How AI Is Changing Work

Learning Objectives

• Explain how AI automation differs from previous waves of technological displacement
• Identify the characteristics that make a task or role vulnerable to AI automation
• Articulate why augmentation — not wholesale replacement — is the most likely near-term outcome for most knowledge workers

The Shift That Is Already Underway

We are not waiting for AI to change work. It is already happening.

Software developers are using AI coding assistants that write, review, and debug code. Lawyers are using AI tools that review contracts in minutes instead of hours. Marketers are generating campaign copy and social content with a prompt. Customer service teams are deploying AI agents that handle the majority of tickets before a human sees them.

The examples are no longer hypothetical. Klarna's CEO announced in 2024 that their AI assistant was doing the work of 700 full-time customer service agents — the company reduced headcount from roughly 5,000 to 3,800 through attrition, explicitly crediting AI. Duolingo replaced contract content creators with AI in early 2025. Chegg, the education company, saw its stock collapse roughly 99% from its peak after CEO Dan Rosensweig directly attributed declining subscriber growth to ChatGPT.

The question is not whether AI will change your field. It is already changing most fields. The useful question is how — and what that means for you personally.

Most people already expect the answer to be bad. In a survey of 37 countries published in September 2026, Pew Research Center found that in 34 of them, people are more likely to expect AI to bring fewer jobs than more over the next 20 years. Worry runs highest in wealthy countries: around seven in ten adults in Australia, South Korea and the United States expect job losses, and the American share is up 7 points in two years. The same survey found feelings more mixed than that suggests, with a median of 41 percent equally concerned and excited about AI and 37 percent mostly concerned. Treat these as a measure of expectations rather than of outcomes, since a survey records what people believe will happen. The rest of this lesson is about what the evidence so far says is actually happening.

What AI Is Actually Good At

To understand which roles AI affects most, you need to understand what AI is actually good at right now.

AI excels at tasks that are:

• Pattern-based: Recognizing patterns in large datasets, text, images, audio — tasks where the answer can be inferred from examples
• Well-defined: Tasks with clear inputs and clear success criteria
• High-volume: Repeating the same type of task thousands or millions of times at low marginal cost
• Language-based: Generating, summarizing, translating, classifying, or extracting text

AI currently struggles with tasks that require:

• Novel physical dexterity: Unstructured physical manipulation in unpredictable environments
• Deep contextual judgment: Decisions involving complex ethical trade-offs, power dynamics, or unstated human needs
• Long-horizon accountability: Work where the consequences unfold over months or years and require sustained relationship management
• Genuine creativity: Truly original conceptual leaps (as opposed to recombining existing patterns — which AI does well)

Why This Wave Is Different

Previous waves of automation — factories, spreadsheets, computer-aided design — primarily affected manual labor or narrow, clearly defined computational tasks. A robot could weld a car frame; it could not write the press release about the car launch.

The difference with modern AI: It affects knowledge work directly.

Large language models can draft, summarize, translate, reason, code, and analyze. For the first time, tasks that required a university education and years of professional experience can be partially automated. This is new.

The scale of potential impact has been noted by major institutions:

• Goldman Sachs (2023): Estimated 300 million jobs globally could be "exposed to automation" — meaning AI could perform a significant share of their tasks, not that all 300 million jobs disappear. Follow-up research in 2024 shifted emphasis toward productivity gains, estimating AI could boost global GDP by 7% over a decade (roughly $7 trillion)
• WEF Future of Jobs Report (2025): Projects 170 million new jobs created and 92 million displaced by 2030 — a net gain of 78 million roles globally, though 59% of the workforce will need reskilling or upskilling. The fastest-declining roles include cashiers, administrative assistants, data entry clerks, and bank tellers
• McKinsey Global Institute: Estimates AI overall could deliver $13 trillion in additional global economic activity by 2030 (2018 estimate), with generative AI specifically adding $2.6–4.4 trillion annually on top of that (2023 estimate). By 2024, 72% of organizations had adopted AI in at least one business function
• IMF (2024): Estimates 40% of global jobs are exposed to AI; in advanced economies, the figure rises to 60%. About half of those exposed jobs may benefit from AI integration, while the other half could see reduced demand
• OECD (2024): Found that 27% of jobs across OECD countries are in occupations at high risk of AI automation — but emphasized that actual job losses have so far been limited, with the bigger near-term effect being task transformation

