What the BLS Outlook Says About AI and HVAC Technician Careers

What is the direct BLS answer?

For the United States, the Bureau of Labor Statistics’ Occupational Outlook Handbook (OOH) projects employment of heating, air conditioning, and refrigeration mechanics and installers to grow 11 percent from 2025 to 2035. It projects about 40,600 openings a year, on average, across that decade. The Handbook commonly calls these workers HVAC technicians. Those figures describe a favorable national demand outlook; they do not say that AI causes the growth, prevents it, or has no effect. The OOH gives a projection and its stated demand factors, not an HVAC-specific estimate of jobs changed or removed by AI. [Heating, Air Conditioning, and Refrigeration Mechanics and Installers: Occupational Outlook Handbook](https://www.bls.gov/ooh/installation-maintenance-and-repair/heating-air-conditioning-and-refrigeration-mechanics-and-installers.htm)

A separate BLS product adds AI-specific occupational information. Its supplemental exposure categories compare occupations using theoretical measures of whether AI could assist with or complete some work, alongside measures mapping observed AI interactions to tasks or work activities. BLS says even these observed-use measures do not directly establish that workers in a particular occupation used AI on the job. The categories are relative comparisons, not forecasts of employment growth or decline, wages, productivity, adoption, or worker replacement. They also do not distinguish automation from augmentation. [Artificial Intelligence (AI) exposure categories](https://www.bls.gov/emp/publications/ai-exposure-categories.htm)

The careful reading is neither that BLS has no AI-specific material nor that its HVACR projection is an AI risk verdict. The OOH addresses labor demand for a defined U.S. occupation over ten years. The supplemental product adds a different signal about potential task overlap. Neither establishes what a particular contractor or building owner will adopt, how a technician’s duties will change, or what will happen to an individual worker’s job. Occupational exposure, employer adoption, and displacement are separate questions; a measure of one cannot answer the others.

For someone considering entering or staying in the trade, or weighing a change, the 11 percent projection is a national baseline, not a guarantee of local work or technological immunity. Annual openings do not mean every opening is a newly created job, and neither figure attributes demand to data centers or AI unless BLS makes that connection. Read the outlook as positive projected demand with AI’s specific occupational effect unresolved by these figures. This keeps the BLS numbers useful without turning them into a personal job-loss prediction they were not designed to provide.

Sources: Heating, Air Conditioning, and Refrigeration Mechanics and Installers: Occupational Outlook Handbook; Artificial Intelligence (AI) exposure categories

What exactly does the HVACR outlook say?

The national outlook is positive for the occupation BLS calls “Heating, Air Conditioning, and Refrigeration Mechanics and Installers,” often shortened to HVAC technicians. The Occupational Outlook Handbook projects employment to grow 11 percent from 2025 to 2035, from 440,900 jobs in 2025 to 489,100 in 2035. That is a projected net increase of 48,200 jobs across the decade. BLS also estimates about 40,600 openings per year on average. These figures describe the U.S. occupation as a whole; they do not say that generative AI creates those jobs, prevents their loss, or leaves every technician’s work unchanged. The different numbers answer different questions. The 11 percent is the percentage change in the occupation’s employment stock between the base and projected years. The endpoints show the estimated size of that stock at each end of the period. The 48,200 is the rounded difference between those endpoints, or net projected employment growth. The annual openings figure is a flow of opportunities expected to arise in a typical year, averaged over the decade. It is not 40,600 additional positions added to the occupation every year. A substantial portion of openings is expected because workers transfer to other occupations or leave the labor force, including through retirement. An opening may therefore replace a departing worker without increasing the total number of HVACR jobs. Reading “openings” as annual new jobs would overstate what the projection says. The two BLS pages present the same broad outlook at different levels of detail. The OOH summary gives the 11 percent growth and average annual openings. Its employment-projections table identifies occupation code 49-9021 and reports the base-year and projected employment, numeric change, and percentage change. That detail matters because “HVAC technician” in everyday speech can sound like one uniform job, while the BLS category is a defined occupational grouping covering heating, air conditioning, and refrigeration mechanics and installers. It is the grouping to which these national figures apply. BLS names commercial construction, including data centers, and residential building construction as expected drivers of employment growth. It also says the growing number of sophisticated climate-control systems is expected to raise demand, and that technicians will be needed to retrofit, upgrade, or replace existing systems. Those are the drivers the OOH states. The page does not assign a portion of projected growth to AI or claim that AI demand in data centers will translate directly into technician hiring. “Including data centers” is part of BLS’s description of commercial construction, not evidence that AI itself explains the forecast. A ten-year national projection is best treated as a demand baseline under the projection’s assumptions, not as a promise about any particular city, employer, or year. National growth can coexist with slower growth, different hiring patterns, or fewer suitable openings in a particular area. The OOH links readers to state and area wage estimates and to state employment projections, which offer a more relevant next comparison for someone deciding whether to train or move. Regional figures still cannot tell a reader whether a role fits their experience, schedule, physical needs, income floor, or training access. For a career decision, keep the units straight: 11 percent is projected occupational growth; 48,200 is projected net growth over ten years; about 40,600 is average annual openings that include replacement needs. These figures are neither an AI exposure score nor a probability that an individual will keep or lose a job. National demand is projected to rise, but the outlook does not settle how AI will alter duties or how much hiring will occur near a given worker. Check local openings, area wages, and entry requirements alongside the projection before treating it as a reason to enter, stay, or leave HVACR.

