Back to Blog

Every generation of workers has faced a technological disruption they were told would end their profession. The printing press was supposed to eliminate scribes. Calculators were going to make accountants obsolete. The internet would kill retail. The honest historical pattern is more complicated than either the panic or the dismissal suggests: technology tends to eliminate specific tasks within jobs rather than entire occupations, while simultaneously creating new kinds of work that didn't previously exist. AI is following this pattern — but the pace and breadth of task-level disruption is genuinely different in scale from previous waves.

The question "will AI take my job" is almost always the wrong question. The right question is: which of the tasks that make up my job can AI do, and what does that mean for the role I'll be doing in five years? This article answers that question industry by industry, drawing on the best research available — and then gives you a practical framework to assess your own position.

300M
full-time-job equivalents of work
exposed to AI automation globally
Goldman Sachs, 2023 [1]
80%
of US workers have at least
10% of tasks affected by LLMs
OpenAI / U. Penn., 2023 [2]
+78M
net new roles projected by 2030
— 170M created, 92M displaced
World Economic Forum, 2025 [3]
7%
of US workers are in jobs where
≥50% of tasks are exposed and substitutable
Goldman Sachs, 2023 [1]
The Essential Framing: Tasks, Not Jobs

A landmark 2023 study from OpenAI and the University of Pennsylvania analyzed 1,016 occupations against GPT-4's capabilities. The researchers found that roughly 80% of the US workforce has at least 10% of their work tasks exposed to LLMs, and about 19% have more than 50% of tasks exposed. Critically, they framed this as exposure, not replacement — tasks that AI can perform alongside a human, making that human faster and more productive.

The same research found that higher-wage, higher-education roles are disproportionately exposed — the reverse of what most automation waves have done. A truck driver is less exposed to GPT-4 than a paralegal. A plumber is less exposed than a financial analyst. This inversion is one of the most important and underreported findings in AI labor research.

Source: Eloundou, Manning, Mishkin & Rock — "GPTs are GPTs" (arXiv:2303.10130, 2023) up

The Industry-by-Industry Breakdown

🗂️
Administrative & Clerical Work
Data entry · scheduling · document processing · basic correspondence · filing
HIGH RISK

This is the category with the most legitimate automation pressure in the near term. The defining characteristic of clerical work is that it involves processing, organizing, and transforming information according to rules — exactly the task profile that large language models excel at. Routine data entry, standard document processing, templated correspondence, and calendar management are already being automated or significantly accelerated by AI tools embedded directly into business software (Microsoft 365 Copilot, Salesforce Einstein, Google Workspace AI).

⚠ High Displacement Risk
Data entry operators
Filing and records clerks
Basic scheduling coordinators
Templated correspondence roles
Form-processing clerks
→ Lower Risk / Evolving
Executive assistants (judgment-heavy)
Office managers (physical coordination)
Administrative directors (strategic)
HR admins who handle sensitive issues
Roles requiring institutional knowledge
Key Insight:The evidence here is weaker than the risk rating suggests, and it is worth being honest about that. In its current 2024–34 projections the US Bureau of Labor Statistics expects little or no change in employment for secretaries and administrative assistants, with the wider office and administrative support group declining 3.9%. [10] That is a slow erosion that predates generative AI, not a collapse caused by it. BLS has not published a projection attributing these declines to AI. The roles that persist will require judgment, relationship management, and cross-departmental coordination — tasks AI still handles poorly — but readers in these jobs should weigh the official projection, which is milder than most coverage implies.
🎧
Customer Service & Support
Tier-1 support · FAQ handling · complaint routing · live chat · call centers
HIGH RISK

Tier-1 customer service — the layer that handles routine, repetitive, policy-based questions — is already being automated at scale. Every major customer service platform (Intercom, Zendesk, Salesforce Service Cloud) now embeds AI chatbots as the first contact point. Companies openly report deflection rates of 40–70% on common queries. The jobs that are disappearing are the roles that answered the same 50 questions a thousand times per shift.

What remains — and what is growing — is complex escalation handling, relationship management with high-value accounts, and emotionally sensitive support (medical, financial, bereavement contexts). These tasks require human empathy, real-time situational judgment, and the ability to operate outside of scripted responses.

