2.1.1 Changing Nature of Work

Elena used to write product descriptions. Not glamorous work, but steady. She had a niche—luxury watches—and her clients loved her eye for detail, the way she could make a titanium chronograph sound like a love letter. She earned about $60,000 a year through freelance platforms, enough to live comfortably in Lisbon while working from her apartment overlooking the Tagus.

In March 2023, three of her five regular clients stopped placing orders. They didn't fire her. They just... went quiet. When she finally asked, one of them was honest: "We're using ChatGPT now. It's not as good as you, but it's fast and it costs almost nothing."

By the end of that year, her income had dropped to $31,000. She hadn't gotten worse at her job. She hadn't missed a deadline or lost her touch. The job itself had changed underneath her.

Elena's story isn't unusual. It's the new normal. And what's happening to freelance writers is just the opening act in a much larger transformation of what work looks like, who does it, how it's managed, and what it means to have a career.

The Freelance Collapse

Within eight months of ChatGPT's launch in late 2022, job posts for automation-prone freelance work on major platforms dropped by 21%. Writing and coding were hit hardest. Translation cratered. Customer service scripts dried up.

This wasn't gradual erosion. It was a cliff.

Researchers studying the freelance market found something particularly unsettling: it wasn't just low-quality, bottom-of-the-barrel work that disappeared. Generative AI tools reduced employment opportunities and income for freelancers across all skill levels. Even top performers, with stellar ratings and loyal clients, weren't safe.

The logic is brutally simple. If a company was paying $500 for a human-written product description and can now get something 80% as good for $2, the math doesn't care about craft, experience, or artistry. It doesn't even matter if the AI output isn't quite as polished. For most use cases, "good enough" is good enough.

Text-centric jobs—writing, translation, support—declined sharply. Interestingly, some categories went the other direction. Video editing saw a 39% increase in demand, perhaps because AI tools made video production more accessible, creating more raw material that still needed human editing. Prompt engineering emerged as an entirely new category. AI ethics consulting became a job title.

The freelance market didn't collapse. It's actually projected to cross $500 billion globally by 2025. But it shape-shifted. The work that remains is different from the work that vanished, and the people doing it need different skills.

Fiverr, sensing the shift, launched a program allowing freelancers to create "Personal AI" models trained on their own work—letting them scale output with AI rather than compete against it. It's a clever adaptation: if you can't beat the machine, merge with it. But most freelancers lack the technical sophistication or capital to build such systems. For them, the market transformation isn't an opportunity—it's a displacement.

The Algorithmic Boss

If you work for a large company in 2026, there's a good chance your boss isn't entirely human.

According to the 2024 European Working Condition Survey, 42.3% of EU workers are now subject to algorithmic management. That number ranges from 27% in Greece to 70% in Denmark. In the United States, the penetration is estimated to be even higher.

Algorithmic management—AM, as researchers call it—uses machine-learning algorithms to automate functions that used to belong to human managers: assigning tasks, evaluating performance, scheduling shifts, even hiring and firing. It's not a future scenario. It's happening now, at scale, across industries.

The most visible version is in gig work. If you've ever driven for Uber or delivered for DoorDash, you've experienced algorithmic management firsthand. The app tells you where to go, how to get there, how much you'll earn, and whether you're performing well enough to keep getting assignments. There's no human manager. There's a dashboard.

But algorithmic management has spread far beyond gig platforms. Warehouses use it to direct workers through optimal picking routes, measuring their speed against algorithmic benchmarks. Call centers use it to score agents in real time, flagging those who deviate from approved scripts. Software companies use it to track developers' code output, commit frequency, and time spent on tasks.

The promise is efficiency. Algorithms can process more information, optimize more variables, and make faster decisions than any human manager. They don't play favorites. They don't have bad days.

The reality is more complicated.

The Watched Worker

A Swedish transport company deployed a ride scheduling system that optimized pick-up and drop-off routes. On paper, it was brilliant: more rides per shift, less wasted time, better service. In practice, the algorithm scheduled drivers so tightly that they couldn't stop for food or bathroom breaks. The system optimized for efficiency and neglected the fact that its workers were human beings with bodies.

