What to do when AI changes or replaces your role

If you are reading this, you have probably felt it. A low hum of professional unease that sharpens every time a colleague mentions a new AI tool, every time a LinkedIn post announces another wave of "efficiency gains," every time your industry newsletter uses the word transformation. You may be watching parts of your role quietly migrate to software. You may have already been told your position is "evolving." Or you may simply be a thoughtful person who reads the news and wonders what comes next.

This article is for you. It is long, and deliberately so, because the question of what to do when AI reshapes your working life deserves more than a listicle. We have synthesised findings from labour economics, organisational psychology, career science, and real-world case studies, drawing on research from institutions including McKinsey Global Institute, the World Economic Forum, the IMF, Goldman Sachs, MIT, Harvard Business School, Brookings, and the OECD to give you an honest, substantiated picture of where things stand.

The core message, which we will substantiate throughout: AI is reshaping white-collar work faster than most expected, but the evidence points to transformation far more often than elimination. Professionals who understand this distinction, attend to the psychological reality of the transition, and take deliberate exploratory action are far better positioned than those who either panic or freeze.

This is not a moment for panic. It is a moment for informed, deliberate strategy.

The gap between predicted disruption and actual job loss

The headlines about AI eliminating millions of jobs rest on real research, but they routinely conflate theoretical automation potential with imminent job loss. The distance between those two things is wide, and understanding it is the first step toward thinking clearly about your own situation.

The World Economic Forum's 2025 Future of Jobs Report, surveying over 1,000 global employers, projects that 92 million jobs will be displaced by 2030, but that 170 million new roles will emerge, yielding a net gain of 78 million positions. The IMF estimates that roughly 60% of jobs in advanced economies are exposed to AI, yet most of that exposure means augmentation: tasks changing, not entire roles vanishing.

McKinsey Global Institute's November 2025 report found that current technologies could theoretically automate activities accounting for 57% of US work hours, but explicitly cautioned this reflects technical potential for change, not a forecast of job losses. Goldman Sachs research from August 2025 offered a more grounded figure: if every current AI use case were deployed economy-wide, only 2.5% of US employment faces immediate displacement risk. Even under the most aggressive adoption scenarios, Goldman's economists note that historically, technology-driven unemployment disappears after approximately two years.

The critical distinction, one that most commentary fails to make, is between roles eliminated entirely and roles transformed. The OECD surveyed firms across seven countries that had already adopted AI and found that 77% reported no impact on the quantity of jobs for affected workers. Half the firms implemented AI to boost production quality rather than reduce headcount, and displaced workers were typically reassigned to different tasks rather than dismissed. A Brookings Institution and Yale Budget Lab analysis from October 2025 examined 33 months of labour market data since ChatGPT's launch and concluded that despite widespread fears, the overall labour market shows more continuity than immediate collapse.

That said, the changes are real and concentrated. Goldman Sachs found that 46% of office and administrative support tasks and 44% of legal tasks are automatable with current technology. The WEF identified postal clerks, bank tellers, data entry clerks, cashiers, administrative assistants, and accountants as the fastest-declining roles. MIT economist Daron Acemoglu, who won the 2024 Nobel Prize in Economics, offered the most conservative estimate: only 4.6% of tasks will be profitably replaced by AI in the near term, yielding modest productivity gains of roughly 0.7% over a decade.

The pace follows a wave pattern rather than a single shock. Administrative and data processing roles face the most immediate pressure, with routine knowledge work and content creation affected next, and complex professional work (aspects of legal analysis, financial planning, medical diagnostics) shifting in subsequent years. Critically, only 9.3% of US companies reported using generative AI in production as of mid-2025. The full transformation will be measured in years and decades, not months.

But for professionals in exposed roles, the individual timeline can feel much shorter than the macroeconomic one. That matters, and we should be honest about it.

Why this feels like grief, and why that reaction is not weakness

The anxiety many professionals feel about AI is not irrational. Research in organisational psychology has established that disruption to one's professional role triggers real psychological distress that follows recognisable patterns. Understanding these patterns does not make them disappear, but it does make them navigable.

The concept of a "career shock," coined by Jos Akkermans and colleagues at Vrije Universiteit Amsterdam in 2018, describes a disruptive event at least partly caused by factors outside the individual's control that triggers a deliberate thought process concerning one's career. AI-driven role change fits this definition precisely. It is not something you chose, it is not something you caused, and yet it demands a fundamental reassessment of professional direction.