⚠ High Displacement Risk
Tier-1 call center agents
Live chat support (routine queries)
Technical support (basic troubleshooting)
Complaint routing roles
→ Lower Risk / Evolving
Escalation specialists
Key account relationship managers
Sensitive / complex case handlers
Customer experience strategists
Key Insight:If your job primarily consists of responding to predictable questions using reference documentation, AI is already doing substantial portions of your role in the organisations ahead of yours. The transition is underway, not pending.
📈
Finance & Accounting
Bookkeeping · financial analysis · tax preparation · investment research · advisory
SPLIT

Basic bookkeeping, routine tax preparation, and standardized financial reporting have been automating for over a decade — AI is accelerating a trend already well underway. Intuit's AI suite, QuickBooks AI, and AI-native tools like Ramp and Brex have made much of routine bookkeeping nearly touchless. The OpenAI/Penn study found financial quantitative analysts among the roles with the highest LLM exposure of any occupation.

Financial advising, complex tax strategy, investment banking deal work, and CFO-level judgment are categorically different. They depend on client relationships, risk appetite calibration, regulatory navigation, and the kind of institutional trust that takes years to build and cannot be delegated to software.

⚠ High Displacement Risk
Basic bookkeepers
Routine tax preparers
Junior financial analysts (report generation)
Accounts payable / receivable clerks
↑ Augmented / Growing
Financial advisors (AI-assisted)
Tax strategists (complex planning)
Investment analysts (original thesis)
CFOs and finance directors
Risk management specialists
Key Insight:The AICPA has acknowledged that AI will transform accounting — but toward advisory services rather than elimination. CPAs who pivot toward strategic tax planning and financial advisory will find AI makes them more productive, not redundant.
🏥
Healthcare
Diagnostics · clinical notes · imaging · nursing · therapy · surgery · patient care
AUGMENTED

Healthcare is the most complex AI labor story of any sector. AI genuinely outperforms radiologists on specific diagnostic image classification tasks in controlled settings — Google's DeepMind has demonstrated ophthalmology AI with expert-level accuracy, and FDA-cleared AI tools are now reading chest X-rays in real clinical environments. At the same time, the physical, emotional, and relational core of most clinical work is deeply resistant to automation.

The net effect is not replacement but significant augmentation: clinicians who adopt AI diagnostic support tools will be more accurate and can see more patients. The roles most at risk are purely administrative clinical roles — medical coding, prior authorization processing, clinical note transcription. The roles most resilient are those with the highest human-contact and physical-judgment components: nursing, emergency medicine, surgery, mental health.

⚠ Automation Pressure
Medical coders and billers
Prior authorization specialists
Clinical note transcriptionists
Routine radiology reads (narrow tasks)
→ Highly Resilient
Nurses and care workers
Surgeons and proceduralists
Mental health clinicians
Emergency physicians
Primary care physicians (relationship)
Key Insight:The WHO projects a global shortage of 11 million healthcare workers by 2030 — revised up from 10.2 million as global progress slowed. AI in healthcare is far more likely to address that shortage through augmentation than to create unemployment within it. The clinical workforce challenge is supply, not displacement.
🎓
Education
Teaching · curriculum development · tutoring · grading · research · administration
AUGMENTED

Evidence on AI tutoring is more mixed than vendor claims suggest. Carnegie Learning's MATHia has genuine randomised-trial evidence behind it — but it is rule-based intelligent tutoring software, not generative AI. For Khanmigo, the widely-quoted proficiency gains come from Khan Academy's own analysis of self-selected users, not a controlled trial; the peer-reviewed study that did randomise found no statistically significant difference in learning outcomes against a plain web-search control. [14] The tasks most exposed are content delivery and assessment of well-defined correctness (multiple choice, formulaic writing, math problem grading). But the tasks that constitute teaching as a human endeavor — motivation, mentorship, identifying struggling students before they disengage, navigating classroom dynamics — are almost entirely outside AI's current capability.

The realistic near-term trajectory is that AI handles the most mechanical aspects of teaching (differentiated worksheet generation, basic grading, administrative communications) while teachers focus more on the relational and adaptive elements of instruction that are genuinely irreplaceable. Curriculum designers and instructional technologists who can work with AI tools will be in increasing demand.

