What Changed in BLS’s 2024–34 AI Occupation Projections?

What changed between the 2023–33 and 2024–34 forecasts?

The headline changed in degree, not direction. In the 2023–33 forecast, BLS projected total U.S. employment to rise 4.0 percent over the decade. In the 2024–34 cycle, the projected increase was 3.1 percent. Several technology-linked occupations still sat well above that economy-wide rate, but their projected growth rates were lower than in the prior release: data scientists moved from 36.0 to 33.5 percent; information security analysts, from 32.7 to 28.5 percent; computer and information research scientists, from 25.6 to 19.7 percent; and software developers, from 17.9 to 15.8 percent. These are revised estimates for two different ten-year periods, not a count of jobs that AI removed or a measure of any one applicant’s chances. BLS’s “Industry and occupational employment projections overview and highlights, 2023–33” and its 2024–34 employment table provide the like-for-like occupation comparisons. The comparison is useful precisely because the pattern is mixed. A switcher looking only at percentages might treat the highest rate as the best destination. But a percentage describes change relative to an occupation’s projected starting size; it does not show how many positions that change represents. In the 2024–34 table, the 33.5 percent projected increase for data scientists corresponds to 82,500 additional jobs. Software developers’ lower 15.8 percent corresponds to 267,700, the largest numerical increase among the selected occupations BLS lists. The preceding 2023–33 overview made the same scale distinction visible: it projected 73,100 additional data-scientist jobs and 303,700 software-developer jobs. Neither quantity tells us how many openings will be entry-level, local, or suitable for a particular career changer. They do show why a high growth rate and a large potential market are different signals. Forecast vintages change because BLS updates the period and underlying assumptions; they are not successive measurements of the same decade. The 2023–33 release described total employment rising from 167.8 million in 2023 to 174.6 million in 2033, while the later release projected 2024 employment forward to 2034. A lower rate in the later cycle therefore does not mean that a corresponding number of jobs disappeared between publications. Nor does the comparison isolate AI’s effect. BLS’s 2024–34 article, “Artificial intelligence, information technology, and employment, 2024–34,” connects expected technology-related demand with occupations such as data science, security, and software development. Its table reports projected employment change, not a causal estimate of jobs created by AI. The safer reading is that BLS’s outlook remained favorable for selected roles while its estimate of their pace, and the overall economy’s pace, was more moderate. For a career decision, that is a shortlist signal rather than a green light. Compare relative growth with the occupation’s absolute scale, then ask what the work requires and whether the route fits your circumstances. The projection can justify investigating data science or software development; it cannot justify quitting, buying a credential, or assuming demand will translate into an accessible offer. Treat the change between cycles as a reminder to compare both measures and verify the target against current duties, prerequisites, and local opportunities before making a costly move.

Sources: Artificial intelligence, information technology, and employment, 2024–34; Industry and occupational employment projections overview and highlights, 2023–33; Employment Projections news release, August 28, 2025

What does 'AI-related' mean in these projections?

Here, “AI-related” describes occupations BLS discusses in connection with building, integrating, analyzing, securing, or using AI systems. It is not a count of jobs caused by AI. The 2024–34 overview links projected demand for computer and mathematical work to AI-based systems and the people needed to develop or integrate them, while also citing cybersecurity, cloud services, data storage, and broader IT investment. These are multiple influences on an outlook, not a measured estimate of AI’s separate contribution. A ten-year projection cannot assign each expected position to one technology or business decision. Read the figures as a national forecast for occupations connected to a changing technology ecosystem, not as an AI jobs tally. [Industry and occupational employment projections overview and highlights, 2024–34](https://www.bls.gov/opub/mlr/2026/article/industry-and-occupational-employment-projections-overview.htm) explains the rationale behind the estimates; it does not provide a causal decomposition.

Several distinct signals sit behind the headline. Technical capability concerns what a system can do under specified conditions. Observed use concerns interactions with the technology; employer adoption concerns whether organizations incorporate it into work. Task redesign concerns how responsibilities change, including who verifies outputs and remains accountable. Labor demand is projected employment; displacement is the reduction or loss of work for people. These can influence one another, but one does not prove the next. A system’s ability to draft code does not show that employers have deployed it, reorganized teams, or reduced headcount. Likewise, BLS’s expectation of increasing AI integration is projection reasoning, not a census of current use or proof of uniform adoption. Nor does growth in a technology-linked occupation establish that AI caused the growth: a forecast association is not a causal estimate.