The emotional response goes deeper than worry about income. Researchers at the University of Paderborn found that AI introduction in the workplace creates what they term AI identity threat, a challenge to one's sense of professional self, not just economic security. Their study identified three central drivers: changes to the nature of work, perceived loss of status, and negative identification with AI systems. A study of physicians published in the Journal of Medical Internet Research found that a third of negative responses to AI centred on threats to professional recognition; doctors feared becoming a conduit for algorithmic decisions rather than exercising the judgment they had spent decades developing. Software engineers showed similar patterns, with some refusing to adopt generative AI tools not out of ignorance but to preserve their sense of competence and autonomy.

A 2025 qualitative study by Sharma and colleagues at Symbiosis International University interviewed 24 Indian IT professionals who had experienced AI-induced job loss or reassignment. They identified six psychological themes: emotional shock, erosion of professional identity, chronic anxiety and anticipatory rumination, social withdrawal driven by shame rather than anger, oscillation between adaptive and maladaptive coping, and a pervasive sense of organisational betrayal. The findings will resonate with anyone who has watched their role quietly shrink. The distress is about meaning and belonging, not just money.

The broader evidence on job displacement and mental health confirms these reactions carry real consequences. Administrative data from Taiwan tracking displaced workers over a decade found 67–68% earnings loss in the year following displacement, with earnings failing to return to pre-displacement levels even after ten years. Mental health outpatient visits increased 15–16%, and associated medical costs rose substantially. An Australian cohort study demonstrated a clear dose-response gradient: 69% of people who lost jobs reported poor mental health versus 59.5% of those with reduced hours and 24.2% of unaffected workers. The gradient matters for the AI context. It suggests that gradual role transformation, while less psychologically damaging than sudden job loss, still carries meaningful distress.

Two frameworks from the research literature are worth knowing because they provide real orientation during a disorienting time.

William Bridges' transition model, developed in 1991, distinguishes between change (the external event) and transition (the internal psychological reorientation). He identified three stages: endings, where you acknowledge what is being lost; the neutral zone, an uncomfortable in-between period of confusion that is also the seedbed for new beginnings; and new beginnings, where a reconfigured professional identity takes shape. For professionals whose roles are being gradually reshaped by AI, the neutral zone concept is particularly apt. It describes that prolonged, disorienting period where you are no longer fully in your old role but not yet established in a new one. Bridges' key insight is that transition starts with an ending, and attempting to skip the emotional processing of that ending undermines the entire journey forward.

Mark Savickas's career adaptability framework, validated across 13 countries, provides a more action-oriented toolkit. His four dimensions, concern (preparing for what lies ahead), control (taking ownership of career decisions), curiosity (exploring possible selves), and confidence (believing in one's capacity to act), function as psychological resources that can be deliberately cultivated. Research published in Scientific Reports in 2026 found that AI anxiety directly undermined career adaptability, with this mediating effect accounting for 63% of the total negative impact on career decision-making. The implication: building adaptability resources is a psychological necessity, not an optional extra.

What history teaches us, and where this time is genuinely different

Every major technological disruption has provoked the same fear: this time, the machines will finally take all the jobs. And every time, the economy has eventually generated new work at scale. These historical parallels offer real comfort. They also have real limits.

US farm workers comprised 41% of the labour force in 1900 and just 2% by 2000, yet total employment grew massively. The famous ATM–bank teller paradox, documented by Boston University's James Bessen, showed that despite ATMs reducing the average number of tellers per branch from 21 to 13, cheaper branch operations enabled banks to open 43% more locations, and total teller employment actually increased from 485,000 in 1985 to 527,000 in 2002, while ATM numbers rose from 60,000 to 352,000.

The computerisation of office work in the 1980s and 1990s followed a well-documented pattern: computers automated routine cognitive tasks (bookkeeping, data entry, clerical processing), producing "job polarisation," with growth in both high-wage cognitive work and low-wage service work, and the middle hollowed out. Personal secretary roles declined sharply, but new categories of work emerged in IT, management consulting, and knowledge work. Each of these transitions unfolded over decades, not months. The Yale Budget Lab's 2025 assessment noted that historically, widespread technological disruption in workplaces tends to occur over decades rather than months or years.

Goldman Sachs economists Joseph Briggs and Sarah Dong observe that predictions of technology eliminating human labour have a long history but a poor track record, noting that 60% of workers in 2022 held occupations that did not exist in 1940, implying 85% of employment growth over eight decades came from technology-driven job creation.

