Silicon Valley GirlStanford's Top AI Economist: The Next 10 Years Will Be the Best AND the Worst in History
At a glance
WHAT IT’S REALLY ABOUT
AI is erasing entry jobs; agency, taste, and policy determine outcomes
- AI is already reducing employment in highly AI-exposed entry-level roles (notably for under-25s), while less-exposed care and physical jobs are growing and “augmentation” use patterns correlate with better outcomes.
- The biggest near-term shift is from doing execution work to defining problems and evaluating outputs—meaning many workers will manage a “fleet” of AI agents rather than perform tasks end-to-end.
- Economic impacts lag technical capability because organizations must redesign processes, train people, and connect AI to core value creation—similar to past general-purpose technologies but likely compressing to a 3–5 year window.
- Job outcomes depend on demand elasticity and task composition: automating a task can shrink jobs in some sectors but expand employment in others (e.g., radiology demand grew despite better image-reading AI).
- The next decade could be “best or worst” depending on choices around diffusion vs concentration of power, transition support (training/education), and mitigating catastrophic risks (misinformation, biosecurity, autonomous weapons).
IDEAS WORTH REMEMBERING
5 ideasEntry-level cognitive roles are the first major casualty—especially in exposed tasks.
Brynjolfsson cites Stanford/“Canaries” results showing ~16% lower employment for under-25s in the most AI-exposed occupations, with declines worsening month by month in areas like coding and call centers.
Think in tasks, not titles, to predict whether a job shrinks or grows.
Few occupations are fully automatable; most are a mix of tasks, so AI may replace specific components (e.g., image reading) while increasing the value and demand for the remaining human tasks (coordination, judgment, patient interaction).
Whether automation cuts jobs depends on demand elasticity, not just efficiency gains.
If AI lowers the “price” of a service and demand is elastic, total consumption can rise enough that hiring increases (radiology/medical imaging example); if demand is inelastic, spending and labor can fall.
The “winning” AI strategy for companies is value-linked process change, not gimmick pilots.
He contrasts capability hype with muted economic impact, arguing firms often build low-value demos (“better lunch menus”) instead of redesigning workflows, retraining teams, and shipping AI into core products.
The new career moat is problem definition and evaluation—agents do the execution.
He frames work as define → execute → evaluate, with AI agents rapidly improving at execution; humans differentiate by scoping the right questions, iterating, and judging correctness amid hallucinations and mis-specified goals.
WORDS WORTH SAVING
5 quotesThe next decade, if we play our cards right, will be the best decade in human history by far, or this could be, like, one of the worst 10 years ever.
— Erik Brynjolfsson
So I don't wanna sugarcoat it. The core folks who are using AI to automate their jobs in places like coding and call centers that are highly exposed, there was double-digit declines in employment. And since we published that paper-... we've continued to track it, and the effect's just getting bigger every month.
— Erik Brynjolfsson
This year, every single student, every single project, they have to have running code, because-... everybody's a coder now. Everybody's a coder now.
— Erik Brynjolfsson
If you don't have any, it doesn't do much for you. But if you've got a plan, this can totally amplify it. So the people in the future are the ones with a lot of high agency.
— Erik Brynjolfsson
If you're not both excited and scared, you're missing at least half the story.
— Erik Brynjolfsson
High quality AI-generated summary created from speaker-labeled transcript.