This isn't an isolated case. It's a pattern.

Algorithms enable continuous, data-intensive, and opaque forms of digital surveillance. Workers are increasingly subjected to real-time tracking, behavior scoring, and predictive assessments that operate invisibly and without recourse. Keystroke logging, mouse movement tracking, screen capture at random intervals, facial recognition to verify presence, sentiment analysis of email tone—these are no longer exceptional measures but routine features of enterprise software.

An estimated 80% of large companies now use some form of employee monitoring software. Much of it is AI-powered, meaning it doesn't just record what you do—it interprets it. It decides whether you're "productive" or "idle." It scores your engagement. It flags anomalies.

When autonomy and agency are reduced this way, research shows, creativity plummets. People stop taking risks. They stop experimenting. They optimize for whatever the algorithm measures, even if that's not what actually matters.

A customer service agent whose calls are scored for average handle time will rush through conversations, sacrificing quality for speed—because the algorithm rewards speed. A software developer whose productivity is measured by lines of code will write verbose, bloated code—because the algorithm doesn't measure elegance. You become what you're measured as. And if the measurement is crude, you become crude.

Machines as Teammates, Supervisors, and Subordinates

Deloitte's research on human-machine collaboration identifies three emerging relationship patterns between workers and AI: machines as subordinates, machines as supervisors, and machines as teammates. Each carries distinct implications for worker autonomy, skill development, and job satisfaction.

Machines as subordinates is the most comfortable arrangement. The worker directs the AI—drafting an email, summarizing a document, generating options—and remains in control throughout. The AI functions as a sophisticated power tool: capable, but subordinate to human judgment. This model preserves human agency and tends to produce the most positive outcomes for worker morale and creative output.

Machines as supervisors is the algorithmic management described in the previous section. The AI assigns tasks, evaluates performance, and determines schedules. Workers execute instructions and are assessed against algorithmic benchmarks, often without visibility into how those benchmarks are set or how performance is weighted.

Machines as teammates is the model most organizations claim to be pursuing. The AI handles what it does well—data analysis, pattern recognition, rapid synthesis of large information sets—while humans contribute judgment, creativity, empathy, and relationship-building. In theory, the two complement each other in ways that produce outcomes neither could achieve alone.

Model	AI's Role	Human's Role	Typical Context
Subordinate	Executes instructions	Directs and decides	Creative work, knowledge tasks
Supervisor	Assigns, evaluates, schedules	Executes, reports	Gig platforms, logistics, call centers
Teammate	Pattern recognition, data synthesis	Judgment, creativity, relationships	Collaborative professional roles

In practice, the teammate model is the hardest to implement and the rarest to find. It requires thoughtful system design, well-defined role boundaries, and organizational cultures that genuinely value human input rather than merely measuring output. What actually happens in most workplaces is a shifting combination of all three, with the balance tilting toward AI-as-supervisor more often than organizations acknowledge—or intend.

The Fragmentation of Jobs

AI isn't just replacing jobs. It's fragmenting them—and this subtler transformation may ultimately prove more consequential than outright automation.

A marketing manager's job used to be a coherent whole: strategy, writing, analysis, client relationships, team management, creative direction. Now AI handles the writing, automates the analysis, generates the creative options. What's left? Client relationships and strategic judgment. Important, but it's not the same job anymore. It's a different job wearing the same title.

This fragmentation is happening everywhere. A lawyer's job gets split into tasks AI can do—document review, legal research, contract drafting—and tasks AI cannot do: courtroom advocacy, client counseling, strategic negotiation. An architect's job divides into AI-generated design iterations and human aesthetic judgment. In each case, the role persists while its content is quietly hollowed out.

The implications for career development are serious. Junior lawyers used to learn by doing document review—tedious, yes, but educational. It's how they developed judgment, spotted patterns, and built expertise. If AI does the document review, where do junior lawyers learn? The same question applies across professions. If AI handles the grunt work, and the grunt work is how people develop expertise, then who becomes the expert in ten years?

Some optimists argue that freeing people from drudgery lets them focus on higher-value work. That's true in theory. In practice, it often means there's less work overall, and what remains is either high-skill and therefore inaccessible to newcomers, or supervisory—managing the AI—which requires a different skill set entirely.