Sources: Heating, Air Conditioning, and Refrigeration Mechanics and Installers: Occupational Outlook Handbook; Occupational Projections and Characteristics

Does projected growth mean BLS assumes AI has no effect?

No. BLS does not say AI has no effect on HVAC employment. Its 2025–2035 employment projections assume technological progress and productivity broadly continue along historical patterns, while the agency researches new technologies and makes evidence-based adjustments where convincing grounds exist. BLS presents the result as a conditional national scenario, not a causal estimate of AI’s contribution to HVACR growth or a guarantee that the assumptions will hold. A projection can include technology without predicting every new tool’s effect in advance. In the “Employment Projections: 2025–2035 Technical Note,” BLS says its projections describe what would be expected under specified assumptions and circumstances. They focus on long-term structural trends rather than anticipating business cycles. When assumptions do not occur, actual values can differ. The figure describes the model’s result, not a promise about employer behavior or an individual technician’s experience. BLS’s technology assumption is not “nothing changes.” The technical note says the agency expects technology to progress and productivity to increase, but projects historical relationships forward on the assumption that the future will behave comparably to the past. That provides an empirical baseline and sets a limit: if technology advances much faster than it has historically, those relationships may no longer hold, and BLS says its methods may not yield reasonable results in that situation. BLS addresses AI directly. Recent AI developments, the technical note says, raise the prospect that technological progress could accelerate. Faster productivity growth could feed into higher overall output in the projection framework. If productivity growth were uniform across industries, BLS explains, its full-employment assumption would mean higher output and productivity but unchanged employment projections. If effects differed by industry, the methods could capture those differences, but BLS says it lacks data on which to base them. The agency therefore presents a scenario grounded in historical relationships instead of speculative adjustments. This explanation is general; it does not identify an AI adjustment specific to heating, air conditioning, and refrigeration mechanics and installers. That choice is a methodological boundary, not evidence that AI cannot alter the trade. BLS says it researches factors that may not yet appear in historical data, including new technologies, and generally applies research-based adjustments conservatively when convincing evidence supports a change. The separate “Artificial Intelligence (AI) impacts on employment projections” factsheet says the Employment Projections program considers potential AI impacts in its annual process. It also stresses uncertainty: the exact labor-market effect of new technology ten years out cannot be predicted precisely. Both points can be true: AI is within BLS’s analysis, and the published occupation projection does not isolate or quantify AI’s contribution. BLS’s annual update cycle is one response to that uncertainty. The agency says each new set incorporates new data, research, and analysis. That does not mean the next release will necessarily revise HVACR employment because of AI; the sources give no such direction. It means projections can change as evidence and assumptions change. When comparing projection vintages, check their periods and methods. The method also recognizes that occupational change can happen through task composition before it appears in total employment. The technical note says technology can alter the weighting of tasks within an occupation without necessarily changing overall demand for that occupation. Thus, positive projected demand and changing work practices are not inherently contradictory. An occupation-wide projection does not say every task remains untouched, every employer adopts the same systems, or workers’ daily responsibilities stay constant. The BLS growth projection is useful evidence about a modeled national demand scenario, but it does not isolate AI’s contribution. Nor does avoiding speculative productivity adjustments amount to a finding of zero effect. The careful reading is narrower: BLS assumes technology continues broadly in line with historical experience, considers new evidence, adjusts conservatively where it can support doing so, and acknowledges unusually rapid change could make historical relationships less reliable. Use the projection to inform, not settle, an individual career decision. These BLS explanations do not answer whether a particular employer will adopt a tool, redesign a service workflow, or change staffing.

Sources: Employment Projections: 2025–2035 Technical Note; Artificial Intelligence (AI) impacts on employment projections

What does BLS's separate AI exposure table add?