⚠ Automation Pressure
Rote content delivery roles
Basic grading and assessment
Administrative education coordinators
Online course content creators (commodity)
↑ Augmented / Resilient
Classroom teachers (K–12)
Special education instructors
University professors (research/mentorship)
Education technologists
School counselors
Key Insight:The US faces a significant teacher shortage in most states, particularly in STEM and special education. As in healthcare, the structural problem in education is undersupply, not oversupply — AI is unlikely to displace teachers at the K–12 level in the foreseeable future.
🎨
Creative Industries
Writing · design · illustration · music · photography · video · advertising · PR
SPLIT

Creative industries present the starkest split of any sector. Commodity creative work — stock photography, generic marketing copy, template-based design, product description writing, background music for commercial use — is already being displaced at significant scale by generative AI. The scale of that displacement is easy to overstate, and this article previously did. Getty Images — the obvious bellwether — has reported paid download volume falling only about 1.5% a year, and its 2025 revenue was the highest in the company's thirty-year history at $981.3 million, up 4.5%. [6] Whatever generative imagery is doing to commodity creative work, it is not yet visible as a collapse in the largest stock-photo business in the world.

Distinctive creative work — work with a recognizable voice, a coherent aesthetic vision, cultural insight, original conceptual thinking — is not only surviving but commands a meaningful premium as AI-generated content floods the market. The paradox is that generative AI simultaneously devalues commodity creativity and increases the value of genuine creative distinction.

⚠ High Displacement Risk
Stock / commodity illustrators
Generic content / SEO copywriters
Background music composers (library)
Template-based graphic designers
Product description writers
→ Resilient / Growing
Creative directors (vision and judgment)
Brand strategists
Distinctive-voice writers and journalists
Art directors overseeing AI workflows
Cultural / community storytellers
Key Insight:The question for creative professionals is not "can AI do what I do" but "is what I do distinguishable from what AI produces?" If the answer is no, the market will price your work accordingly. If yes — and if your audience can tell the difference — human creative work is entering a period of heightened value.
{ }
Software & Technology
Engineering · QA · DevOps · product · data science · IT support · architecture
AUGMENTED

Software development is one of the most AI-augmented professions and one of the most frequently mischaracterized in the "AI is coming for your job" media narrative. AI coding tools are usually said to make developers faster, and the number always quoted is the 55.8% speed-up Peng et al. measured on GitHub Copilot in early 2023. [8] That study is real, but it is narrower than its fame suggests: recruited freelancers, one artificial task, and a Codex-era product that is not the tool anyone uses now. The most careful follow-up points the other way. METR's 2025 randomised trial put experienced open-source developers on real tasks in repositories they already knew well, and they were 19% slower with AI assistance — while believing they had been about 20% faster. [13] The honest summary is that AI helps most on unfamiliar, well-specified, greenfield work, and can cost you time on a mature codebase you know intimately. It does not make experienced developers redundant.

The roles under genuine pressure are in quality assurance (automated test generation is maturing rapidly), IT helpdesk (tier-1 IT support is following the same trajectory as customer service), and junior developer roles — though the reason is not the one usually given. The evidence does not show AI widening the gap between senior and junior engineers. It shows the reverse: in the largest field study to date, the least experienced workers gained 34% while the most experienced gained almost nothing. [5] AI compresses the skill gap rather than widening it. The pressure on junior roles comes from somewhere else — if a tool lets a senior engineer absorb the routine work that used to be delegated, the junior post is cut even though the junior would have benefited most from the tool. That is a hiring decision, not a capability finding.

⚠ Automation Pressure
Junior / entry-level developers (boilerplate)
Manual QA testers
Tier-1 IT helpdesk
Basic data analysts (reporting only)
↑ Augmented / Growing
Senior and staff engineers
AI / ML engineers
Security engineers
Platform and infrastructure architects
Product managers with AI domain knowledge
Key Insight:The technology sector will not shrink due to AI — software is the medium through which AI is delivered, and demand for robust, secure, well-architected systems is expanding. The entry-level hiring pipeline will narrow; the demand for senior judgment will grow.
🔧
Skilled Trades & Physical Labour
Electricians · plumbers · HVAC · construction · mechanics · welders · carpenters
LOW RISK

This is the most straightforwardly good news in the entire AI labor landscape for workers in it. Skilled trades are structurally resistant to AI displacement for a simple reason: the work requires physical presence, dexterous manipulation in unpredictable environments, and the kind of real-world problem-solving that robotics still handles poorly and expensively. An AI can answer questions about electrical wiring — it cannot run new circuits through the walls of a 1950s house.

The US is in the middle of a skilled trades shortage that has been building for two decades, driven by demographic transition and under-investment in vocational training. That shortage will not be solved by AI. The trades are, for the foreseeable future, one of the safest career categories relative to automation risk — and among the most in-demand and well-compensated at the journeyman and master level.