The erosion of entry-level roles also has structural consequences. Many institutions—law firms, hospitals, consultancies, architecture practices—are built around pyramidal staffing models in which junior work is both profitable and pedagogical. AI disrupts this structure by making junior work less necessary while leaving the organizational model unchanged. The result is a growing mismatch between the work available and the people trying to enter the professions that do it. The job doesn't disappear. It dissolves.

The Always-On Economy

AI doesn't sleep. It doesn't take vacations. It doesn't call in sick. This creates a subtle but relentless pressure on the humans who work alongside it.

If AI can generate reports at 3 AM, why can't you review them by 7 AM? If AI can respond to customer inquiries instantly, why is your response time measured in hours? The expectation of constant availability—already a problem in the smartphone era—gets amplified by AI's tirelessness. The gap between what is technically possible and what is humanly sustainable becomes a source of chronic organizational tension.

Research on workplace well-being shows that AI-augmented roles often come with higher expectations, tighter deadlines, and increased cognitive load. One study found that while AI reduced time spent on certain tasks, total working hours didn't decrease. The time saved was immediately filled with additional tasks, higher output expectations, or the administrative overhead of managing AI tools themselves—reviewing outputs, correcting errors, and navigating the complexity of human-machine workflows. Researchers describe this as a "productivity paradox": AI makes workers more efficient per task, but efficiency gains translate into expanded workloads rather than reduced hours.

There is also evidence of skill atrophy in heavily AI-assisted environments. When AI handles routine tasks automatically and continuously, workers may gradually lose the ability to perform those tasks independently—a phenomenon analogous to GPS dependency reducing navigational ability. If systems fail or produce errors, workers who have grown reliant on AI scaffolding may lack the foundational competence to catch mistakes or operate without it. The always-on AI simultaneously increases output and quietly erodes the capabilities it was designed to augment.

The New Geography of Work

AI is also reshaping where work happens—and who gets to do it.

Remote work, already accelerated by the pandemic, gets another boost from AI. If your job involves interacting with AI tools, it can be done from anywhere with a laptop and an internet connection. AI handles coordination, scheduling, translation, and communication across time zones, making distributed teams more functional and the office more optional.

But there is a structural consequence to this flexibility: if a job can be done remotely with AI assistance, it can be done remotely by anyone, anywhere. Location no longer provides protection from global competition. A company in New York no longer needs to hire a New Yorker at New York wages. It can hire someone in Manila or Nairobi who uses the same tools and charges a fraction of the price.

This dynamic creates contradictory pressures. AI democratizes access to tools and platforms, lowering the barriers that once separated workers in high-income countries from those in lower-income ones. For workers in the Global South, that can mean genuine new opportunity—access to international clients and markets previously out of reach. For workers in higher-wage economies, it means intensified competition from a global labor pool that is increasingly equipped with equivalent capabilities.

The result is a form of global wage convergence, though not a comfortable or equitable one. High-income country workers experience downward pressure on pay for remote-compatible work, while some workers in lower-income countries gain access to better-compensated opportunities than their local markets could provide. Where these trajectories meet is not a negotiated settlement—it is determined by market forces that tend to favor cost efficiency over worker welfare. AI democratizes access to tools but globalizes competition for work. The market expands, but the slice per person shrinks.

The Meaning Crisis

Beyond the economics, there is a deeper question that AI is forcing many workers to confront: what is work for?

For most of human history, work was survival. In the 20th century, it became identity—the first question at every social gathering. Your job defined your status, your daily structure, and your sense of self-worth. AI is now complicating that identity in ways that feel genuinely novel.

Workers whose skills are suddenly replicable by software at near-zero cost face a disorienting disconnect: their competence is not in question, but the market no longer assigns it significant value. A skilled professional can produce excellent work and still find that clients are satisfied with a cheaper, AI-generated alternative. The issue isn't quality—it's substitutability. And for people who built their identities around a craft, the distinction matters less than it might seem. When the market stops caring about the difference between human and machine output, the psychological effect resembles obsolescence, even when the technical reality is more nuanced.