But the AI parallel genuinely breaks down in important ways, and we should be honest about that.

David Autor's 2024 NBER analysis identifies the core difference: previous computerisation automated routine tasks that followed explicit, codifiable rules. AI breaks through what philosopher Michael Polanyi called the paradox of tacit knowledge. It can now perform tasks that humans understand intuitively but cannot express as rules: legal writing, medical diagnosis, software coding, creative work. Previous waves of automation primarily displaced blue-collar and mid-skill workers. AI disproportionately targets white-collar, educated, higher-income knowledge workers. Brookings research found that workers with graduate degrees face four times the AI exposure of those with only high school education.

Daron Acemoglu, in Power and Progress (co-authored with Simon Johnson), makes a harder argument: technology's benefits are not automatically broadly shared. During the British Industrial Revolution, he notes, wages fell significantly and it took roughly a century, more than three generations, for gains to be widely distributed. He warns that AI is currently being developed primarily as what he calls anti-worker technology, focused on automating tasks rather than creating new ones for workers. Carl Benedikt Frey, whose widely cited 2013 Oxford paper estimated 47% of US jobs at risk, acknowledged in a 2024 reappraisal that generative AI confounded expectations: the first casualties were artists and writers, not factory workers. Yet Frey also found something unexpected: AI disproportionately benefits lower-skilled workers by lowering barriers to expertise, much as GPS navigation enables anyone to navigate like an experienced taxi driver.

The honest assessment is that historical patterns offer real comfort and real grounds for concern. Mid-career professionals should plan for both possibilities, and the practical frameworks for doing so are, fortunately, well-researched.

Practical strategies for navigating what comes next

The research on effective career transitions converges on several evidence-based principles that are more nuanced, and more useful, than the generic advice to "learn to code" or "embrace change."

Start with identity, not skills

This runs counter to what most career guidance suggests, but it may be the most important point in this article.

Herminia Ibarra, professor at London Business School and author of Working Identity, argues that conventional "plan-then-implement" career change advice has the process backwards. Her research, spanning two decades, shows that successful career reinvention operates through three mechanisms: experimenting with new professional activities (what she calls "identity experiments"), interacting in new networks of people, and making sense of what is happening in light of emerging possibilities. Ibarra's central insight is that changing careers means changing ourselves, and this change cannot be analytically planned in advance. It must be tested through small, exploratory actions. The transition process is rarely linear; people oscillate between clinging to their old professional identity and reaching toward a new one because they have lost the narrative thread of their professional life.

What this means: before updating your CV or enrolling in a course, spend time with the more uncomfortable question of who you are becoming professionally. Talk to people in adjacent roles. Volunteer for projects at the edge of your current expertise. Take on a small freelance engagement in a field that interests you. These are not distractions from the "real" work of career transition. According to Ibarra's research, they are the real work.

Map your skills, not your job title

Research on skill transferability demonstrates that the specific combination of skills a professional possesses matters far more than their job title in determining mobility options. The Federal Reserve Banks of Cleveland and Philadelphia built the Occupational Mobility Explorer mapping 2,300+ skills across 600+ occupations, finding that communication, management, writing, planning, and analytical skills transfer broadly across industries and roles.

Research by Geel and Backes-Gellner applying Lazear's skill-weights framework found that mobility within a skill cluster yields an average 6.8% wage increase, while mobility between clusters produces wage losses. The takeaway: career moves that build on existing skill foundations succeed more reliably than dramatic reinventions into unrelated fields. A marketing director who moves into customer experience strategy is leveraging the same skill cluster. A marketing director who retrains as a data engineer is starting from scratch, and the data suggests this is a riskier path than it appears in the bootcamp advertisements.

To map your own transferable skills, consider not what your job title says but what you actually do each day: the decisions you make, the relationships you manage, the problems you solve, the judgment calls you navigate. These capabilities are far more portable than most people realise.

Understand what actually works in reskilling, and what does not

The evidence on retraining effectiveness is mixed but instructive, and it is worth knowing before investing significant time or money.

The OECD's meta-analysis of 207 studies found that intensive one-on-one job counselling is the most cost-effective intervention for displaced workers, while training programmes show modest short-term effects but stronger medium-term benefits over two to three years. The University of Chicago's Becker Friedman Institute found that effective reskilling generates $3.20 in additional returns for every dollar invested beyond direct labour market gains, and prevents one case of depression for every three participants. However, Brookings researchers identified a critical failure pattern: programmes frequently retrain workers from one automation-susceptible occupation to another, moving people laterally across a shrinking landscape rather than forward.