BLS’s AI exposure categories address potential applicability of AI to occupational tasks. They combine theoretical measures with measures based on observed AI interactions mapped to tasks or work activities; the latter do not directly establish whether workers in an occupation used AI on the job. BLS says exposure does not imply job loss, productivity gains, automation probability, or wage effects. The categories are not forecasts of employment growth or decline, adoption probability, or worker replacement, and they do not distinguish automation from augmentation. Their inputs also have limits: some theoretical measures reflect capabilities no later than mid-2023, and some observed measures may lean toward early adopters. Exposure and projected demand can therefore differ without contradiction. An occupation may include tasks AI could assist while employment is projected to grow for other reasons. [Artificial Intelligence (AI) exposure categories](https://www.bls.gov/emp/publications/ai-exposure-categories.htm) sets out these boundaries.

The BLS factsheet on AI impacts says the agency researches emerging factors that historical data may miss, but generally adjusts projections conservatively when evidence for change is convincing. It also says AI’s labor effects remain highly uncertain and cannot be predicted precisely ten years ahead; projections are updated annually as new evidence becomes available. In the 2024–34 overview, AI development and integration are among several reasons for expected demand in computer and mathematical occupations, alongside cybersecurity, cloud, and data needs. That is a conditional outlook, not realized hiring, a measure of employer adoption, or proof that projected growth offsets displacement elsewhere. For a switcher, use the figures to shortlist occupations for closer investigation, then check their duties, entry requirements, and local opportunities. Do not treat a growth rate as proof AI created those jobs or an exposure label as a forecast of decline. [Artificial intelligence (AI) impacts on employment projections](https://www.bls.gov/emp/factsheets/ai-impacts.htm) describes BLS’s method and uncertainty; neither source gives an individual job-loss probability.

Sources: Industry and occupational employment projections overview, 2024–34; Artificial Intelligence (AI) exposure categories; Artificial intelligence (AI) impacts on employment projections

Why does a high growth rate not equal an accessible career opening?

A growth rate is relative change in jobs over a stated period. Net employment change is the projected difference between starting and ending employment. Annual openings are a separate flow: they include openings from growth and from replacing workers who leave an occupation permanently. The measures answer different questions. None says how many vacancies will appear in your city, suit a new entrant, or match your qualifications.

The 2024–34 table in “Artificial intelligence, information technology, and employment, 2024–34” shows the scale distinction. BLS projected data scientists to grow 33.5%, a net increase of 82,500 jobs, while software developers’ 15.8% represented 267,700 additional jobs. A percentage compares pace relative to each occupation’s base; the count shows projected scale. Neither counts applicants or guarantees a hiring opportunity.

Annual openings add another lens, but they are not net new positions or a pool reserved for career changers. The Occupational Outlook Handbook’s “Computer and Information Technology Occupations” page estimates about 280,000 openings a year on average for the group in 2025–35, from growth and replacement of workers who leave permanently. This is a broad group estimate for a different cycle, not 280,000 entry-level jobs each year. Replacement openings can sustain hiring even when total employment changes little; net growth does not show how hiring is distributed by employer, seniority, or place.

Entry requirements further separate projected demand from an individual’s access to work. In the BLS Occupational Outlook Handbook profile “Data Scientists,” the agency says workers typically need at least a bachelor’s degree in mathematics, statistics, computer science, or a related field; some employers require or prefer graduate degrees. The profile also notes the mathematical and statistical study involved. This is a national description of typical preparation, not a rule that every employer applies. Still, it is a reason to compare your current preparation with actual duties before treating data science’s projected growth as a personal opportunity. A short course may help with a defined gap, but the projection itself does not show that a certificate substitutes for the preparation employers seek.

BLS’s current “Occupational projections and worker characteristics” table places projected change and annual openings beside typical education and related-work experience. For 2025–35, it lists a bachelor’s degree and less than five years of related experience as typical for information security analysts; for computer and information research scientists, a master’s degree. These summaries are not universal hiring rules. They show why growth alone can mislead: strongly projected fields may demand different preparation, with different costs in study time, tuition, and forgone earnings.