→ Highly Resilient Roles
Licensed electricians
Plumbers and pipefitters
HVAC technicians
General contractors
Welders (specialized / structural)
→ Also Resilient
Mechanics and auto technicians
Elevator installers / inspectors
Sheet metal workers
Carpenters and finish workers
Millwrights and industrial technicians
Key Insight:The Oxford study that originally predicted 47% of US jobs were at high risk of automation (Frey & Osborne, 2013) rated electricians at a 15% automation probability — genuinely low, in the bottom third of its 702 occupations. Plumbers, pipefitters and steamfitters came in at 35%, which falls inside the paper's own medium-risk band and ranks 249th. Worth noting what that table also says: several desk professions scored lower than plumbers. The trades are not uniformly safe, and the safest jobs in that dataset were not all manual. [9]

The Five-Question Personal Risk Framework

Industry-level analysis is a starting point, not a verdict. The same industry can contain roles with radically different automation exposure. Here is a practical self-assessment framework — five questions that will help you locate your specific role on the risk landscape.

Assess Your Own Role
Answer these five questions honestly about your current job. More "safe" signals = lower automation exposure. More "risk" signals = examine your task composition carefully.
1
What percentage of your work follows a predictable, rule-based process?
Tasks that can be described as "if X, then Y" — following scripts, applying policies, filling templates, routing decisions according to predetermined criteria — are the tasks AI handles most readily. Estimate what fraction of your average week consists of this kind of work.
Safe: <20% rule-based Risk: >60% rule-based
2
Does your work require a physical presence or dexterous manipulation in a variable environment?
Robotics and physical AI remain expensive, limited, and unreliable outside of highly controlled factory settings. If your job requires you to be somewhere in person — responding to the specific conditions of a building, a patient, a customer, a scene — automation is structurally harder.
Safe: physical presence required Risk: fully desk-based and digital
3
Do people pay for your work because of your specific judgment, taste, or accumulated expertise?
If clients, employers, or customers would accept an equivalent output from anyone (or anything) else, that work is more substitutable. If they specifically want your perspective, your history, your relationships, or your aesthetic sensibility, that is harder to automate. The personal brand premium is a genuine structural buffer.
Safe: personal expertise valued Risk: output is fungible
4
Is human connection, trust, or emotional engagement a core part of what you deliver?
Therapy, negotiation, teaching, pastoral care, mentorship, and high-stakes sales all depend on a quality of human relationship that AI cannot replicate because the relationship itself is the product. People will accept an AI receptionist — they will not accept an AI therapist for their trauma, or an AI surgeon making intraoperative judgment calls.
Safe: relationship is the product Risk: relationship is incidental
5
Does your work involve novel problem-solving in ambiguous, poorly-defined situations?
AI is excellent at pattern matching against known domains. It struggles with genuinely novel situations — where the parameters aren't clear, where competing values have to be balanced, where the right framework for thinking about the problem is itself part of the problem. Original research, startup building, crisis management, and complex negotiation all live in this territory.
Safe: novel ambiguous problems Risk: pattern-matching on known data

What to Actually Do About It

The research is clear that workers who use AI tools outperform those who don't — within AI's capability range. The practical implication is that the most effective protective strategy is not to resist AI adoption but to get ahead of it in your own domain.