Online learning platforms face a sobering reality. MOOC completion rates average just 12.6%, dropping to 3% for free courses. Paid, certificate-bearing programmes with instructor engagement perform much better, with completion rates reaching 46% among paying students. The lesson: free and frictionless does not equal effective. Structure, accountability, and some financial commitment consistently predict better outcomes.

Coding bootcamps show surprisingly competitive results: 72% of graduates find relevant employment within six months, with average salary increases exceeding $25,000. But these outcomes are most relevant for people whose existing skills and interests align with technical work. They are not a universal prescription.

Know the realistic timelines

Career pivot data suggests realistic timelines for mid-career professionals. Most successful changers land new roles within six to twelve months, though industry transitions typically take six to twelve months longer. Harvard Business School research found that successful mid-life transitioners spend three to six months in an active exploration phase before making concrete moves. The average age of career change is 39, and 73% of professionals who change careers at 40 or older report higher job satisfaction within two years. Only 17% of career changers regret the move, and the most common regret is not making the change sooner.

Invest in your professional network strategically

Professional networks remain the single most powerful career transition mechanism. Research consistently finds that 70–85% of jobs are never publicly posted, and referred candidates are significantly more likely to be hired than those applying through job boards. Ibarra identifies building new networks as one of three essential strategies for career reinvention. Not networking in the transactional, card-collecting sense, but deliberately cultivating relationships in the professional communities you want to enter.

The most effective approach is to contribute before you need anything. Share what you are learning. Offer perspective from your existing domain expertise. Ask genuine questions. The professionals who navigate career transitions most successfully tend to build bridges before they need to cross them.

What's becoming more valuable

Understanding what is becoming more valuable matters as much as tracking what is declining. The evidence here is more encouraging than the headlines suggest.

Autodesk's 2025 analysis of three million job listings found AI Engineer postings growing 143%, AI Content Creator up 135%, and AI Solutions Architect up 109% year-over-year. But the biggest surprise was that design skills have surpassed coding as the top in-demand skill in AI-specific job listings, a signal that the ability to shape how humans interact with AI systems matters as much as building those systems. PwC's analysis of roughly one billion job postings found that AI-skilled workers command a 56% wage premium, up from 25% the prior year, one of the fastest-growing skill premiums in recent labour market data. And 51% of job postings requiring AI skills are now outside IT and computer science, spanning healthcare, finance, legal, marketing, and education.

The World Economic Forum's 2025 skills forecast places analytical thinking, creative thinking, resilience, curiosity, and leadership alongside AI literacy in its top ten most important skills. Skills in AI-exposed jobs are changing 66% faster than in less-exposed roles, creating both urgency and opportunity.

The clearest framework for understanding which human capabilities retain their value comes from MIT Sloan researchers Isabella Loaiza and Roberto Rigobon, who analysed approximately 19,000 work tasks across 950 occupations to develop the EPOCH framework: Empathy and emotional intelligence, Presence and connectedness, Opinion and ethical judgment, Creativity and imagination, and Hope, vision, and leadership. Their analysis found that tasks scoring high on EPOCH dimensions are less susceptible to automation but strong candidates for augmentation, and that newly added tasks in the O*NET database in 2024 have higher EPOCH levels than existing tasks. The economy, in other words, is already shifting toward more human-intensive work. Rigobon's framing is worth sitting with: "We deliberately don't call these 'soft' skills. A 'hard' skill, like solving a math problem, is comparatively easy to teach. It is much harder to teach hope, empathy, and creativity."

How the best human-AI collaboration actually works

The most important empirical study on human-AI collaboration, conducted by researchers from Harvard Business School and Wharton with 758 BCG consultants using GPT-4, revealed what its authors call the "jagged technological frontier." The frontier is not a clean line between what AI can and cannot do; it is irregular, unpredictable, and different for every task. For tasks within AI's capabilities, consultants using AI completed 12% more tasks, 25% faster, at 40% higher quality. The lowest-performing consultants improved by 43%. But for tasks outside AI's frontier, consultants with AI performed 19 percentage points worse than those without it, victims of miscalibrated trust in AI's capabilities.

Separately, Stanford and MIT researchers studying 5,179 customer support agents found that AI tools boosted average productivity by 14–15%, with novice workers gaining 34% while experienced workers saw minimal change. The AI essentially disseminated the best practices of top performers to newer colleagues, compressing the learning curve from six months to two.

