Column
Highly Capable AI Might Be Damaging Your Team — A Study on Workplace Perception
An experiment with 50 participants found that low-competency, low-proactivity AI produced better outcomes for employee ownership, job meaningfulness, and team dynamics than the high-performing alternative. For HR and AI implementation leaders, the design implication is significant.
Hi, I’m Keito Inoshita from Affectosphere Group.
Here’s a premise worth questioning: the more capable the AI system you deploy in the workplace, the better the outcome for everyone involved.
A paper published on arXiv in May 2026 (Kuntal Ghosh, Marc Hassenzahl, Shadan Sadeghian et al., arXiv:2606.00182) runs an experiment that complicates this assumption. In a study of 50 participants, the researchers varied AI systems on two dimensions — competency level and proactivity level — and measured how each combination affected employee self-perception and peer perception in a workplace context.
The finding that inverts the intuition: low-competency, low-proactivity AI produced better outcomes on ownership, sense of meaning, and role dynamics. High-competency, high-proactivity AI degraded professional identity and team dynamics.
3 Points for Today
- Value: AI designed thoughtfully can complement employee expertise and reinforce their sense of ownership.
- The risk: “High-capability and highly proactive” AI may erode role identity and interpersonal dynamics in ways performance metrics won’t capture.
- Implementation: Calibrating AI proactivity and autonomy by role type and seniority level could be a direct lever for reducing attrition risk.
① Why “More Capable AI” Can Backfire
When organizations evaluate AI tools for deployment, capability tends to dominate the conversation. Accuracy rate. Task coverage. Autonomy in execution. Speed relative to human baselines.
This paper asks a different question: what happens to the people working alongside that AI?
The 2×2 experimental design crossed high vs. low AI competency with high vs. low AI proactivity. Participants experienced each condition and reported on their self-assessment (sense of contribution, professional competence) and their perception of how they were viewed by colleagues (social image, role within the team).
The worst combination was high-competency, high-proactivity.
The researchers’ interpretation: an autonomous, highly capable AI system that actively generates proposals makes visible what it can do — and in doing so, implicitly reframes what the human contribution is worth. Employees begin to feel that their role could be performed by the system. Job meaningfulness erodes. Peer perception shifts as well: being seen as “the person who relies on AI” changes how someone is socially positioned in the team.
② Who Is Most Affected
The effect is most pronounced in roles where professional identity is tied to analytical or judgment-based expertise.
Consultants, analysts, strategists, managers. In these roles, “being the person who knows things and decides things” is a core part of how people understand their contribution. When a highly capable AI can produce the same output faster, the question “what exactly is my role here?” becomes harder to avoid.
Roles where AI functions more clearly as a tool — data cleaning, format conversion, scheduling — may be less affected because the worker’s identity isn’t as directly implicated in what the AI is doing.
Seniority also matters. Senior employees and high-value contributors tend to have more established professional identities, and may be more vulnerable to this displacement effect. Newer employees may be better positioned to relate to AI as something they’re learning from, rather than something that is replacing them.
Caveat: this is a single study with a relatively small sample size, using a specific scenario design. The effect sizes and patterns may vary substantially across industries, team cultures, and specific AI implementations. It shouldn’t be treated as a definitive finding, but as a useful hypothesis to test in your own context.
③ What HR and AI Implementation Teams Can Do Now
Translating this into practice doesn’t require waiting for more research.
Calibrate AI proactivity by role type and seniority
Rather than maximizing AI suggestion frequency and autonomous execution range uniformly, consider differentiating these settings. For senior knowledge workers, a design where AI provides information on request rather than proactively generating proposals may preserve the ownership dynamic. For operational roles with repetitive components, narrower autonomous execution may be a better fit.
If you’re formalizing this into a deployment policy, the relevant KPIs to track: job satisfaction scores at 6 months post-deployment, split by role category; and changes in role fulfillment ratings in engagement surveys.
Design for “collaborative agency”
The goal is a structure where the employee decides and the AI supports — not a structure where the AI proposes and the human confirms.
Specific design choices that maintain this: requiring human sign-off on all final decisions, presenting AI outputs as reference information rather than recommendations, and giving employees the ability to configure the AI’s scope of action within their workflow. These are design decisions, not just UX choices.
Run a small experiment before full deployment
As a near-term practical step: before scaling an AI deployment, run a low-proactivity version alongside a high-proactivity version with separate user groups, and measure both efficiency gains and satisfaction scores. Ask in post-deployment surveys whether employees experienced any change in how meaningful their work feels or how they perceive their role on the team.
The results will tell you something performance benchmarks won’t.
Performance Is Not the Only Design Variable
The underlying challenge this paper surfaces is about the design premise of workplace AI.
“More capable is better” is a technology-side assumption. From the perspective of the people working with the technology every day, having a highly capable AI as a colleague reframes the meaning of their work — and not always in a direction that helps the organization retain them.
Designing organizations where AI and humans work well together over the long term requires building structures where employees can answer “yes” to the question of whether they belong there. That won’t come automatically from deploying better-performing models.
That’s all for today!
References
- Kuntal Ghosh, Marc Hassenzahl, Shadan Sadeghian (2026). The New Social Image: How AI Competency and AI Proactivity Influence Self- and Peer-Perceptions in the Workplace. arXiv preprint.
* This article was written in part with AI assistance and may contain inaccuracies.