1
Audit your task composition, not your job title
Write out every task you perform in a typical week. For each task, ask: could a well-prompted AI do 80% of this adequately? That exercise will reveal where your genuine irreplaceability lives — and where you should be spending development time. Job titles are lagging indicators. Task exposure is the leading indicator.
2
Become the person in your field who uses AI best
In every industry in the augmented or split category, there is a clear emerging advantage for the practitioner who can deploy AI tools effectively. A lawyer who can use Harvey to do in 20 minutes what used to take a first-year associate two days is structurally more competitive. In most fields, that person hasn't emerged yet. See our guide to prompting AI effectively for a practical starting point, and our AI tools guide for category recommendations.
3
Build skills that compound on the human side of the ledger
Communication, negotiation, cross-functional leadership, deep domain expertise, and relationship management are the skills that become more valuable as AI handles more of the mechanical work. If your current role is heavy in AI-exposed tasks, the question is what you can build alongside them — the judgment and relational layer that makes the AI-assisted output yours rather than anyone else's.
4
Think in terms of adjacent roles, not career reinvention
Most workers displaced by automation historically move into adjacent roles that require similar underlying skills — not into completely different fields. A customer service representative at risk of displacement likely has strong communication, de-escalation, and problem-solving skills that transfer into coaching, training, complex sales, or healthcare support roles. The transition is rarely as wide a leap as the headlines suggest.
5
Watch where your industry is hiring, not just where it's cutting
Every sector undergoing AI disruption is simultaneously creating new roles. Prompt engineers, AI trainers, AI safety auditors, workflow automation designers, and AI governance specialists barely existed three years ago. The WEF's 2025 Future of Jobs Report projects 170M new roles created by 2030 against 92M displaced — a net gain of 78M. That is worth dwelling on, because the 2023 edition of the same report projected net losses of 14M, and this article cited that older figure until August 2026. The direction of the headline forecast reversed between editions. The new roles are real — they are just not evenly distributed across the same organizations that are contracting, and a net global gain is no comfort to anyone inside the 92M. [3]

New Roles Being Created by AI

These are not hypothetical future jobs — they are roles hiring now in significant numbers, most of which didn't meaningfully exist five years ago.

AI Prompt Engineer / Specialist
Organizations need people who can systematically get the best outputs from AI systems across use cases. Growing especially in marketing, legal, and customer experience functions.
AI Trainer / RLHF Specialist
Human feedback and preference labeling remains essential to model improvement. Requires domain expertise — medical AI trainers, legal AI trainers, and coding AI trainers all command premiums.
AI Governance & Ethics Officer
Enterprises need oversight of AI systems for bias, safety, regulatory compliance (EU AI Act, emerging US frameworks), and reputational risk. A growing specialization in legal and compliance.
AI Workflow Designer
Operational roles that design how humans and AI systems collaborate within business processes — where automation starts, where human review applies, how exceptions are handled.
MLOps / AI Infrastructure Engineer
Deploying and maintaining AI systems in production — model monitoring, data pipeline management, cost optimization, and reliability engineering. One of the fastest-growing engineering specializations.
AI-Augmented Domain Expert
Not a separate job title but a premium on existing roles: the radiologist who uses AI diagnostics, the lawyer who uses Harvey, the financial analyst who uses AI research tools. These practitioners are now in higher demand than their non-AI-using peers.
The Honest Bottom Line
The jobs most at risk are the ones we've been under-investing in preparing people to leave for decades. The jobs most resilient are often the ones we've systematically undervalued.

Routine data processing, basic customer service, templated creative work, and rote legal research were never the parts of those jobs that required the most human intelligence. They were the parts that paid the most entry-level workers because there was no cheaper alternative. AI is the cheaper alternative now, and that will cause real disruption to real people's livelihoods — particularly those at the beginning of careers in affected industries.

But the research does not support the apocalyptic narrative. The same economy that is automating data entry is creating an unprecedented demand for AI engineers, AI governance professionals, AI-augmented domain experts, and roles that don't yet have names. The historical pattern of technological transitions — painful for specific workers in specific industries, expansionary in aggregate — appears to be repeating. That aggregate expansion doesn't automatically protect any individual, which is why the practical steps above matter more than the macro narrative.

The most durable career strategy isn't to guess which jobs AI will never touch. It's to build the human capacities — judgment, relationship, domain depth, adaptability — that make the human-AI combination more valuable than either alone.