These findings give rise to what Wharton's Ethan Mollick terms the "centaur" and "cyborg" models of human-AI collaboration. Centaurs maintain a clear division of labour (the human decides the strategy, the AI executes defined tasks) like the distinct halves of the mythical creature. Cyborgs integrate human and AI effort seamlessly, with constant back-and-forth collaboration. Both models outperform either humans or AI working alone. The origin story is telling: after Garry Kasparov lost to IBM's Deep Blue in 1997, he discovered in subsequent "advanced chess" tournaments that a weak human plus a machine plus a better process was superior to a strong computer alone, and, more remarkably, superior to a strong human plus a machine with an inferior process. The quality of the collaboration design matters as much as the quality of either participant.

David Autor's 2024 NBER paper frames this well. Rather than viewing AI as an automation tool that eliminates expertise, he argues it can work as a force multiplier for expertise, enabling a larger set of workers with complementary knowledge to perform high-stakes decision-making previously reserved for elite experts. If developed and deployed deliberately, AI could help rebuild the middle-skill, middle-class heart of the labour market that decades of automation and globalisation have hollowed out. Whether that potential is realised depends not on the technology itself but on how organisations and societies choose to deploy it.

The practical takeaway for individual professionals: learn to recognise the jagged frontier in your own domain. Understand where AI is reliably excellent, where it is confidently wrong, and where your human judgment creates irreplaceable value. That discernment is the skill.

Organisations that get it right redeploy rather than discard

The corporate case studies of successful AI workforce transitions share a pattern: invest in reskilling before displacement occurs, build on employees' existing competencies, and treat the transition as a strategic investment rather than a cost-cutting exercise. These are not utopian outliers. They are evidence of what is possible when organisations take the long view.

IKEA provides perhaps the clearest example. When the company deployed its AI chatbot "Billie" to handle 47% of customer service inquiries, it chose to reskill rather than dismiss. 8,500 call centre workers were retrained as interior design advisors, remote sales specialists, and complex customer service agents. The results were substantial: IKEA generated €1.3 billion in revenue through remote customer meeting points by the end of the fiscal year, representing 3.3% of total sales. The company plans to train 70,000 employees in AI literacy by 2026.

AT&T's $1 billion "Future Ready" programme reskilled 180,000 of its 203,000 employees when it discovered that nearly half lacked the STEM skills its evolving business required and 100,000 hardware jobs would become obsolete. Over 2.7 million online courses were completed, and 40–50% of internal job openings were subsequently filled by existing employees.

In professional services, the pattern holds. Morgan Stanley deployed a GPT-4-powered assistant across its 16,000 financial advisors, achieving 98% adoption. Document retrieval efficiency jumped from 20% to 80%. The role of financial advisor did not shrink; it shifted toward client relationships and personalised advice, with administrative burden sharply reduced. A&O Shearman, the first "magic circle" law firm to deploy generative AI, saw one in four lawyers using the Harvey AI system daily during its beta trial of 40,000 queries. Junior attorneys' work shifted from manual contract review toward strategic judgment, validation, and oversight. The head of the firm's innovation group offered an important caveat that applies to every human-AI workflow: "You must validate everything coming out of the system."

The Associated Press uses AI to automate corporate earnings stories, freeing reporters for investigative work. Bloomberg built its own large language model for financial document analysis. In each case, the pattern is consistent: AI absorbs the routine, and the human role migrates toward judgment, context, and relationship management.

The lesson for individual professionals is simple: if your organisation is investing in reskilling, engage fully. These programmes represent significant investment in your continued relevance. If your organisation is not, take that as a signal to invest in yourself.

The policy picture: what support exists and what does not

On the policy front, responses vary widely, and professionals should understand what resources are available to them.

The UK's AI Skills Boost programme, announced in mid-2025 and expanded in January 2026, targets 10 million workers by 2030 (a third of the UK workforce) with free AI training through a national AI Skills Hub, backed by partnerships with Accenture, Amazon, Google, IBM, Microsoft, and the NHS. Singapore's SkillsFuture programme provides every citizen over 25 with credits for approved training, with enhanced support for workers aged 40 and above including a $300 monthly training allowance. Denmark's flexicurity model combines easy hiring and firing with generous unemployment support and mandatory retraining, achieving low unemployment alongside strong worker protection.

The EU AI Act, which took effect in August 2024, requires AI literacy for all staff deploying AI systems but contains no specific provisions on job displacement or worker reskilling. The US picture is fragmented; the current administration's approach favours deregulation without workforce-specific measures. The Bipartisan Policy Center has proposed an "AI Adjustment Assistance" programme, but no such programme currently exists at the federal level.

