In The Last Human Job¹, sociology professor Allison Pugh researches the profound effects of automation and artificial intelligence (AI) on professions centred around human connections. Through interviews and shadowing of individuals who are involved in what she calls connective labour, she argues that the increasing emphasis on efficiency, data, and automation threatens to erode the essential human-centric roles that rely on empathy, spontaneous human interaction, and mutual recognition of humanity. This connective labour is practised mainly by professionals like physicians, teachers, chaplains, and therapists. Pugh wonders what effect AI systems are having in moments when we express and experience our humanity and what their impact will be on belongingness, which is crucial to human thriving.

Connective labour is a deeply interactive, two-way process between provider and client, involving recognition of the other by the provider, as well as recognition of the provider by the other — seeing and being seen. This requires compassion and emotional attunement. The results of such relationships between physician and patient — dignity, purpose, and understanding — are well known to increase the chances of therapeutic success.

Industrial logic, introduced in factories by Taylorism at the beginning of the 20th century, has now also infiltrated connective labour, whether it be driven by the bottom line in for-profit businesses or in resource-deficient public institutions, such as health care. In many industries, counting, measuring, and applying all kinds of data is on the rise, infringing on time for people to pursue human connections. The cultural ascendance of data as authority has resulted in counting and measuring to make caring professions more effective and productive like factory work. As the demand for and costs of connective labour increase, many industries try to increase managerial control by introducing systems for data collection and analysis, imposing manuals and checklists, and implementing evaluation and assessment plans.

Occupations are being transformed, as these complex interpersonal jobs are reorganized to increase predictability and accountability. This is happening both in the private sector, with the goal of extracting profit, and in the public sector, with the goal of managing austerity in the context of limited time and resources. Both sectors impose very similar pressures on interpersonal work, emphasizing outcome over process. Scripting of the process of interactive service work threatens creativity and autonomy, transforms clients or patients into standardized objects, and demoralizes workers. For health care workers, this contributes to burnout and job dissatisfaction. For patients, it leads to a depersonalization crisis already visible in increasing social alienation and isolation, fragmentation of social connectedness, a growing societal trust gap, and loneliness. While the intention of increasing efficiency by scripting is to decrease time and resources needed, the depersonalization crisis increases the need for human connection at the point of service.

The power of connective labour lies in its capacity to create belonging. Belonging is an emotional state with real consequences, whose contours are honed by powerful societal forces shaping who is in and who is out. Belonging matters; research evidence shows that physical and mental well-being are influenced by whether someone is excluded. If the explosion of all things data severely removes or erases connective labour, it will result in a redefinition of what it means to be human!

Better than nothing

In light of increasing workloads and an insufficient number of providers, some say that AI is “better than nothing.” Pugh argues that not dealing with the root cause of resource shortages will increase the danger of eliminating the human connection in AI services. The resulting depersonalization of care might lead to reduced patient engagement and less adherence to treatment plans. It might also generate communication challenges and misunderstandings, as AI-generated responses so far lack contextual awareness or cultural competence that a human clinician brings to sensitive conversations. For instance, AI chatbots might address access to mental health support services, but they often lack the depth of understanding or empathy that human professionals provide. Systems are increasingly trained to simulate empathic or connective responses. Your chatbot apologizes for your frustration, your virtual assistant encourages you, your mental health app listens without judgement. But these systems don’t feel anything; they just know what to say. “Empathetic” algorithms are starting to outperform our managers at recognizing distress, while lacking a moral compass to decide what to do with that information. Performance is replacing presence, replacing not only connective labour, but also our emotional responsibility to one another.

However, AI continues to develop at light speed. In April 2025, only one year after the publication of this book, a research group in Dartmouth reported how a generative AI chatbot for mental health meaningfully reduced the clinically significant symptoms of adults with major depressive disorder, generalized anxiety disorder, or at high risk for eating disorders.² In contrast to Pugh’s concerns, users appeared to develop a bond with the chatbot. Therabot received ratings comparable to those of human providers when participants were asked whether they felt their provider cared for them and whether AI and the patient could work toward a common goal. This finding is important because the therapeutic alliance is often one of the best predictors of whether psychotherapy works or whether therapeutic suggestions are accepted.

Better than humans

AI is better than humans for non-connective tasks, such as pattern recognition in diagnostic images or pathology slides, designing new drugs, or managing the overwhelming volume of knowledge databases. Systemically induced biases are still possible and sometimes unnoticeable.

Better together

Pugh acknowledges that, when used thoughtfully, automation can augment human roles. They can be better together when they distinguish between thought or data and feeling or emotion, with the latter reserved for humans. Automatic transcription in medical settings frees up time for professionals to engage more deeply with patients. Another example would be a virtual intake nurse who does the rote work before human interaction, creating space to enhance rather than replace the human elements of connective work. This “better together” relationship is good as long as the freed-up time creates more contact between the professional and the patient and
not heavier workloads resulting in job dissatisfaction and burnout.

The research in this book is meticulous, blending interviews and observations across diverse fields to illustrate how connective labour fosters community and individual well-being. The book’s unideological approach, focusing on socioeconomic realities rather than on moralizing, lends credibility to Pugh’s warnings. Her writing is clear and easy, but her explanations are sometimes tedious and slow moving because of a repetitive structure in the nine chapters. The core message — that human work must be prioritized in an automated world — resonates powerfully. The book is thought-provoking for professionals and students in health care, education, or any care-giving jobs, and for policymakers grappling with the societal impact of AI.

At a time when AI continues to permeate various aspects of life, Pugh makes a compelling call to recognize and preserve the irreplaceable value of genuine human connection in the workforce. As a society and as caring professionals, we must ensure that technological advancement doesn’t come at the cost of our shared humanity. For AI to become a partner rather than a replacement for humans, we will need ethics, design choices, and policies to protect the connective care of the jobs that involve empathy, caring, and compassion. The danger is not AI itself but who uses it, how it is used, and for what purpose.

AI is penetrating our daily lives everywhere, from the annoying virtual assistant during telephone inquiries to the self-help checkout in grocery stores. The loss of a human voice or contact adds to the explosion of loneliness. Connective labour is facing an existential threat, but the blame does not only fall on AI. What countermeasures do we, as individuals, take? Do we take off our headphones to be more present in stores or waiting rooms, or put down our ever-present mobile phone to create opportunities for conversation? As we encounter people in our daily lives, are we present, do we give them our full attention?

Reference

1. Pugh A. The Last Human Job: The Work of Connecting in a Disconnecting World. Princeton: Princeton University Press; 2024

2. Heinz M, Mackin D, Trudeau B, et al. Randomized trial of a generative AI chatbot for mental health treatment. NEJM AI 2(4), 2025. https://ai.nejm.org/doi/full/10.1056/AIoa2400802