What is AI going to do to my retirement?

This summer the research confirmed what many people over 55 suspected: AI is ending some careers early. None of the coverage attaches a dollar figure, so this post does.

Boston College's Center for Retirement Research matched job exits against occupational AI exposure and found that AI "could be pushing some older workers" out early, naming programmers and accountants among the most exposed. In AARP's March 2026 survey, 24 percent of workers 50-plus see AI as a threat in their line of work. This sits on top of an old risk: ProPublica and the Urban Institute found that 56 percent of workers over 50 are pushed out of a job at least once, and only one in ten ever again earns as much. Anthropic CEO Dario Amodei told Axios that AI could eliminate half of all entry-level white-collar jobs and push unemployment to 10 to 20 percent within five years.

The household

A single accountant in New Jersey, age 57, earning $120,000. Savings of $900,000: $700,000 in a 401(k), $150,000 in a taxable account with a $90,000 cost basis, $50,000 in cash. Spending is $66,000 a year after tax, in today's dollars. Social Security is $36,000 a year at the full retirement age of 67, and the plan was to work to 65. Each strategy runs 5,000 simulated lifetimes, each with its own market, inflation, and lifespan path.

Working to 65, the plan succeeds in 89 percent of lifetimes. With income stopping at 57 and nothing else changed: 24 percent.

The collapse: eight years of salary and 401(k) contributions (and their growth) gone, the portfolio funding $66,000 a year from day one so early down years lock in, and eight years of pre-Medicare health insurance. One offset: with earned income near zero, the ACA prices the first marketplace years at almost nothing.

Can you live on less?

Yes, and it is the strongest lever. Cutting spending 10 percent, to $59,400 in today's dollars, lifts success from 24 to 38 percent; cutting 20 percent, to $52,800, reaches 59 percent. A dollar of spending cut is a dollar not withdrawn this year and not funded for the next thirty.

Can you earn anything at all?

More than the amount suggests. Consulting income of $30,000 a year from 57 through 62, a quarter of the old salary, moves success from 24 to 42 percent, because the portfolio is spared its deepest withdrawals in the years when a bad market does permanent damage. Six years of 1099 work after a layoff is an assumption, not a guarantee; the model does not price the job market's demand for 57-year-olds.

Should you claim Social Security early?

If nothing else changes, yes. Claiming at 62 for a reduced $25,200 a year beats the planned claim at 67, 28 versus 24 percent, and delaying to 70 for $44,640 does worse at 22 percent. When we ran 62-versus-70 for a household in good shape, the two paths came out even; here delay drains the portfolio faster now for larger checks later, and in a thin plan the money can run out before they arrive.

Repair the plan first, with the 10 percent cut and the consulting income, and the ranking flips back: 70 beats 62, 61 versus 58 percent. The claiming answer depends on which repairs came first.

The scoreboard

The deep re-plan lands at 84 percent, five points shy of the plan the accountant lost, without a day of full-time work after 57. Living on $52,800 instead of $66,000 is a real loss, but it is the cheapest path back to a plan that works.

What if AI does not stop with your job?

Suppose AI transforms the whole economy. This scenario writes one version down as assumptions: stocks return 9.5 percent instead of 8 as capital captures more of the economy; equity volatility rises from 16 to 24 percent as gains concentrate; inflation averages 4.5 percent instead of 3, with bonds at 5 percent and more risk; and the labor market never wants the accountant back.

In that world the re-plan drops from 61 to 46 percent and the deep re-plan from 84 to 66 percent. Owning capital helps, since the index funds participate in the boom, but 4.5 percent inflation against a fixed budget, on a drawn-down portfolio at twice the volatility, eats most of the benefit. This is one imagined path, not a forecast; it is fully specified so any assumption can be changed and rerun.

What this looks like from 35, 45, and 55

At 35, AI is a career-length risk, not a retirement risk: test your savings rate against an income that stops at 55 instead of the straight-line career the plan assumes. Kiplinger reports parents delaying retirement to fund children frozen out of entry-level jobs.

At 45, "work to 65" has become the assumption holding up the plan. Run it with income ending at 55: above 70 percent there is slack; below 40, the plan depends on the meeting never happening.

At 55, the CRR data says the exposure is now. Have the re-plan ready before it is needed: the cut-to spending number, the realistic bridge income, and the claim age the repaired plan prefers.

This is one household, one state, one set of assumptions; your balances, taxes, and spending floor move every number here. What carries over is the shape: spending is the strongest lever, bridge income defends the decisive years, and claiming timing depends on which repairs came first.