The evidence on what types of support actually help most is clear in broad strokes. Intensive, personalised job counselling consistently outperforms generic programmes. Training that connects directly to real job vacancies works; training disconnected from employer needs often does not. Income support during transitions cushions wage losses and encourages acceptance of suitable new roles. And yet OECD countries have been decreasing investment in training, with average spending falling to just 0.1% of GDP in 2023 (a 30% decline from the 2010 peak) even as the need accelerates. This means that for many professionals, the safety net is thinner than they might assume, and proactive self-investment is correspondingly more important.

Bringing it together: principles for deliberate action

The evidence assembled across hundreds of studies, institutional reports, and real-world case studies points toward several conclusions that mid-career professionals can act on with confidence.

The transformation is real but slower than the headlines suggest. Thirty-three months after ChatGPT's launch, the broader labour market shows more stability than disruption, and most economists expect the full transition to unfold over years and decades. But specific roles and industries are already changing materially, and waiting for certainty is itself a strategy, a passive one with compounding costs.

Most roles will be reshaped, not replaced. The consensus across every major research institution is that the vast majority of white-collar jobs will see 25–50% of tasks automated, with the role evolving rather than vanishing. The professionals who thrive will be those who understand the jagged frontier of AI capabilities and position themselves on the human side of that boundary: judgment, ethics, emotional intelligence, strategic thinking, and the ability to know when AI is confidently wrong.

The psychological dimension deserves as much attention as the practical one. Career shock, identity threat, and anticipatory anxiety are documented, normal responses. Acknowledging the emotional reality of professional disruption, rather than suppressing it with toxic positivity, is supported by every strand of the research. Bridges' neutral zone, Savickas's adaptability resources, Ibarra's experimental approach to identity: these are frameworks grounded in decades of empirical work, not self-help platitudes.

The most effective action is exploratory, not dramatic. The evidence favours small experiments over grand reinventions: testing adjacent roles, building new networks, acquiring targeted skills that compound existing expertise rather than starting over. Career pivots succeed most reliably when they build on transferable skill clusters rather than abandoning them. The professionals who navigated previous technological disruptions successfully, from bank tellers becoming relationship bankers to journalists becoming data-augmented investigators, did not become entirely different people. They carried forward what made them valuable and applied it in reshaped contexts.

Invest in the right things, in the right order. Attend to the psychological transition first. Acknowledge what is changing and give yourself permission to feel uncertain. Then map your transferable skills honestly. Begin small experiments: conversations with people in adjacent roles, pilot projects that stretch your capabilities, structured learning that builds on your existing expertise. Cultivate your professional network deliberately, contributing before you need to ask for anything. And develop AI fluency, not to become an engineer, but to understand the jagged frontier in your own domain well enough to know where your human judgment is irreplaceable.

This is not a crisis to survive. It is a transition to navigate, and the evidence says you can.

Key research cited in this article

For readers who wish to explore the evidence further, these are among the most important sources referenced above:

On the scale and pace of AI workforce impact: WEF Future of Jobs Report 2025 · McKinsey, "Agents, Robots, and Us" (2025) · Goldman Sachs, "How Will AI Affect the Global Workforce?" · Brookings, "No AI Jobs Apocalypse — For Now" (2025) · Yale Budget Lab, "Evaluating the Impact of AI on the Labor Market" (2025)

On the economics of AI and work: Autor, "Applying AI to Rebuild Middle Class Jobs" (NBER, 2024) · Acemoglu, "What Do We Know About the Economics of AI?" (MIT) · Frey, "Twelve Years After 'The Future of Employment'" (2024)

On human-AI collaboration: Dell'Acqua et al., "Navigating the Jagged Technological Frontier" (HBS, 2023) · Brynjolfsson, Li, and Raymond, "Generative AI at Work" (QJE, 2025) · MIT Sloan, EPOCH Framework

On the psychology of career transition: Ibarra, Working Identity (London Business School) · Bridges, Managing Transitions · Savickas, Career Adaptability Framework · Sharma et al., "Psychological Impacts of AI-Induced Job Displacement" (2025)

On reskilling and career transition effectiveness: Brookings, "AI Labor Displacement and the Limits of Worker Retraining" · Becker Friedman Institute, "Reskilling and Resilience" · Cleveland Fed, Occupational Mobility Explorer