What Jobs Will AI Replace by 2040?
Nobody has a credible, named forecast for specific jobs AI will replace by 2040. That's not an evasive answer, it's the accurate one, and the reason is worth understanding rather than skipping past: 15-year labor forecasts have a bad track record, and the specific failures are well documented.
The track record of 15-year forecasts
In 2013, Oxford researchers Carl Benedikt Frey and Michael Osborne estimated that 47% of US jobs were at high risk of automation "perhaps in a decade or two" (Oxford Martin School). That's a 2013-to-roughly-2033 window, close enough to a 15-year horizon to be directly relevant here. Total US employment did not collapse. It grew. A later, more methodologically granular study out of the University of Mannheim put the comparable US figure at closer to 9%, not 47%, after correcting for the original study's tendency to flag entire occupations like fashion models and school bus drivers as highly automatable based on a handful of automatable sub-tasks (ITIF). Frey and Osborne's own paper measured technical exposure, not a timed prediction, but the number traveled through public discussion as if it were one.
Jeremy Rifkin's 1995 book "The End of Work" predicted a labor market breakdown as automation displaced human work at scale. Instead, the following 25 years saw nearly continuous US employment growth, from 124 million jobs in 1995 to 158 million by early 2020 (The Conversation). During the 1960s and 1970s, a wave of economists predicted office automation would permanently eliminate clerical employment. A 1986 study reviewing clerical employment from 1950 to 1982 found no persuasive evidence that automation had caused any significant decline, and concluded the earlier forecasts had rested on overoptimistic assumptions about how fast and how completely the technology would substitute for people (ERIC). Clerical employment in the US actually doubled between 1947 and 1956, from 4.5 million to 9 million, and kept growing through the 1980s even as computers spread through offices, because the same computers also generated more paperwork, more accounting, and more compliance and scheduling work that needed someone to handle it (Trends). It didn't eliminate the category, it changed what clerical work looked like. ATMs were widely predicted to replace bank tellers outright, and the actual record is more specific than that summary suggests. Between 1988 and 2004, the number of tellers needed per branch fell from roughly 20 to 13 as ATMs spread, yet urban bank branches expanded 43% over the same period, and total teller employment grew, because cheaper branch operations let banks open more of them (American Enterprise Institute). ATMs did not shrink the job. What did was mobile banking two decades later: bank branches peaked near 99,550 in 2009 and had fallen to under 78,000 by 2023, and BLS now projects teller employment to decline 8% over the coming decade (Vanguard). The lesson isn't that automation never shrinks a role. It's that the technology everyone blamed in advance (the ATM) wasn't the one that eventually did it, and the gap between the two was almost 30 years.
Why the 15-to-25-year horizon breaks forecasting
Two structural problems compound at this distance. First, occupations bundle many tasks together, and organizations restructure work around new technology slowly. BLS has noted this gap between a technology being capable of a task and that task actually stopping being done by a human has historically been long (BLS). A 2040 forecast has to get both the technology curve and the organizational-adoption curve right, and compounding two uncertain curves over 15 years multiplies the error, it doesn't average it out.
Second, and more specific to generative AI: nobody agrees on whether it behaves like past general-purpose technologies (electricity, computers, the internet), which took one to two generations for labor markets to fully absorb, or whether it moves faster because it's a more general tool. Economists are genuinely split on this, not in agreement with a minority dissent. Any 2040 forecast has to pick a side of that live disagreement and build on top of it, which is one more reason to treat specific 2040 job-loss numbers with real skepticism.
What's actually defensible for 2040
Direction, not numbers. The task-level pattern already visible in 2026 data, routine and low-accountability work under the most pressure, will very likely still be the operative pattern in 2040, because it's grounded in what AI is structurally good at (pattern completion across large amounts of text and data) versus what it isn't (physical action, legal accountability, in-person trust). That structural gap isn't going away by 2040 on any credible current roadmap. What's not defensible is a specific percentage of jobs lost by that date, because no methodology has a track record of getting that number right at this distance, and the 2013 Oxford example above is the clearest documented case of it going wrong.
If you're trying to plan a career around a 15-year horizon, the more useful question isn't "what percentage of my job will be automated by 2040." It's "which of my current tasks already show automation pressure, and is that pressure likely to compound or plateau." That's a task audit, not a date, and it's the same audit that applies at any horizon.
The tasks you keep decide how replaceable you are
The tasks you can still do without leaning on AI are what make you hard to replace here. The free 5-Day AI Reset is a five-email course built around exactly that: Day 2 has you take one task back and do it unassisted. One small change per day, and it stays useful no matter which way by 2040 moves.
Frequently asked questions
Is there a credible forecast for jobs AI will replace by 2040?
No named, methodologically sound forecast makes specific occupational predictions for 2040. The forecasts that exist for 2030 (WEF, McKinsey, Goldman Sachs) explicitly stop short of this horizon; going further requires assumptions none of them are willing to make with confidence.
Why don't economists trust 15-year automation forecasts?
Because the closest historical comparison, Frey and Osborne's 2013 estimate that 47% of US jobs were at high risk within a decade or two, didn't hold up. A later study found the real US figure closer to 9%, and total US employment grew rather than shrank over the period in question (ITIF).
Did past predictions about office automation and bank tellers come true?
Not as predicted. 1960s-70s forecasts of clerical employment vanishing due to office automation, and later predictions that ATMs would eliminate bank tellers, both proved overstated. The roles changed and shrank in some dimensions, but weren't eliminated (Slate).
What should I actually plan around for 2040?
Direction, not a number. Routine, low-accountability, digital-only tasks will very likely keep facing rising automation pressure. Build judgment, accountability, and relationship-based skills into your role now rather than waiting for a specific year.
Is generative AI different enough from past technologies to break the historical pattern?
That's genuinely disputed among economists, with credible researchers on both sides. Past general-purpose technologies took one to two generations to be absorbed by labor markets; whether generative AI moves faster is an open question, not a settled one.
Keep reading
For a sector view, see Jobs That AI Can't Replace.
Plan around your own tasks, not a distant date
A 2040 forecast can't tell you anything about your specific job. Take the free How AI-Proof Is Your Job? assessment to see which of your current tasks already show automation pressure.