Psychologists describe this as a "meaning crisis," and it is emerging across professions. Artists whose styles are mimicked by image generators. Programmers whose code is assembled by Copilot. Translators whose output is indistinguishable from DeepL. The work still exists, in some form, but it no longer feels like the same work. The sense of unique contribution—the feeling that one's particular skill matters and could not simply be reproduced—is central to how many people derive meaning from their labor. When that uniqueness is challenged, the psychological impact can be severe even when the financial impact is modest.

This is not simply a matter of workers needing to adapt or retrain. Retraining addresses economic displacement, but meaning is not an economic category. A professional who successfully transitions to AI-augmented work may maintain their income while losing the craft that gave their work purpose. The transition is successful on paper and hollow in experience. How societies address this dimension of AI's impact—not through retraining programs alone, but through reimagining the relationship between work, identity, and value—is one of the more difficult and underappreciated challenges ahead. The psychological dimensions of this question are explored more fully in Chapter 4.

What Work Becomes

Predicting the precise shape of work a decade from now would require more certainty than the evidence supports. But several trajectories are clear enough to describe with confidence, even if their full implications remain to be seen.

Work is becoming more fluid. The stable, decades-long career at a single employer is already rare and growing rarer. In its place, portfolio careers are emerging—combinations of freelance engagements, part-time roles, and project-based arrangements that can add up to a livelihood but rarely provide the security, benefits, or identity continuity of traditional employment. This fluidity offers autonomy for some workers and precarity for many others.

Work is also becoming more intensively supervised. Whether through human managers wielding AI dashboards or through algorithms operating directly, workers are being monitored, measured, and evaluated at a level of granularity that would have been both technically impossible and socially unacceptable in previous generations. The information asymmetry between employer and worker is growing, with significant implications for bargaining power and working conditions.

At the same time, work is polarizing. AI tends to enhance the productivity of high-skill workers who exercise judgment and creativity, while automating or algorithmically managing lower-skill routine work. Middle-skill roles—the clerical, administrative, and junior analytical jobs that formed the backbone of the 20th-century middle class—are the most vulnerable. This hollowing out of the middle is not a new observation, but AI is accelerating a process that labor market reforms have so far failed to adequately address.

Finally, work is becoming more global. AI tools reduce the technical barriers that once confined competition for knowledge work to proximate labor markets. This expands opportunity for workers in lower-income countries and intensifies wage pressure for those in higher-income ones.

These trajectories do not resolve cleanly into either optimism or pessimism. They coexist: more opportunity and more competition, more efficiency and more surveillance, more flexibility and less security. Navigating them will require not just individual adaptation but deliberate institutional responses—in policy, in education, and in how organizations choose to deploy the AI tools now available to them.

Key Takeaways

Displacement is already underway in automation-prone fields. Generative AI has reduced both the volume and the value of human-produced freelance work across all skill levels, not just the most routine, with job postings in writing, translation, and customer service falling sharply following the introduction of capable language models.
Algorithmic management is now mainstream. Roughly 40% of EU workers and an even higher proportion of US workers are subject to some form of AI-mediated oversight. Documented effects include reduced autonomy, increased stress, and distorted incentives when workers optimize for what the algorithm measures rather than what actually matters.
Three models of human-AI work relationships exist—subordinate, supervisor, and teammate—representing a spectrum of organizational choices, not an inevitable progression. The teammate model offers the most promise but requires deliberate design; the supervisory model is the most common default.
Job fragmentation, not just job elimination, is the most pervasive near-term effect. AI is disaggregating roles into AI-compatible and human-residual tasks, disrupting career ladders and eroding the entry-level work through which expertise has historically been developed across many professions.
Productivity gains from AI have not consistently translated into reduced working hours. Time saved tends to be reallocated to additional tasks or absorbed by the overhead of managing AI systems, and some evidence points to gradual skill atrophy in workers who become dependent on AI scaffolding.
The meaning crisis is real and distinct from economic displacement. Workers across income levels and skill categories are confronting the challenge that AI poses to professional identity—not just to wages and employment. This dimension of AI's impact is underappreciated in policy discussions and is explored further in Chapter 4.

Sources:

Last updated: 2026-02-25