The Bureau of Labor Statistics projects 750,000 office and administrative support jobs will vanish from the U.S. economy by 2035. That number is real. The panic it triggers is not.
Here is what the data actually shows when you stop reading headlines and start reading the underlying research.
The 60 percent that the human does became even more valuable, because it's an economic complement to the 30-40 percent that now became cheaper.
The 68% Fact That Changes Everything
The BLS evaluated 831 occupations for “AI exposure” — a measure of how much of a job’s tasks an AI system could theoretically perform or assist with. Of the 206 occupations ranked in the highest exposure quartile, 141 are still projected to grow over the next decade. That is 68 percent.
High AI exposure is not a synonym for job destruction. It is a synonym for task reallocation.
Andrew Ng, the Stanford computer scientist and AI educator, put it plainly in a recent interview: “The 60 percent that the human does became even more valuable, because it’s an economic complement to the 30-40 percent that now became cheaper.”
The mechanism is straightforward. Automate the middle of a workflow — the execution step — and the remaining definition and evaluation steps become scarcer, not less important. Economists call this the complementarity effect. Most coverage of AI and employment ignores it entirely.
Where the Growth Actually Is
The BLS projections through 2035 rank the fastest-growing high-paying occupations. The list does not look like a technology conference agenda.
Nurse practitioners are the single fastest-growing occupation in the United States, with a projected increase of 41 percent. The median salary is $132,000. The BLS classifies this role as having high AI exposure. Yet it is growing faster than any other job in the country. AI can draft clinical notes, flag drug interactions, and summarize patient records. It cannot perform a physical exam, assume legal liability for a diagnosis, or sit with a frightened patient for 20 minutes.
Physician assistants earn $136,000 and are projected to grow 20 percent. They sit in the lowest quartile of AI exposure — one of the few roles in the entire BLS database that combines high pay, strong growth, and minimal overlap with current AI capabilities.
Construction managers are adding 55,000 jobs. The connection to AI is indirect but structural. Every data center, chip fab, and power plant announced this year becomes a construction project with a manager in a hard hat. Jensen Huang, CEO of Nvidia, told Fox Business in August that “hundreds of thousands of jobs are being created right now” in the physical infrastructure layer of AI.
Software developers — the profession that internet commentary has been writing obituaries for since 2023 — are projected to add 174,000 jobs through 2035, the largest absolute gain of any occupation paying over $100,000. The median salary is $135,000. AI exposure is high. Job growth is still positive.
The TechCrunch analysis of SignalFire hiring data tells a more textured story. Engineering hires at large tech companies have fallen roughly 25 percent since 2019. But engineers now account for 55 percent of all new hires at those same companies, up from 46 percent. Engineering is the most resilient function in technology, not the most fragile. The Indeed hiring lab found that U.S. software job postings rose 15 percent after Claude Code launched — while overall postings fell 7 percent. But 71 percent of those gains went to senior roles. The entry point is narrowing. The profession is not shrinking.
The Five Occupations That Are Shrinking
The BLS also identifies five occupations with the fastest projected decline. All five rank in the highest quartile of AI exposure.
Word processors and typists lead the decline at negative 34 percent. Data entry keyers follow at negative 25 percent. Payroll and timekeeping clerks at negative 15 percent. Accounting and auditing clerks at negative 5 percent — but that is 85,000 jobs on a base of 1.5 million workers. Customer service representatives also decline by 5 percent, a seemingly small percentage that conceals 141,000 lost positions because 2.6 million Americans do this work.
A Goldman Sachs analysis cited by CNBC found that U.S. call center employment is running 39 percent below trend. Klarna’s CEO stated publicly that its AI agent handled the equivalent work of 850 full-time employees. The company also said it wanted to offer human support as a premium VIP experience. That is the future of the role — smaller, higher-value, not eliminated.
The Task Framework That Explains the Difference
Erik Brynjolfsson, the Stanford economist who has spent two decades studying how technology affects labor markets, offers a three-column model for understanding any job.
Define the question. Execute the work. Evaluate the result.
For most of history, humans did all three. AI agents are now competitive on the middle column — execution — across a growing range of tasks. That changes the value of the other two columns. A professional who can define the right question and evaluate whether the answer holds up under scrutiny becomes more valuable, because the execution cost dropped to near zero.
The second-order effect that most coverage misses is the seniority premium. If AI automates execution, the premium shifts to judgment and accountability. That is excellent news for professionals who already have those skills. It is a problem for mid-career workers who spent 15 years building execution expertise but have not yet developed the strategic judgment that now commands the premium. The labor market is not eliminating their profession. It is stranding them in the middle of it.
What the Data Means for a Professional
The BLS data and the expert commentary converge on three conclusions that matter for someone making a career decision today.
First, do not confuse AI exposure with job risk. The single fastest-growing occupation in the country has high AI exposure. The occupations with the highest AI exposure are still more likely to grow than shrink. Exposure measures task overlap, not replaceability.
Second, the entry point is compressing even where the profession is growing. Software engineering is adding jobs, but the new jobs are senior. The same pattern will repeat in any profession where AI can handle the junior-level execution work. The path from entry to seniority may require deliberate skill-building in the definition and evaluation columns — not just more years of execution.
Third, the physical economy is a hedge against the digital one. Construction managers, nurse practitioners, and physician assistants are all growing faster than most technology roles. None of them can be fully automated because none of them are purely digital. The “safe career” advice of the past two decades — get a job behind a screen — may be exactly wrong for the next decade.
The Practical Test
Brynjolfsson recommends a simple exercise. Take your last two weeks of work. Write down what you actually did. Then sort every task into one of three columns: define, execute, evaluate.
The tasks in the middle column are candidates for automation. The tasks in the outer columns are where your career value is moving. If most of your time is in the middle column, the trajectory is not mysterious. The question is whether you start moving toward the edges before the market moves for you.
The BLS publishes its full occupational projections annually. The AI exposure ratings are in the public dataset. The data is not hidden. Most professionals simply never look at it.
- How many office jobs will be lost by 2035?The Bureau of Labor Statistics projects 750,000 office and administrative support jobs will vanish by 2035.
- What percentage of high-AI-exposure occupations are still growing?68% of occupations in the highest AI exposure quartile are projected to grow over the next decade.
- Which occupation is the fastest-growing in the US?Nurse practitioners are the fastest-growing occupation with a 41% projected increase.
- What is the complementarity effect?Automating the execution step makes the definition and evaluation steps scarcer and more valuable.
- What does Andrew Ng say about AI and job value?Andrew Ng states that the 60% humans do becomes more valuable as it complements the cheaper automated portion.
