Jobs Replaced by AI: Which Roles Are Most Exposed, and How to Prepare in the Next 2 Years
In August 2026 the Bureau of Labor Statistics did something it had never done before: it published an AI exposure category for 831 occupations, next to its own ten-year projections. This page uses that file rather than anyone's intuition — including ours.
Which jobs are being replaced by AI?
The occupations most exposed to AI are almost entirely office and computer-based: programming, web and database work, data entry, bookkeeping, technical writing, customer service, administrative support, paralegal work and design. That is the Bureau of Labor Statistics' own classification, not a prediction of ours.
But exposed is not the same as disappearing. BLS says so explicitly, and its own data proves it — several of the most exposed occupations are projected to grow strongly. Read the exposure column and the growth column together, or you will reach the wrong conclusion.
- Estimated time
- 6 min read
- Cost / impact
- Free · sourced to BLS
- What you need
- Your occupation
What “exposed” actually means
Before the table, the definition — because almost every article on this topic gets it wrong, and the error changes the conclusion entirely. BLS built its categories by combining five published studies, taking median percentile ranks within the theoretical and observed families, and clustering the result into four groups. Then it said this:
Exposure signifies that AI technology could be or has been used to assist or complete some of the work performed in an occupation. Exposure does not imply job loss, productivity gains, automation probability, or wage effects.
The exposure category is relative to other occupations. It does not imply an absolute “high” or “low” level of exposure.
So exposure measures overlap — how much of what an occupation does is work these systems can now do or assist with. It is relative to other occupations, and it says nothing on its own about whether the occupation shrinks. Here is that distinction in one comparison, both rows from the same file, both in the highest band:
Information security analysts
Very highexposure+21.0%
projected employment change, 2025–35
Network and computer systems administrators
Very highexposure-4.1%
projected employment change, 2025–35
The roles most exposed to AI
Every occupation below is one BLS places in its highest relative-exposure band. They are ordered by the Eloundou task-exposure measure — the share of an occupation's tasks a language model could halve the time on — so the ranking rests on a published number rather than our opinion.
Read the last column with the others. A high exposure score beside strong projected growth means the work is changing, not ending.
| Occupation | Tasks driving exposure | Task exposure | Projected 2025–35 |
|---|---|---|---|
| Proofreaders and copy markers43-9081 | Reading text against a style guide and marking deviations. | 98% | -0.9% |
| Computer programmers15-1251 | Writing, translating and debugging routine code from a specification. | 95% | -7.3% |
| Database administrators15-1242 | Routine tuning, backup scripting and access administration. | 94% | -0.1% |
| Web developers15-1254 | Building standard page and component code from a design. | 93% | +3.8% |
| Database architects15-1243 | Producing schemas, migrations and query optimisation from requirements. | 91% | +9.4% |
| Data entry keyers43-9021 | Transcribing and validating structured data between systems. | 89% | -25.5% |
| Software developers15-1252 | Generating, refactoring and documenting code; writing tests. | 87% | +10.2% |
| Telemarketers41-9041 | Delivering scripted calls and handling scripted objections. | 81% | -21.4% |
| Bookkeeping, accounting, and auditing clerks43-3031 | Coding transactions, reconciling ledgers, chasing exceptions. | 80% | -5.6% |
| Word processors and typists43-9022 | Producing formatted documents from dictation or drafts. | 74% | -34.4% |
| Network and computer systems administrators15-1244 | — | 73% | -4.1% |
| Technical writers27-3042 | Turning source material into structured documentation. | 70% | +0.8% |
| Data scientists15-2051 | Exploratory analysis, feature engineering and model boilerplate. | 69% | +34.6% |
| Information security analysts15-1212 | Alert triage, log analysis and report writing. | 65% | +21.0% |
| News analysts, reporters, and journalists27-3023 | Drafting copy from source material to a house format. | 65% | -5.9% |
| Executive secretaries and executive administrative assistants43-6011 | Scheduling, correspondence, briefing notes and meeting summaries. | 65% | 0.0% |
| Human resources specialists13-1071 | Screening applications, drafting job ads, answering policy questions. | 64% | +6.4% |
| Secretaries and administrative assistants, except legal, medical, and executive43-6014 | Correspondence, records, scheduling and document preparation. | 64% | -6.0% |
| Computer network architects15-1241 | Producing standard designs and configuration from requirements. | 64% | +7.7% |
| Computer network support specialists15-1231 | Diagnosing and resolving routine network faults from tickets. | 63% | +1.2% |
| Computer and information research scientists15-1221 | Literature synthesis, experiment scaffolding, and write-up. | 63% | +21.8% |
| Tax preparers13-2082 | Applying rules to a set of documents to produce a return. | 63% | +4.7% |
Exposure band:BLS “Relative AI exposure”, published 27 August 2026. Every row shown is “Very high”. Task exposure: the beta measure from Eloundou, Manning, Mishkin and Rock — the share of tasks a language model could halve the time on, counting app-mediated exposure at half weight. Note these are not fully independent: BLS used that paper as one of its five inputs, so treat the column as texture rather than corroboration. Projected change: BLS Employment Projections 2025–35.