Geography and household constraints belong in the same check. National projections do not establish that jobs exist near you, meet your salary floor, allow remote work, or fit care responsibilities. They also do not show whether you can train while keeping income. A national median wage is not an entry-level offer in your location.

Before paying for a credential or leaving a role, review recent postings for one target occupation where you could work. Note recurring duties, tools, credentials, experience, location, and listed pay; distinguish repeated requirements from preferences and compare them with skills you can demonstrate. If the same gap recurs, choose a learning step or bounded work sample to address it. If prerequisites require substantial time or cost, test an adjacent role using your domain experience first. That does not mean adjacent hiring is easier.

Use the BLS figures as a reason to investigate a field, then let task fit, local evidence, entry requirements, and your constraints determine the next move. A high projected rate can support a shortlist; it cannot make the opening accessible by itself.

Sources: Artificial intelligence, information technology, and employment, 2024–34; Occupational Projections and Worker Characteristics; Computer and Information Technology Occupations; Data Scientists

How should a reader use the newer 2025–35 cycle?

For a decision now, use BLS’s 2025–35 projections as the current national baseline, while reading 2024–34 as the historical forecast this article examines. Each cycle is a new ten-year estimate built from then-current data and assumptions, not a rolling correction that turns the prior forecast into a record of what happened. BLS’s “EP Notices” lists the 2025–35 release on August 27, 2026, and dates the 2024–34 release to August 28, 2025. Keeping both vintages visible answers two questions: what did BLS expect earlier, and what does its latest cycle now project?

The current “Occupational Projections and Worker Characteristics” table shows why “AI jobs” is too broad a label for choosing training. For 2025–35, data scientists are projected to increase 34.6 percent, from 275,600 to 371,000 jobs, a net gain of 95,400. Computer user support specialists are projected to decline 3.5 percent, or 26,200 jobs; computer programmers, 7.3 percent, or 8,100. Software developers are projected to grow 10.2 percent, or 174,700 jobs. These figures describe different occupations and scales; they do not show that AI caused any change. The data-scientist rate alone also cannot tell a switcher how many suitable jobs will be nearby.

The table places employment change beside annual-average openings, median annual wages for 2025, typical education, and related experience. A growth rate is relative to an occupation’s base; net change shows national scale, while annual openings are another measure and not a count of new jobs. Typical entry education is a starting description, not a universal employer rule. Check the detailed occupation and requirements repeated across current postings where you could realistically work. Salary needs, caregiving, health, and study time can make a national trend impractical.

BLS’s separate “Artificial Intelligence (AI) Exposure Categories” answers a different question. The agency supplements its 2025–35 projections with categories combining theoretical exposure and measures of observed AI interactions mapped to occupational tasks. It cautions that these measures do not directly observe whether workers in a given occupation used AI on the job. Exposure therefore cannot explain projected employment changes by itself or establish displacement. Occupations bundle tasks, and a category cannot tell an individual how a particular employer will redesign work.

A switcher can use the new cycle without abandoning the historical question: take 2024–34 as evidence of what that forecast said, then use 2025–35 to refresh a present-day shortlist. For one target occupation, compare the current projection with its duties, preparation, openings, and a sample of local postings. If a full transition would cost too much time or income, test an adjacent task or project first. If postings repeatedly require credentials or experience outside your means, the national rate should not overrule that constraint. This check keeps the next learning step tied to actual work, not just a headline. The newer cycle is a reason to update the investigation, not a verdict on which career to choose.

Sources: Occupational Projections and Worker Characteristics; Artificial Intelligence (AI) exposure categories; EP Notices

What does the forecast comparison say about a career switch?

