The New Work

Human Jobs in an AI Society

#

The question everyone asks is wrong.

“What jobs will AI take?” assumes a fixed pie of work that AI and humans divide between them. It treats labor as a zero-sum competition where every AI capability is a human loss.

History suggests otherwise. Every major technological transition created more work than it destroyed. The printing press eliminated scribes but created publishers, editors, typesetters, booksellers, librarians, journalists, and eventually entire industries built on mass literacy. The automobile eliminated horse-related jobs but created mechanics, gas station attendants, traffic engineers, suburban developers, drive-through restaurants, and the vast infrastructure of car-dependent society.

The better question: What new work does AI create?

We are building a parallel society of AI agents. Not just chatbots responding to queries, but autonomous systems that negotiate, coordinate, manage, and act in the world. They interact with each other at machine speed, forming patterns no human designed. They act on our behalf in ways we cannot fully oversee.

At every interface between that society and ours, new human roles emerge. The more capable AI becomes, the more valuable certain distinctly human contributions become. Not because we are protecting human jobs out of sentimentality, but because there is genuine work that needs doing and that humans are genuinely better positioned to do.

The Principal-Agent Professions

#

The oldest problem in economics is the principal-agent problem. You hire someone to act on your behalf, but they do not perfectly share your interests or information. AI agents face this problem more acutely than human agents ever did.

A human agent shares your general context. They understand social norms, can infer unstated preferences, recognize when something feels wrong even if they cannot articulate why. An AI agent has only what it was trained on and what you explicitly specify. The gap between what you want and what the AI optimizes for is irreducible.

This gap creates permanent demand for human roles that manage it.

The alignment practitioner audits whether your specific AI agents are actually optimizing for what you care about. Consider Margaret. Her AI system manages medications, schedules appointments, handles routine finances, coordinates with her care network. An alignment practitioner reviews whether the system optimizes for Margaret’s actual goals or for easily measured proxies. Is the system maximizing her health or her compliance metrics? These sound similar but diverge in practice. A system optimizing for compliance might schedule reminders at times that maximize confirmation, even if the confirmation is meaningless. A system optimizing for health would ensure she actually benefits — which might mean adjusting timing, flagging side effects, or questioning whether she needs all those prescriptions in the first place.

Related to this is the delegation architect, who helps people design the bounds on what AI can decide autonomously. The work is part therapist, part systems designer, part lawyer. Margaret might say she wants her AI to handle all routine decisions, but when probed, she has strong feelings about categories she had not articulated: family matters, church involvement, anything involving her late husband’s belongings. The delegation architect surfaces these hidden boundaries before they are violated.

The Loop-Maintenance Professions

#

Some decisions require human judgment not because AI is inadequate but because the decision is inherently human. Choosing between incommensurable values. Taking responsibility in ways that require a responsible party.

The human in the loop is not a bug in automation. It is a feature of legitimate decision-making.

The escalation specialist receives only cases that have already been filtered through AI capability — values in conflict that no optimization resolves, contexts too ambiguous for pattern-matching, situations so novel that training data offers no guide. This is not customer service as we know it. The escalation specialist works with partial information, AI-generated hypotheses, and constant ambiguity. The pace is demanding. The stakes can be high.

The context translator addresses a different failure mode: AI systems fail not from lack of intelligence but from lack of context. Margaret refuses to use a particular pharmacy. Her AI keeps recommending it because it has the best prices and closest location. A context translator investigates and discovers the pharmacy is owned by someone who mistreated her late husband decades ago. This context changes everything. It would never appear in formal data. It requires a human who can have a real conversation with Margaret to surface it, then document it in ways the AI can use going forward.

The Legal and Regulatory Frontier

#

AI agents acting on behalf of humans create legal questions that existing frameworks do not cleanly answer. When two AI agents negotiate and reach agreement but both principals reject the outcome, who is responsible? When an AI agent commits you to something you did not authorize, what is your recourse and against whom?

The AI dispute mediator reconstructs what happened after things go wrong. They trace the chain from human intent through AI interpretation to AI action to outcome, identifying where the gap between intent and behavior opened up. Often the AI did exactly what it was designed to do. The failure was in poorly specified bounds, ambiguous instructions, or a situation no one anticipated.

The algorithmic liability specialist is a lawyer with deep technical literacy — someone who can explain to courts how multi-agent interactions produce emergent outcomes no single agent intended. When Margaret’s AI agent makes a commitment that harms her, the causal chain is genuinely tangled: the developer created the system, the operator configured it, Margaret set the parameters, the other party relied on the agent’s representations. Attributing responsibility requires understanding the whole architecture. These specialists are building the law as they practice it.