Every figure above is a projection. Two studies are not, and they disagree in an instructive way:

• Stanford Digital Economy Lab, "Canaries in the Coal Mine?" (revised August 2026): Brynjolfsson, Chandar and Chen measured what actually happened, using payroll records covering millions of US workers. Employment for 22-to-25-year-olds in the most AI-exposed occupations now sits about 19% below where it would be had it tracked similarly aged workers in less-exposed jobs — up from 15% a year earlier. Workers in the same occupations with more experience show no comparable decline
• CESifo working paper, "The Early Impacts of AI on Employment Among Recent College Graduates" (September 2026): Robert Fairlie and Jane Wu used US Census survey data and found "no evidence of any significant, widespread displacement or reduction in hiring of recent college graduates." Unemployment among graduates aged 22 to 25 was 7.3% in summer 2026, inside the range of every summer since 2022, and graduates in the most AI-exposed jobs fared no worse than comparison groups

The mechanism matters more than the number. The gap comes from reduced hiring, not layoffs — firms are opening fewer junior roles rather than cutting existing staff — and it shows up in headcount rather than wages. It concentrates in work built on codified knowledge that a model can reproduce, while roles depending on tacit knowledge built through judgment and experience hold steady.

This is uncomfortable for the advice people usually give. "Gain experience, then you'll be safe" assumes there is a rung to stand on. If the contraction lands on entry-level hiring, the ladder is being pulled up from the bottom while the people already standing on it are fine. That is a different problem from mass unemployment, and it explains how aggregate figures showing limited job losses can be true and beside the point at the same time — accurate for the workforce as a whole, useless to a new graduate.

The second study is the check on that reading, and the two are measuring different things. Stanford tracks headcount inside particular occupations from payroll records; CESifo asks whether recent graduates are finding work at all. Both can be true at once if graduates are being hired into other roles rather than the exposed ones. The CESifo authors are careful to call summer 2026 a first test, and they expect the classes of 2027 and later may feel more of an effect as workplace AI use deepens. The honest summary for a new graduate is that the exposed entry rungs are narrowing while the overall job market for graduates has not yet moved.

One employer has now shown the same pattern from the inside. In October 2026 Anthropic published a study of its own workforce, built from hiring records, calendars, surveys and interviews. Junior roles fell from 21 percent of its technical staff to 14 percent from early 2025, and junior hires from 27 percent of technical hiring to 9 percent. Over the same period, hours worked by its AI agents grew to six times its employees' hours, and output per person rose in code, sales and contracts. Hiring managers said they now weigh autonomy and judgment above technical expertise. Read it as one company that has pushed AI further into its own work than almost anyone, writing about itself, rather than as a forecast for every employer. It is still the clearest first-hand account yet of what the Stanford figures look like inside a firm.

There is a second question underneath the hiring one: does working with AI still teach you the job? The first long field experiment on it, published in October 2026 by economist David Autor and Google researchers through the National Bureau of Economic Research, gave an AI drafting assistant to 133 patent lawyers at eleven law firms for three months and withheld it from a control group. With the tool, everyone's drafts got better, and junior lawyers also finished faster. The real test came at 90 days, when every lawyer had to find and fix the errors in a flawed patent with no AI allowed. Senior lawyers who had used the tool beat their peers clearly, so three months of working alongside it had sharpened their judgment. Junior lawyers, on average, gained nothing. Their scores split instead: more good results, but also more very poor ones.

The lesson for someone early in a career is specific. AI can make your output look like a more experienced person's before you have that person's judgment, and judgment is the thing employers pay seniors for. Using the tool is not the risk; letting it do the part of the work that would have taught you is. Check its output against your own reasoning, and ask why it made each change rather than only accepting it.