Tasks driving exposureis LayoffNext's summary of the kinds of task the underlying studies score highly for each occupation. It describes the shape of the work, not any employer's plans.
Jobs that are least exposed
The pattern at the other end is consistent: the least exposed occupations are done with your hands, in a specific place, on a person or a machine — skilled trades, clinical and diagnostic healthcare, therapy, equipment repair and field safety work. Almost no desk job sits in the two lower bands, which is a finding rather than an omission. We list 46 of them with pay, projected growth and entry routes on their own page.
AI-proof jobs: 46 lower-exposure careersHow to tell whether your own job is at risk
Two different questions get muddled here. Exposure is a property of your occupation and it is published — look it up rather than guessing, because intuition about this is demonstrably unreliable. Risk is a property of your specific job at your specific employer, and none of the national data can see it. These are the local signals that can:
- Tooling is being built or bought specifically to handle your team's workflow.
- Hiring is frozen in your function while it continues elsewhere in the company.
- Targets are rising without more people, and “do more with less” has entered the vocabulary.
- Contractors are replacing employees in your area rather than supplementing them.
- Two roles have been quietly merged when someone left and not backfilled.
- Work that used to come to you is being routed somewhere else first.
How to prepare in the next two years
Two years is long enough to change your position materially and short enough that the work has to start now. The order below matters more than the content: the cheap, reversible moves come first, and the expensive irreversible one comes last.
Find out where you actually stand
- Look up your occupation's exposure band and its projected change — both columns, not one.
- Write down every task you did last week and mark the ones that match the “driving exposure” column above. That list, not the job title, is your real position.
- Build the savings buffer while you still have income. Every option below is wider with more runway behind it.
Move up your own occupation, not out of it
- Take on the parts of your role that are least like the exposed tasks — judgement calls, client relationships, work nobody has written down.
- Get fluent with the tools that are changing your field. Being the person who evaluates and supervises the output is a different job from being the person who produces it by hand.
- Document decisions you owned, not tasks you completed. That distinction is what survives in the record.
Build the option before you need it
- Identify one lower-exposure role you could genuinely reach, and find out what its credential costs and how long it takes where you live.
- Talk to two people already doing it. Ten minutes of that beats ten hours of reading about it.
- Keep your network warm while you are still employed — it is far easier to ask for a conversation than for a job.
Commit, or consciously decide not to
- If you are retraining, start while you still have income rather than after a layoff forces the timing.
- If you are staying, make that a decision rather than a default, and revisit it when the next projections release lands.
- Either way, keep the runway number current. It is what converts a shock into an inconvenience.
One thing not to do: enrol in something expensive in the first month after a layoff. That is when the fear is highest and the information is worst. Work out your runway, keep searching in your existing field, and make the retraining decision from a calmer position — the job you take now is what funds the retraining you start later.
Related LayoffNext Tools
Keep the momentum going with the tools people use most alongside this guide.
Frequently asked questions
Which jobs are most likely to be replaced by AI?+
Does high AI exposure mean the job is going away?+
How many jobs will AI replace by 2035?+
Will AI replace software engineers?+
What skills are hardest for AI to replace?+
How do I know if my job is at risk?+
Sources & methodology
Exposure bands, projected employment change and the occupation titles on this page are the Bureau of Labor Statistics', reproduced verbatim. The task-exposure percentages are the beta measure from Eloundou et al. The “tasks driving exposure” column is LayoffNext's summary of the task types those studies score highly, and is labelled as ours in the table notes. No figure on this page is estimated.
- AI exposure categories and 2025–35 employment projections — U.S. Bureau of Labor Statistics, Employment Projections programEvery exposure band on this page, the definition of what exposure means, and BLS's statement that it does not imply job loss. · Last verified Sep 16, 2026
- Employment Projections, 2025–35 — U.S. Bureau of Labor StatisticsProjected employment change 2025–35 for every occupation shown. · Last verified Sep 16, 2026
- Eloundou, Manning, Mishkin & Rock, "GPTs are GPTs: Labor market impact potential of LLMs" — Science 384(6702), 2024The task-exposure percentage in the table, and one of the five studies BLS combined to build its categories. · Last verified Sep 16, 2026
Important note
This article is educational and is not career, financial, or legal advice. The exposure categories are a national statistical classification, not a prediction about any individual job, employer or region. Projections are ten-year national estimates, and BLS itself notes that the effect of a new technology a decade out cannot be predicted with precision. LayoffNext is not affiliated with or opposed to any employer.
Deepak Middha is the founder of LayoffNext and a Chartered Accountant (ICAI, India). A U.S. immigrant with nearly 20 years of experience — and 17 years in hedge fund and private equity administration, including as Vice President of Fund Accounting at NAV Fund Administration Group and Associate Director of Private Equity and Real Estate at SS&C Technologies — he builds free, plain-language layoff tools and guides for employees, H-1B workers, and immigrant families.
Layoff recovery tips for employees
Get practical layoff recovery tips, financial planning reminders, job-search guidance, and new free tools.
Free weekly email. No spam. Unsubscribe anytime. Learn what's inside