Corrections — 16 August 2026:This article was fact-checked after publication and several claims did not survive. The largest change: its headline projection was running the World Economic Forum's 2023 Future of Jobs Report, which the January 2025 edition supersedes — and the sign reversed, from a net loss of 14M roles by 2027 to a net gain of 78M by 2030. A claim that Getty Images reported a dramatic decline in stock-photo downloads was false; Getty's own SEC filings show downloads down about 1.5% a year and record 2025 revenue. Source [4] had fused two different studies into one citation, crediting Brynjolfsson, Li & Raymond with the BCG/Harvard "jagged frontier" finding, which is Dell'Acqua et al.; they are now listed separately. Frey & Osborne rated plumbers at 35%, not 26%, which places them mid-table rather than "among the lowest." The BLS and WHO figures were superseded editions. The claim that AI widens the gap between senior and junior engineers inverted the study cited to support it. One dead citation link was replaced.
Research & Sources
[1]
Goldman Sachs Global Investment Research — "The Potentially Large Effects of Artificial Intelligence on Economic Growth" (March 2023). Key findings: 300M full-time jobs exposed globally, 2/3 of US jobs have some automation exposure, 7% face >50% task automation.
goldmansachs.com up
[2]
Eloundou, T., Manning, S., Mishkin, P., & Rock, D. — "GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models" (arXiv:2303.10130, March 2023). OpenAI / University of Pennsylvania. Analyzes 1,016 occupations; finds 80% of US workers have ≥10% of tasks exposed to LLMs.
arxiv.org/abs/2303.10130 up
[3]
World Economic Forum — "Future of Jobs Report 2025" (January 2025). Projects 170M roles created and 92M displaced by 2030 across 1,000+ employers in 55 economies — a net gain of 78M. Supersedes the 2023 edition, which projected 69M created against 83M displaced by 2027, a net loss of 14M.
weforum.org up
[4]
Dell'Acqua, F., McFowland III, E., Mollick, E., et al. — "Navigating the Jagged Technological Frontier" (Harvard Business School Working Paper 24-013, 2023). Field experiment with 758 BCG consultants; source of the "jagged frontier" — AI delivers large gains inside its capability range and degrades performance outside it.
papers.ssrn.com up
[5]
Brynjolfsson, E., Li, D., & Raymond, L.R. — "Generative AI at Work" (NBER Working Paper 31161, 2023). A separate study of 5,179 customer-support agents: 14% average productivity gain, 34% for the least experienced, near zero for the most experienced.
nber.org/papers/w31161 up
[6]
Getty Images Holdings — Q3 2025 results (SEC Form 8-K, Exhibit 99.1). Paid download volume 93m vs 94m year over year; FY2025 revenue $981.3m, up 4.5% and the highest in company history.
sec.gov up
[7]
McKinsey Global Institute — "Generative AI and the Future of Work in America" (July 2023). Projects up to 12 million occupational transitions in the US by 2030 due to automation, concentrated in food service, customer service, and office support.
mckinsey.com up
[8]
Peng, S., Kalliamvakou, E., Cihon, P., & Demirer, M. — "The Impact of AI on Developer Productivity: Evidence from GitHub Copilot" (arXiv:2302.06590, February 2023). 55.8% faster task completion for developers using Copilot in controlled study.
arxiv.org/abs/2302.06590 up
[9]
Frey, C.B. & Osborne, M.A. — "The Future of Employment: How Susceptible Are Jobs to Computerisation?" (Oxford Martin School, September 2013; journal publication 2017). Estimates 47% of US employment in the high-risk category. The 2017 date is the journal version, not a revision — Frey & Osborne have never reissued revised probabilities, so the 2013 figures stand as originally published and should be read as a decade-old, pre-LLM methodology.
oxfordmartin.ox.ac.uk up
[10]
US Bureau of Labor Statistics — Occupational Outlook Handbook (2024–2034 projections). Data on projected employment changes across occupational categories, including administrative, legal, healthcare, and technology roles.
bls.gov/ooh up
[11]
World Health Organization — "Health Workforce" projections. Global healthcare worker shortage of 10 million projected by 2030, concentrated in low- and middle-income countries.
who.int up
[12]
Accenture — "A New Era of Generative AI for Everyone" (March 2023). Analysis across 22 job categories and 20 industries; finds 40% of all working hours can be impacted by LLMs, with augmentation the dominant near-term pattern in high-skill sectors.
accenture.com up
[13]
METR — "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity" (July 2025). Randomised controlled trial; experienced developers were 19% slower with AI assistance on real tasks in familiar repositories, while self-reporting a 20% speed-up.
arxiv.org/abs/2507.09089 up
[14]
Journal of Teaching and Learning 19(4) (2025). Randomised comparison of Khanmigo against a web-search control across 69 undergraduates: significant learning gains in all conditions, but no statistically significant difference between groups.
jtl.uwindsor.ca up
[15]
Autor, D. — "Work of the Past, Work of the Future" (AEA Papers and Proceedings, 2019). MIT labor economist's analysis of how automation historically polarizes labor markets — eliminating middle-skill routine work while increasing demand at both high and low ends. Foundational for understanding the current AI transition.
aeaweb.org up
S
Sheldon Valentine
Founder · Dear Tech

Sheldon covers AI technology, its labor market implications, and practical strategies for navigating the transition — grounded in the research rather than the hype cycle.

Share
Back to Blog