The Automation Risk Spectrum

Not all roles face equal risk. Think of it as a spectrum:

High automation risk — roles where AI can perform the majority of the core tasks with high accuracy:

• Data entry and data processing roles
• Basic customer service (FAQ-type queries, ticket routing, form processing)
• Standard legal document review (contract redlining, discovery document review)
• Medical transcription and clinical documentation
• Basic financial analysis (variance reports, standard ratio analysis)
• SEO content writing targeting keyword density rather than insight

Being transformed or augmented — roles where AI handles a significant portion of the work, but human judgment, relationship, and creativity remain central:

• Software engineering (AI writes code; humans design systems, make architectural decisions, debug complex issues)
• Medicine (AI assists with diagnosis, documentation, triage; physicians maintain clinical judgment and patient relationships)
• Law (AI handles research and drafting; lawyers exercise judgment, strategy, and advocacy)
• Marketing (AI generates content at scale; marketers set strategy, manage brand voice, build relationships)
• Education (AI personalizes learning, handles grading; teachers facilitate human development)

Growing because of AI — roles that did not exist a few years ago or are dramatically expanding:

• AI Engineer (building and deploying AI systems in production)
• AI Agent Developer (building multi-step autonomous agent systems — one of the hottest roles in 2025-2026)
• AI Product Manager (bridging technical AI capabilities and business needs)
• AI Safety / Evaluation Specialist (alignment, red-teaming, benchmarking — driven by EU AI Act and corporate risk management)
• Chief AI Officer (CAIO) — a C-suite role adopted by 20-25% of Fortune 500 companies by early 2025, up from under 5% in 2022
• AI Integration Specialist (deploying AI tools into existing enterprise workflows)
• LLM Ops / MLOps Engineer (deployment, monitoring, cost optimization for AI systems)
• AI Trainer / RLHF Specialist (providing expert feedback to improve model outputs — Scale AI, Outlier, and internal AI labs are major employers)
• Prompt Engineer (evolving from a standalone role into an integrated skill expected of many technical positions; specialized prompt engineering for enterprise deployments and red-teaming remains in demand)

Resilient Role Characteristics

Some characteristics make a role more resilient to automation. If your role has several of these, you are in a stronger position:

Human judgment in high-stakes decisions: Courts, medical diagnoses, hiring decisions, investment calls — society is still not comfortable with AI making these decisions without human accountability. This creates durable demand for humans.

Physical presence in unpredictable environments: Plumbers, electricians, nurses, surgeons, childcare workers — physical manipulation in variable environments remains difficult to automate.

Emotional labor and relationship-building: Therapists, coaches, teachers, salespeople with deep client relationships, social workers — the human connection itself is often the service.

Creativity and original conceptual work: Not all creative work is equally safe, but original conceptual thinking, aesthetic judgment, and brand identity work at the highest level remains human-driven.

Interdisciplinary synthesis across contexts: Connecting insights from law + technology + human behavior + organizational dynamics — this kind of broad contextual integration remains a human strength.

The Likely Outcome: Augmented Professionals

The realistic near-term outcome for most knowledge workers is not replacement — it is a new kind of role where being effective requires AI fluency.

The lawyer who uses Harvey AI to do research in 20 minutes instead of 4 hours does not lose their job. They become a more productive lawyer who can serve more clients, work on more complex matters, and bill for judgment rather than research time. Harvey AI now serves roughly 100,000 lawyers across 50 of the top 100 Am Law firms — and the lawyers who refuse to engage with these tools increasingly look uncompetitive against those who do.

This creates a two-tier dynamic: AI-fluent professionals and AI-avoidant professionals — and the gap between them will widen over time.

Key Takeaways

• AI is transforming knowledge work now — the shift is already underway, not hypothetical
• AI automation targets pattern-based, well-defined, high-volume, language-based tasks first
• Most near-term change will be augmentation (AI + human) rather than wholesale replacement
• Roles with human judgment, physical presence, emotional labor, and genuine creativity are more resilient
• The emerging career risk is not "AI takes my job" but "a human using AI takes my job"
• In a three-month trial with patent lawyers, AI improved everyone's work but only built judgment in senior lawyers, so early-career workers need to keep doing the reasoning AI could do for them
• Anthropic's own October 2026 workforce study showed the same entry-level squeeze from inside one company: junior technical hires fell from 27 percent to 9 percent as its AI agents took on more of the work