The comparison supports a shortlist, not a resignation letter. A switcher can use the 2024–34 projections to decide which fields deserve closer checking, but a high growth rate alone says little about whether a particular person can enter a particular role. A plausible move must fit existing skills, the preparation employers expect, local opportunities, acceptable pay, available study time, and responsibilities outside work. Treat the projection as a national demand filter; treat occupational profiles as a check on duties and preparation; then use current local postings to test whether the route is practical. Consider an example: an insurance-operations worker is curious about data science. Experience interpreting claims processes or explaining recurring operational problems may supply useful domain context, but it does not by itself establish readiness for data-science work. BLS’s Data Scientists profile describes duties including identifying useful data, collecting and analyzing it, creating and testing models, presenting findings, and making recommendations. It says entrants typically need at least a bachelor’s degree in mathematics, statistics, computer science, or a related field; some employers require or prefer graduate study. The profile also notes that some employers seek industry-related experience or coursework. That combination matters: familiarity with insurance may be relevant, while quantitative and programming preparation may still be a substantial gap. It places projected change, employment scale, annual openings, wages, and typical education alongside one another for the 2025–35 cycle. A national outlook cannot show that suitable openings exist within reach, meet a salary floor, or accept the worker’s experience. Nor does a typical education category guarantee that every employer uses the same hiring rule. The practical test is to inspect a small, current sample of postings in the worker’s reachable geography. Record repeated requirements: degree or certificate expectations, statistical and coding skills, years of relevant experience, domain knowledge, work location, and stated compensation where available. Compare these against a candid inventory of demonstrated skills. If the same prerequisite appears repeatedly and is genuinely missing, identify what kind of learning or experience could supply evidence of it before buying an expensive program. If postings mainly ask for applied analysis within an industry the worker already knows, an adjacent analytics or reporting role may be worth testing first. This is a hypothesis to validate, not a claim that adjacent jobs are easier to obtain. A useful comparison is between a full switch and a staged test. A full switch may make sense when the person already has much of the required foundation, can meet financial and time constraints, and finds local evidence consistent with the national outlook. A staged route could involve building a small, reviewable analysis of a work-related process using permitted data, then asking whether the resulting skills map to actual roles. The aim is to learn whether the work and prerequisites fit before committing to a long retraining path. In short, the 2024–34 forecast can justify investigation, not a costly credential or abrupt departure from current work. Use domain knowledge, verify duties and entry requirements, and test skills against reachable openings. Make the larger move when that evidence, the required preparation, and household constraints line up.

Sources: Data Scientists; Occupational Projections and Worker Characteristics; Operations Research Analysts

Should a switcher retrain fully or test an adjacent move first?

An adjacent move is often the lower-commitment first experiment when it lets someone use existing domain knowledge while building one missing skill. A full switch is more defensible when the target's prerequisites, finances, local opportunities, and work itself already fit. Neither route is automatically easier or safer. National BLS projections can guide what to investigate, but they do not establish which route is attainable for a particular worker.

Compare three paths by duties, proof expected, missing preparation, and cost in time, money, location, and pay. The first path upgrades the current role. Someone handling insurance operations might test a more systematic way to summarize claims patterns or check a recurring report, while retaining responsibility for definitions, exceptions, and decisions. This hypothetical exercise is not evidence of employer AI use or new roles; it can test interest, data access, and skill gaps.

The second path moves toward an adjacent function, such as reporting or operations analysis in a familiar sector. That route may preserve knowledge of how claims, customers, rules, or service delays work while adding data handling, visualization, or quantitative reasoning. It still requires verification. BLS's Data Scientists profile describes work that includes finding useful data, collecting and cleaning it, analyzing it, testing models, visualizing findings, and communicating recommendations. It says data scientists typically need at least a bachelor's degree in a quantitative or related field; some employers require or prefer graduate study. These duties and typical requirements are comparison points, not proof that adjacent roles have the same entry bar or that sector familiarity replaces technical preparation.

The third path is a larger occupational change, such as aiming directly for data science or another technical specialty. It may fit someone who already has substantial quantitative preparation, wants the core work, and can meet the cost of deeper study. The BLS profile notes that some employers seek industry-related experience or education, making prior domain knowledge potentially relevant. Relevance is not a hiring guarantee. Compare current local postings with your evidence of skills rather than infer accessibility from a national growth rate or title.

Operations research offers a comparison, not a shortcut. The BLS Occupational Outlook Handbook lists operations research analysts among occupations with duties related to data scientists and identifies a bachelor's degree as typical entry education. A small project could document a service bottleneck, define a measurable question, and present an analysis with limitations. It cannot certify readiness for every employer or replace prerequisites that postings repeatedly specify.

