Will AI Kill Jobs or Create Them? The Real Risk Is the Vanishing First Job
Bill Gates warns AI can replace human cognition. Jensen Huang sees bigger markets and more work. The data so far fit neither camp cleanly: total employment has held up, but the door into the labor market is narrowing first.
Why Gates and Huang look at the same AI and reach different conclusions
In August 2026, Bill Gates spelled out why AI differs from past automation. Earlier machines mostly replaced physical labor or routine tasks. Generative AI, he argued, can replace human cognition itself, and it spreads much faster. He said the shock could reach law, customer service, health care, software and manufacturing, and that entry-level and mid-level jobs could come under pressure first.
Jensen Huang starts from a different place. Even if AI automates tasks such as coding, reading medical images and drafting documents, he argues, a company's goals and appetites are not fixed. When costs fall, firms can build more software, run more diagnoses and make more products. Huang sees this as productivity gains feeding back into growth and hiring.
The gap between them comes down to one question: which comes faster and bigger, the substitution effect or the demand-expansion effect?
The data show a weaker career ladder, not mass unemployment
- 19%: Employment gap for 22-to-25-year-olds in AI-exposed occupations as of June 2026, versus a path in which they grew as fast as low-exposure occupations.
- 11%: Actual decline in employment for 22-to-25-year-olds in high-exposure occupations between November 2022 and June 2026.
- 10%: Employment growth for the same age group in low-exposure occupations over the same period.
The Stanford Digital Economy Lab tracked the US labor market using ADP payroll data. Its findings come down to two sentences. Broad AI-driven unemployment is not yet visible across the economy. But young workers in heavily AI-exposed occupations are clearly falling behind. The adjustment shows up more in reduced hiring than in layoffs.
That difference matters. Before AI lays off senior staff in bulk, companies can hand over the research, first drafts, simple coding and customer support that juniors once did. The overall unemployment rate may barely move, yet the effect on a generation's first jobs and the path to building skills could be large.
The bigger risk may be the learning process, not the job
A randomized experiment released in September 2026 by David Autor and colleagues sharpens the point. It gave an AI drafting tool for three months to 133 patent attorneys at 11 US law firms. Output quality rose when they used AI, and juniors saw large immediate gains. But when professional judgment was tested without AI three months later, seniors had improved while juniors showed no average gain in skill.
There is counter-evidence. An experiment released in October with 673 German IT apprentices found that AI use raised success rates on hard tasks beyond their formal training by 26 to 31 percentage points, with no drop in immediate comprehension.
So the conclusion is not that using AI makes people dumber. When AI substitutes for practice, skill formation can weaken. When AI pushes people to attempt a harder task, it can widen what they are able to do. Design choices by companies and schools change the outcome.
White-collar work may weaken as blue-collar work strengthens
Huang's optimism is most persuasive in physical AI infrastructure. Data centers, chip plants, power grids and cooling systems cannot be built with software alone. At Carnegie Mellon's 2026 commencement, Huang told electricians, plumbers, steelworkers, technicians and construction workers, "This is your time." As AI buildouts expand in the US, shortages of skilled tradespeople and rising wages are appearing together.
That does not mean white-collar work disappears while only blue-collar work survives. As robotics improves, physical labor will face automation pressure too. A more realistic picture is repricing at different speeds by occupation. Repeatable digital execution gets cheap fast. Work tied to context and the physical world, such as on-site tasks, final judgment, accountability and client relationships, becomes relatively scarce.
| Path | Early effect | Variables that decide the long-run result |
|---|---|---|
| Cognitive substitution | Automation of entry-level office work, fewer hires, pressure on labor income | AI reliability, AI cost versus labor cost, how deeply firms cut headcount |
| Task expansion | Lower unit costs, more products and services | Demand elasticity, size of new markets, growth investment |
| Skill reshaping | Execution loses value; evaluation and accountability gain value | Training methods, mentoring, ability to judge without AI |
| Physical infrastructure | Data center and power spending lifts demand for skilled trades | Durability of AI capex, power access, construction permitting, pace of robotic automation |
Investors should watch where the productivity dividend goes
A company's claim that AI lifted productivity by 20% says little about shareholder value on its own. Where that 20% goes matters more. If it only cuts labor costs, near-term margins may improve, but growth could slow if demand across the industry is weak. If the gains flow into lower prices, new products, more customers and new markets, the demand-expansion path Huang describes opens up.
The winners of the AI era are therefore likely to be not simply the companies that cut the most people, but those that built a bigger market with the same workforce. In the labor market, likewise, the scarcer worker may not be the heaviest AI user but the one who can verify AI output, take responsibility for it and connect it to new tasks.
Insight Times Editorial Desk
