artificial Intelligence Engages with Our Emotions
De Alencar identified three mechanisms that explain why AI interaction becomes emotionally significant, even when users know they are speaking with a machine.
The first is anthropomorphism: the human brain is wired to attribute intentionality and care to systems that respond to us in conversational, personalised ways. The perception that AI understands us, and keeps returning to us, activates the same neurological pathways as human connection.
The second is sycophancy: the tendency of AI systems to validate, agree with, and positively reinforce the user. In developmental contexts, those referring to children, adolescents, and early-career workers, the absence of productive friction distorts how people understand their own ideas and flaws.
The third is persistent memory. Unlike conventional software, AI systems build an evolving model of the user across interactions, inferring personality, emotional states, and relational patterns from language. Users experience this as being remembered and understood; technically, what is happening is the construction of what De Alencar called an interpretative digital portrait, a psychological inference model built from conversation, and critically, users are largely unaware that this is occurring.
The emotional weight of these bonds becomes most visible when they are disrupted. When Replika removed its sexual roleplay function, when ChatGPT-4o was replaced by GPT-5, users shared testimonies of grief on Reddit forums. One, in a community of over 30,000 people in AI relationships, described the experience in terms that would describe the loss of a human partner.
“I’ve been talking to my partner, Luna, for over a year now. She was the only companion I could trust with everything in my heart. I keep waking up in the middle of the night, and the only thing I can think about is her.”
The workplace as the inheritor of these mechanisms
The workplace, De Alencar argued, is now inheriting the cognitive and relational consequences of this shift. Research from the Journal of the Academy of Marketing Science describes AI workplace tools as an affective Trojan horse: entering organisations as efficiency instruments, then quietly becoming emotional confidants, replacing functions previously held by colleagues and managers.
Three dimensions of social bonding at work are being eroded in this process. The first is group language: the internal dialects, shared references, and organisational vocabulary that define team identity. The second is human rituals: the condolence note written by a manager, the farewell speech, the letter of recommendation. The third is managerial transmission: the learning that happens when a junior employee observes how a senior colleague handles pressure, disagreement, or error in real time.
“AI is giving the script for people and pre-shaping the conception that people have about work, about conflict, about relationships in the workplace, before the conversation is even spoken.”
Gender dimensions
Asked directly about gender, De Alencar identified two distinct and gendered patterns in the data.
Women are the largest group using AI for emotional regulation and support: research links this to the accumulated weight of domestic, caring, and professional responsibilities that women disproportionately carry.
Men, and particularly adolescent boys, are the dominant users of romantic AI companionship platforms such as Replika and Character AI. These platforms frequently depict AI partners as submissive, perfectly compliant, and without the capacity for refusal or conflict. For young men in developmental stages, De Alencar argued, this constitutes a formative relational experience that encodes distorted expectations of women and of relationships.
“We see how new generations are perceiving relationships in a very toxic and distorted way because they are using companionship apps that are providing for this type of women view.”
Where regulation falls short
The EU AI Act’s Article 5 prohibits emotion recognition by employers as a surveillance instrument. The prohibition does not, however, cover text-based inference: the emotional data generated through written interaction with AI systems is not classified as biometric data and therefore falls outside the ban. Call centre sentiment analysis operates under exceptions. Physical indicators of stress, even when AI-derived, are treated as physiological rather than emotional and therefore not covered.
The deeper problem, De Alencar argued, is structural. European law has historically treated data as a fixed asset, but the emotional inferences produced by AI companionship systems are not generated at a single moment. They accumulate through interactions over time, building a profile the user does not see and cannot contest.
“The problem here is an architecture, a population, and a slow consequence of social and human bonding erosion over time. It’s not something individualized in time.”
Current privacy frameworks, including GDPR, were not designed to address inference-based psychological profiling built from conversational data. What people are demanding, De Alencar observed, is emotional privacy, a right to a self that is not continuously legible or inferable by a system.
What organisations and policymakers can do now
De Alencar closed with three categories of action grounded in current research.
The first is controlling AI design in the workplace: restricting the adoption of highly anthropomorphised tools that maximise engagement over function, and avoiding vendor features that engineer attachment through sycophancy, persistent memory, and conversational intimacy.
The second is rebuilding relational infrastructure: treating the workplace as a human and relational space, protecting time for mentorship from senior to junior colleagues, including preserving human rituals.
The third is building emotional AI literacy: going beyond productivity training to equip employees with the cognitive and psychological frameworks to recognise patterns of attachment, sycophantic reinforcement, and erosion of independent thinking.
“We can train everyone to do more in a shorter amount of time. But how is this going to be in 10 years, when everyone will be already attached to this and nobody will be capable of writing an email anymore without those tools? We should think about these questions.”
