AI insiders and the rest of us are picturing two different 2030s
Anthropic asked nearly 11,000 Americans and its own staff what AI will do by 2030. They agree on today and disagree wildly on tomorrow, especially about their own jobs. Here's the gap, explained, with what Canadians think.
Based on: The AI expectations gap, by Anthropic Research ↗Same AI, same year. One group sees a gentle breeze. The other sees a blizzard.
Ask two neighbours what next winter will be like. One checked the long-range forecast. The other is going by how the last few winters felt. You'll get two very different answers, and both will be sincere.
Anthropic just did that with AI. It surveyed 10,980 American adults and 1,334 of its own staff, the people who build AI and use it all day, and asked where AI is headed by 2030. The two groups mostly agree on what AI can do today. They live in different worlds when it comes to what happens next.
- By 2030, 91% of Anthropic staff expect AI to make a Nobel-level discovery. Only 39% of US adults do.
- The biggest gap is about people's own jobs: staff expect AI to be able to do 75% of their work by 2030. US workers say 28%.
- Both groups reason the same way: the more of your own job you think AI can do, the higher you expect unemployment to be.
- People who already use AI at work sit somewhere in between, which hints that expectations shift with experience.
- This was a US survey. Canadian surveys show a similar "not my job" pattern.
Meet the cast
Cartoon stand-ins. The real study surveyed US adults (through the research firm Morning Consult) and Anthropic employees.
Today: they basically agree
The researchers listed eight tasks, from writing routine emails to making a Nobel-level scientific discovery, and asked when AI could do each one as well as a skilled professional. Both groups ranked the tasks in nearly the same order of difficulty. About a quarter of each group even thinks AI can already design, run and publish a scientific study.
Then you ask about the future, and the canoe splits in two.
"AI will make a Nobel-level discovery by 2030"
Share of each group who agree.
Source: Anthropic, August 2026 surveys.
The starkest difference: depending on the task, 19% to 40% of US adults said AI will never be able to do it. Among staff, it was 2% at most.
The real gap: "how much of MY job?"
Part of the gap is simply experience. Ask how long a task that suits AI (like summarizing a document) takes with AI, counting the time spent checking and fixing its work:
"With AI, the task takes a quarter of the time or less"
Share who say so.
Same story with the big list of tasks. Among Americans who regularly use AI at work, 56% expect AI to handle at least seven of the eight tasks by 2030, compared with 41% of all US adults and 28% of people who have never used an AI tool. Staff? 96%.
- Automation vs. augmentation
- AI doing the task on its own, versus AI helping a person do it faster. By 2030, most staff expect AI to work alone or with a person just checking. Four in ten US adults expect humans to still do most of the work.
- Diffusion
- How fast a technology actually spreads into everyday use. US adults expect AI to be used for 42% of the work tasks it's able to do by 2030; staff expect 65%.
- GDP growth
- How fast the whole economy grows in a year.
The surprising part: they think the same way
The researchers told everyone that US unemployment is about 4.2% today and asked them to guess the rate in 2030. Here's the twist. Both groups follow the same rule of thumb.
US workers who expect AI to do less than half their work forecast 7% to 8% unemployment. Those who expect AI to do 80% or more forecast 12.2%, which is above its peak during the Great Financial Crisis. Among US workers, every ten extra percentage points of "AI can do my work" goes with about 0.4 points more expected unemployment. At the same level of expected automation, staff and the public forecast about the same unemployment rate.
In other words, the disagreement isn't about how economies work. It's about what people have seen AI do up close.
Run everyone's answers through Anthropic's economic model and the gap gets dramatic:
GDP growth in 2030 implied by each group's answers
Median respondent, run through Anthropic's economic scenarios model. Not a forecast.
Even the public's 5.4% is more than twice the economy's typical growth rate, according to the authors. The quiz-takers were 23,075 people from 140+ countries who chose to try Anthropic's online scenario tool.
Bonus plot twist: even though staff expect AI to do most of their own work, 95% of them expect Anthropic's headcount to grow over the next year. As the authors note, automating tasks and cutting jobs aren't the same forecast.
- It's an American survey, run online with opt-in panelists, so it misses people without internet access. Responses were weighted to be nationally representative.
- The staff survey was an internal poll of personal views (1,334 people, unweighted). The authors stress these are staff members' own views, not Anthropic's plans or forecasts.
- The GDP numbers are what a model produces from people's answers. They're a way to compare beliefs, not predictions.
- Views move fast: when the survey was repeated in September, the share of US adults expecting a Nobel-level AI discovery by 2030 rose from 39% to 44%.
Why this matters in Canada
Nobody asked Canadians in this study, but Canadian surveys show the same split between "AI will change jobs" and "not mine, though."
- In a 2026 survey of about 2,000 Canadian workers with the Angus Reid Group, 26% said machines would likely be doing much of their job within a few years, up from 17% in 2023. Yet more than three-quarters believed most other workers would say their jobs were at risk. The researchers call it "pluralistic ignorance."
- An Angus Reid Institute poll in May 2026 found 45% of Canadians expect AI to significantly reduce the number of jobs over the next decade. Just 3% expect a significant increase.
- Meanwhile, use is climbing. Statistics Canada found 35.9% of workers used generative AI tools at work in the 12 months to March 2026, and 19.2% of businesses used AI to produce goods or deliver services.
These surveys asked different questions, so don't line the numbers up against the American ones. But the pattern rhymes: Canadians expect big changes "out there" while most expect their own work to be fine. Anthropic's finding suggests that as more of us use AI daily, our expectations will move. In which direction, and how far, depends on whether AI ends up doing as much work in a Winnipeg office as it does inside an AI lab.
What you can do tomorrow
- Form your own view from your own work. Pick one real task you do every week and try it with an AI tool. Time it honestly, including checking and fixing. That's exactly the experience that separated the groups in this study.
- Think in tasks, not job titles. List the parts of your job. Which ones could AI help with, and which need you? That's a better guide than any headline.
- Treat forecasts as beliefs. Whether it comes from a moose or a beaver, a 2030 prediction is a guess. Ask who is guessing and what they've seen.
Pop quiz: what share of US adults said AI will never be able to do some of the eight tasks?
Between 19% and 40%, depending on the task. Among Anthropic staff, it was 2% at most. Never say never, eh.
This is our plain-language take on Anthropic's research, with our own analogies and cartoons. For the full survey, charts and methods, read the original.