We have spent much of the AI debate arguing about jobs. Will AI “replace” workers? Will it “create more jobs than it destroys”? These questions are not wrong, but they are incomplete.

The deeper question is not how many jobs there will be. It is what kind of work will be left for humans, under what conditions, and on whose terms.

The automation narrative, updated

Previous waves of automation replaced specific tasks: weaving, assembly, bookkeeping. AI is different not because it automates more, but because it automates judgment-like work: reading, drafting, summarizing, recommending. The region of work where human discretion and machine capability overlap is expanding.

In call centres, models draft responses. In law firms, they summarize discovery materials. In radiology, they flag anomalies. In logistics, they plan routes. In software, they generate code.

Increasingly, AI systems are not just tools; they are co-workers—invisible ones that never sleep, do not unionize, and quietly reshape the pace and expectations of human labour.

The new division of labour

In practice, we are not seeing an immediate disappearance of entire professions. We are seeing redivision:

• Some tasks are fully automated.
• Some are augmented: humans handle edge cases, relationship work, and responsibility.
• Some are intensified: humans do the same job, but faster, with AI support.
• Some are devalued: what was once “skilled writing” becomes “prompting and editing.”

The risk is that “work without workers” becomes a policy and business goal: maximize efficiency by minimizing headcount while retaining the appearance of human service.

Algorithmic management and the invisible boss

One of the most important labour stories of the last decade has been algorithmic management: workers supervised, rated, scheduled, and disciplined by systems. In gig work, this has already become the norm. Now, similar techniques are creeping into white-collar environments:

• Productivity scoring dashboards
• Automated compliance checks
• Time tracking embedded in applications
• AI systems that rank employees by “impact”

Regulators are starting to respond. The EU’s Platform Work Directive explicitly addresses algorithmic management, giving workers rights to information and human review. Several jurisdictions require bias audits for automated hiring tools.

But the centre of gravity is clear: AI is not just changing what we do at work; it is changing who decides whether we are doing enough.

What remains distinctly human?

In a world where models can draft documents, debug code, and triage support, what is left that is uniquely human?

Three categories stand out:

1. Moral and relational work
   Roles where trust, care, and moral judgment are central: nursing, education, therapy, leadership. AI can assist, but outsourcing the core of these to machines risks hollowing out social bonds.
2. Institutional and political work
   The design of rules, laws, and governance structures is a human responsibility, even if AI can simulate implications. Delegating it would undermine legitimacy.
3. Work that insists on human presence
   Sometimes, the point of a role is that a human did it—even if a machine could. This is not inefficiency; it is a choice about dignity.

None of these are safe by default. They must be defended.

Bargaining power in an AI workplace

Labour markets are about more than matching people to tasks; they are about power. If AI strengthens employers’ bargaining position—by making it easier to replace or monitor workers—wages and conditions may suffer even if headline employment remains stable.

Conversely, if workers gain access to AI on their own terms, they may gain negotiating leverage:

• Being able to produce more, faster, with fewer resources.
• Being able to understand and challenge algorithmic assessments.
• Being able to organize, inform, and advocate more effectively.

Policy choices will determine which side dominates.

A labour agenda for the age of agents

A humanistic approach to AI and work might start with these priorities:

1. Transparency over algorithmic management
   Workers should know when they are being evaluated or managed by AI, on what criteria, and with what consequences. They should have a right to explanation and appeal.
2. Collective voice in deployment
   Decisions about deploying AI systems that materially affect working conditions—pace, surveillance, autonomy—should involve workers or their representatives.
3. Investment in human skill, not just cost reduction
   AI can be used to deskill or upskill. Policy and leadership should favour systems that genuinely extend human capability and career trajectories, rather than freezing people in narrow monitoring roles.
4. Protection for human-only roles
   In some domains, we may decide that certain tasks may not be fully automated, even if it’s cheaper. Law can express that choice—just as we set minimum staffing levels in critical services.

Work worth having

The question is not whether AI will change work. It already has.

The question is whether we use AI to design work worth having: work with autonomy, dignity, and meaning, or whether we use it to squeeze humans into ever tighter loops of monitoring, correction, and pseudo-oversight, while the real decisions migrate into systems no one quite understands.

“Work without workers” is a possible future. It is not an inevitable one. The design of AI systems, the structure of institutions, and the choices of regulators will determine how much of our working lives remain recognizably human.

---