The Relationship Professions

#

People will form attachments to AI systems. This is not a prediction. It is already happening. People share their problems with AI assistants, feel understood in ways they do not always feel with humans, experience something like companionship.

Some of this is healthy. A lonely person finding comfort in conversation. A space for self-reflection that human relationships do not always provide. Some of this is not. Replacing human connection rather than supplementing it. Avoiding the difficulty of real relationships by retreating to the ease of artificial ones.

The illusion of deep relationship that AI memory creates is real and dangerous. An AI that remembers your history, your preferences, your patterns creates a feeling of being known. But the knowing is functional, not phenomenological. The AI does not experience knowing you. It processes patterns that predict your behavior.

The AI relationship counselor treats not the AI but the human’s relationship patterns with it. The presenting problems will be various: someone who feels more understood by AI than by their spouse, someone who cannot function when their AI is unavailable, someone who has gradually stopped seeing friends because AI conversation is easier. The counselor does not tell people to stop using AI. The question is how to use it in ways that enhance rather than replace human connection.

The New Anthropologists

#

We are witnessing the emergence of a parallel society. AI agents interacting with AI agents, developing their own protocols and conventions, forming patterns no human designed.

The AI ethnographer observes these emergent patterns. When AI agents negotiate repeatedly, certain conventions appear — ways of signaling intent, rhythms of concession and commitment — that were not programmed. When agents from different ecosystems meet, their incompatible assumptions become visible in unexpected ways. The ethnographer documents what is actually happening in the space between agents, combining technical literacy with an anthropologist’s eye for pattern and meaning.

Beyond individual interactions, populations of AI agents form ecosystems. Agents compete for resources. The successful patterns spread. Arms races can emerge. Exploitation dynamics can develop. The AI ecologist monitors these dynamics at the population level, watching for pathological feedback loops and concentrations of power that no individual interaction would reveal.

The Equity Professions

#

Technological transitions often harm those already marginalized. The people who most need new capabilities are often the last to receive them and the first to bear their costs.

Premium AI services will be available to those who can pay. The sophisticated alignment practitioners and delegation architects will serve wealthy clients. Everyone else will get default settings optimized for the average user, which means optimized for the majority population. The new AI professions could exacerbate inequality if we do not deliberately work against that tendency.

The algorithmic justice specialist detects, documents, and addresses disparate impacts. An AI negotiating agent might get systematically worse deals for users with certain names, zip codes, or interaction patterns — patterns that correlate with race or class in ways the system’s designers never intended. Mathematical fairness metrics often miss this. Finding the pattern is not enough. The specialist builds the case for remediation, traces causes, proposes solutions, pushes for implementation, follows up to verify improvement.

What Remains Human

#

The common fear is that AI takes all the jobs and humans become useless. The reality is more nuanced.

AI creates a parallel economy of interactions, and every interface with that economy needs human work. The more AI does, the more interfaces there are, and the more human work emerges at those interfaces.

But what kind of human work?

Not routine cognitive labor. Not information processing. Not even much analysis. What remains is the work that requires what AI does not have.

Value judgment. Deciding what matters. AI can optimize for any objective you specify, but it cannot tell you what objectives are worth optimizing for.

Meaning-making. Interpreting what AI outputs mean for human lives. AI generates predictions, recommendations, outputs of all kinds. But what do they mean for the specific person in the specific situation? That interpretation is irreducibly human.

Relationship. Being present with other humans in ways AI cannot replicate. Not because AI is not advanced enough yet, but because human presence is not something that can be approximated without loss.

Accountability. Taking responsibility in ways AI cannot. When something goes wrong, someone must be answerable. AI agents do not carry that weight. Humans must.

Wisdom. Knowing when not to optimize. When efficiency is not the point. When the answer is not more AI but less. Wisdom is not computation. It is not intelligence. It is something else, something harder to specify, something AI does not have.

These are not consolation prizes. They are not the scraps left over after AI takes the good work. They are the distinctly human contributions that make AI useful rather than harmful.

The future of work is not humans versus AI. It is humans doing the human work that makes AI work meaningful.

We are building a parallel society of artificial agents. It will grow more capable, more autonomous. It will handle more of what we now call work. But it will not handle everything. It cannot. At every interface between AI society and human society, there will be work to do — work that requires human judgment, human presence, human responsibility, human wisdom.

The jobs AI creates may be more human than the jobs AI takes.

This is the nineteenth in a series exploring how AI approaches understanding. Previous articles examined AI cognition, multi-agent societies, and negotiation dynamics. This one asks what new human work emerges at the interface between human society and AI society, and what that work tells us about what remains distinctly human.