Choose the smallest test that answers a consequential uncertainty. Review local postings for recurring duties, tools or credentials, experience expectations, location, and pay range where available. Compare these with current strengths and gaps. If several postings ask for the same missing quantitative or programming foundation, follow a work sample with structured study rather than treating the sample as a credential. If the duties appeal but a transition would jeopardize a salary floor or conflict with family responsibilities, an internal project or adjacent role may be more proportionate while the larger option remains open.

Treat an upgrade, adjacent move, and full switch as routes to test, not a universal ranking. When uncertainty is high, prefer a reversible experiment: produce a reviewable work sample, compare it with real role requirements, and check whether local opportunities fit practical constraints. Keep the full switch in view if the test confirms interest and preparation and local evidence supports the move. Change course when task fit, requirements, finances, or openings point elsewhere.

Sources: Data Scientists; Operations Research Analysts; Occupational Projections and Worker Characteristics

What would make the recommendation change?

The staged route should change when the evidence stops pointing to a useful experiment and starts supporting a feasible transition. If you already meet the target role’s prerequisites, have relevant work samples, and can identify current openings in places you can realistically work, a full application or structured retraining plan may be more sensible than spending months proving an adjacent step. If local searches repeatedly reveal no feasible adjacent roles, reconsider the route without assuming another occupation is automatically safer. Start with postings that are current and geographically relevant. Compare several examples of the same target job, and record the duties, required versus preferred skills, credentials, experience, location, and disclosed pay range. Distinguish recurring requirements from one employer’s preferences, then compare them with your demonstrated skills and gaps. The BLS table, “Occupational projections and worker characteristics,” provides national projections alongside typical entry education and related-work-experience categories; It frames questions, but cannot establish local openings or your eligibility; typical education is not a universal employer rule. For data science specifically, BLS’s “Data Scientists” profile describes collecting, analyzing, and modeling data, and says a bachelor’s degree in a quantitative or related field is typical preparation, with some employers preferring graduate study. If your background includes quantitative work and sound analysis, test direct applications against local requirements. If you lack a recurring prerequisite, a focused course or project may be a smaller test than enrolling in a degree program. Make the project resemble a target task, use permitted data, and show your reasoning. It tests fit and gaps, not hiring readiness. Also put a number on the transition’s practical cost for your household: tuition or course fees, study hours, time away from paid work, commute or relocation needs, and the income floor you cannot cross. If those costs do not fit, an adjacent role may remain the better near-term choice even when a direct switch looks attractive on paper. If postings consistently require a missing credential, assess whether a recognized program is worth its cost; a projection rate does not prove payoff. Finally, keep forecast dates in perspective. BLS’s “Projections Archive” preserves prior cycles, while the current occupational table reflects a newer national estimate. Revisions show that assumptions and estimates change; they do not document realized losses or settle your local prospects. Recheck the latest cycle when making a current decision, but let local duties, actual requirements, finances, and evidence from a work sample determine whether to upgrade, move adjacently, or make a larger change. Change course when those decision-relevant facts change, not merely because a headline forecast moves.

Sources: Occupational Projections and Worker Characteristics; Data Scientists; Projections Archive

What is the next practical step?

Choose one target occupation and make the next decision clear. Start with five recurring tasks from your week: preparing a report, reconciling records, answering a customer question, checking an exception, or explaining a recommendation. Note the inputs, where judgment or approval is required, and what a mistake would cost. This describes your work; it does not predict whether your role will disappear. Compare the list with the relevant BLS Occupational Outlook Handbook profile and current local postings. The BLS Occupational Projections and Characteristics table gives national context on projected change, annual openings, and typical education. It cannot show which openings are entry-level or available to you. Record requirements that recur across postings, alongside experience, location, and stated pay range. Verify one-off requirements before treating them as barriers. Choose one repeated skill gap and test it in a small project using public data or a non-sensitive work sample. Define the task and check the output against a reliable reference. If the work fits and the gap is learnable within your time and budget, compare a focused course, guided project, or deeper qualification against that requirement before enrolling. If it does not fit, test an adjacent role that draws on more of your experience. The free task checker can help organize this review. Its change-pressure signals are not a probability of displacement and cannot replace local posting research. Commit to a larger switch only when task fit, prerequisites, finances, and credible local evidence align.

Sources: Occupational Outlook Handbook; Occupational Projections and Worker Characteristics

Questions readers ask