AI isn't replacing most workers — it's making them dramatically faster. Here's the hard data from 2024-2026 randomized studies, and how to pick a major that captures the productivity dividend instead of getting flattened by it.
What the research actually shows in 2026
The headline-grabbing studies of the AI productivity revolution were not opinion pieces — they were randomized controlled trials. Three of them define what we now know:
The MIT/Stanford customer support study
A landmark 2023 study by Brynjolfsson, Li, and Raymond at Stanford and MIT tracked 5,179 customer support agents using a generative AI assistant. Average productivity rose 14%, but the gain was concentrated among less-experienced workers, who improved 34%. Top performers barely moved. The implication: AI lifts the floor faster than the ceiling.
The Harvard Business School BCG study
In 2023, Dell'Acqua and colleagues at Harvard Business School ran a randomized trial with 758 Boston Consulting Group consultants. Those given GPT-4 completed 12.2% more tasks, finished them 25.1% faster, and produced work rated 40% higher quality on tasks within AI's "frontier" of capability. On tasks outside that frontier, AI users performed worse than the control group — a finding the researchers nicknamed the "jagged frontier."
The MIT writing study
Noy and Zhang (MIT, 2023) studied 453 college-educated professionals on writing tasks. ChatGPT users finished tasks 40% faster with 18% higher quality ratings. Again, weaker writers gained the most.
The 2025 GitHub Copilot field study
GitHub's controlled study of 4,867 developers (published 2024, follow-up 2025) found Copilot users completed coding tasks 55% faster with no significant drop in quality — the largest measured productivity gain to date in any white-collar domain.
Pick a Major That Captures the AI Dividend →
The data is clear: in AI-augmented fields, the right major isn't just "safe" — it pays more because the AI multiplier compounds your output.
The productivity gain varies wildly by job type
Averages hide the real story. Here's what the 2024-2026 evidence shows by job category:
| Job category | Measured productivity gain | Source |
|---|---|---|
| Software engineering | 26-55% faster task completion | GitHub 2024, 2025 |
| Customer support | 14% avg, 34% for novices | Brynjolfsson et al. 2023 |
| Management consulting | 25% faster, 40% higher quality | Dell'Acqua et al. 2023 |
| Legal research | 32% time savings | Stanford HAI 2024 |
| Marketing copy & content | 40% faster | MIT 2023 |
| Medical diagnostics (radiology assist) | 10-15% accuracy lift | Nature Medicine 2024 |
| Skilled trades, nursing, teaching | 0-5% (limited applicability) | McKinsey 2024 |
Notice the bottom row. Hands-on work — nursing, trades, teaching — shows essentially no productivity lift from generative AI. That's why these careers remain among the most durable. But it also means workers in these fields don't get the AI compounding effect.
Who actually captures the productivity gains?
This is the question almost nobody is asking, and it's the most important one. When a worker becomes 40% more productive, three groups can capture the value:
- The worker (higher pay, faster promotion)
- The employer (lower headcount, higher margins)
- The customer (lower prices, more output)
A 2025 paper by Acemoglu and Restrepo at MIT examined the early evidence. In tight labor markets — software, healthcare specialties, skilled trades — workers captured most of the gain through wage increases. In loose labor markets — generic content writing, junior accounting, entry-level legal research — employers captured most of it through reduced hiring.
The lesson for major choice: credentials that are scarce capture AI's productivity dividend; credentials that are abundant give it away.
The 2026 wage data confirms it
BLS Occupational Employment and Wage Statistics released in May 2025 (covering May 2024 data) shows median wages for software developers rose 6.8% year-over-year, registered nurses 7.1%, and electricians 5.9% — well above the 4.1% all-occupation average. Meanwhile, median wages for general office clerks and entry-level paralegals rose only 2.4% and 2.1%, respectively.
What this means for your major choice
Use this 3-part filter when evaluating any major in 2026:
1. Does this major lead to a job where AI is a tool, not a replacement?
Strong examples: nurse practitioner, civil engineer, software developer who works on systems (not just code), occupational therapist, accountant who handles audit and advisory (not just data entry), high school teacher.
2. Does the credential carry licensing or scarcity?
Licensing creates a moat AI can't cross. RN, PE (Professional Engineer), CPA, JD with bar admission, MD/DO, electrician with state license — all create durable wage premiums.
3. Does the work require trust, judgment, or physical presence?
Therapy. Teaching. Building. Healing. Negotiating. These are the categories where, even if AI makes the worker 30% more productive, the worker still needs to show up.
Majors that fit all three filters: nursing, civil/mechanical/electrical engineering, accounting (if pursued through CPA), education (especially special ed and STEM teaching), social work, occupational/physical therapy, computer science (with focus on systems and architecture, not boilerplate code), construction management, and the skilled trades.
Find Your AI-Resilient Major →
Our quiz uses 60 questions and real BLS labor data to match your strengths to majors that capture the AI productivity dividend — instead of getting flattened by it.
Frequently Asked Questions
Is the 40% productivity number real or hype?
Both — the 40% figure comes from a real MIT study, but it applies to a specific task (writing) under specific conditions. Across all knowledge work, the average productivity gain is closer to 14-26%. Still real. Still significant. But not universal.
Will AI raise my salary or just let my employer hire fewer people?
It depends on how scarce your credentials are. In licensed or specialized fields (RN, PE, CPA, MD), workers are capturing most of the gain through higher wages. In abundant-credential fields (general business administration, generic communications), employers are capturing it through reduced hiring.
Should I just learn to use AI tools instead of going to college?
For most students, no. AI tools are a complement to a credential, not a substitute for one. The 2024 BCG study showed that consultants who already had domain expertise gained 40% from AI; novices given AI without expertise made worse decisions.
Which majors have the lowest AI productivity boost — and is that bad?
Hands-on majors — nursing, the trades, teaching — show 0-5% AI productivity lift. That's not bad. It means the work itself can't be automated, which is why these jobs are among the most secure and the wages are rising fastest in 2026.
Sources
- Brynjolfsson, Li, & Raymond. "Generative AI at Work." NBER Working Paper 31161, 2023. (Stanford / MIT)
- Dell'Acqua et al. "Navigating the Jagged Technological Frontier." Harvard Business School Working Paper 24-013, 2023.
- Noy & Zhang. "Experimental Evidence on the Productivity Effects of Generative AI." Science, 381(6654), 2023.
- GitHub Research. "Quantifying GitHub Copilot's Impact on Developer Productivity." 2024 update, 2025 follow-up.
- Bureau of Labor Statistics. Occupational Employment and Wage Statistics, May 2024 release (May 2025 publication).
- Acemoglu & Restrepo. "Tasks, Automation, and the Rise in U.S. Wage Inequality." NBER, 2025 update.
- McKinsey Global Institute. "The state of AI in 2024," and follow-up 2025 report.
- Stanford HAI. "AI Index Report 2025."