BLS adds a comparative signal, not an AI verdict on HVAC work. Its 2025–35 exposure categories sit alongside employment projections and compare occupations using five external sources: three theoretical measures of whether AI capabilities could assist with or complete work, and two measures based on observed AI interactions mapped to tasks or work activities. BLS cautions that observed measures do not directly show whether workers in a particular occupation used AI on the job. Capability, recorded interaction, adoption, and employment outcomes are separate questions. (U.S. Bureau of Labor Statistics, “Artificial Intelligence (AI) exposure categories.”) The inputs differ. Theoretical sources relate AI applications to workplace abilities or assess whether a large language model might reduce time spent on tasks. Observed sources map work-related Claude interactions to O*NET tasks and Microsoft Copilot activity to intermediate work activities linked to tasks. BLS converts source results to percentile ranks, calculates theoretical and observed dimensions, then groups occupations into four relative categories: low, moderate, high, and very high. This is a comparison framework, not a count of technicians using AI. The sources reflect distinct capabilities, datasets, and task mappings; combining them does not erase those differences. “Relative” is central to interpreting a category. BLS ranks occupations against others within each source; the result is not an absolute measure of how much AI will affect a job. A low category does not establish safety from change, and a high one does not mean employment will decline. BLS says categories are not forecasts of employment growth or decline, wages, productivity, AI adoption, or worker replacement. They also do not distinguish automation, where technology performs work, from augmentation, where it assists a person. The table can show how occupations compare under selected exposure measures. It cannot convert that comparison into an HVAC technician's chance of job loss or a count of AI-created or removed positions. The method has further limits for any occupation row. Results depend on source coverage, occupational crosswalks, missing-value imputation, normalization, and category choices. BLS reports some sources did not cover every occupation and missing ranks were imputed; these decisions can affect the category. The theoretical sources reflect capabilities available no later than mid-2023, while observed-use measures may lean toward early adopters. Most inputs focus on language models, not image or video generation. BLS notes the sources treat occupations as bundles of tasks or abilities without accounting for bottlenecks and complementarities. A tool's ability to assist with one information task would not show that a full HVAC service visit can be completed autonomously. That is an interpretation of the measure's limits, not a measured HVAC result. BLS links a downloadable table pairing categories with 2025–35 projections, but its XLSX could not be inspected in this research pass because the browser could not fetch the spreadsheet. This section therefore does not state an HVACR category or repeat an outside site's label as BLS data. To find the classification, open the current BLS download and methodology, confirm the occupation definition and columns, and check the update date. Treat a verified row as a prompt to examine tasks: which involve information handling, which require physical access or measurement, and where must a person verify and take responsibility? The table adds comparative context to the outlook, not an AI forecast or individual outcome.

Sources: Artificial Intelligence (AI) exposure categories

Which parts of the HVAC task bundle make the BLS signals coexist?

O*NET’s profile for heating, air conditioning, and refrigeration mechanics and installers helps explain how AI exposure can coexist with projected growth: HVACR combines information work, customer interaction, inspection, decisions, and physical service. Tasks include discussing malfunctions, testing circuits, consulting manufacturer recommendations, inspecting systems, replacing components, and keeping records. O*NET lists information processing alongside mechanical repair and handling objects. These descriptions do not show how a technician divides a week or whether an employer uses AI. The BLS Occupational Outlook Handbook describes the occupation and national outlook, not local adoption. Assisting one step differs from taking responsibility for a service visit. Consider an illustrative workflow: after recording a fault code and measurements, a technician might use software to find a manual passage or organize a service note. These are plausible applications, not evidence of HVAC adoption. The result must match equipment and conditions. Information retrieval may be easier to digitize than reaching equipment, but that alone does not show a task is automated or removed from a job. O*NET describes physical inspection and repair: testing circuits, checking leaks, connecting wiring, installing components, and replacing parts. These steps may require access, instruments, noticing conditions absent from records, and confirming a repair worked. Digital diagnostics might support some work, but O*NET does not measure system capability or whether a worker can safely rely on it. A suggestion is not a verified measurement; an instruction is not a completed repair. Someone remains responsible for checking the result. Customer judgment is part of the bundle. O*NET lists discussing malfunctions with users to isolate problems or verify repairs; the BLS Occupational Outlook Handbook describes explaining problems and recommendations. Listening to a customer and clarifying when a fault occurs connect findings to a particular building. A text tool might help prepare an explanation, but cannot establish that it fits the equipment or diagnosis. The task includes context and consequences, not only words. Occupational growth and task change can coexist. BLS’s projection concerns jobs in a broad occupation under a national scenario; exposure asks whether tasks may overlap with AI capabilities or observed interactions. Growth does not show that every task stays unchanged. The International Labour Organization’s *Generative AI and Jobs: A 2025 Update* analyzes nearly 30,000 tasks. It helps explain varied relationships to generative AI, but is not evidence of U.S. HVAC adoption or employment outcomes. The OECD’s *The OECD AI Exposure Measure* compares capabilities with occupational requirements across cognitive, social, and physical domains. It explains what a capability-gap measure assesses, not what HVAC employers deploy or what happens to a technician. A useful personal check starts with actual duties, not the occupation label. List recurring tasks involving information or records, measurement and interpretation, physical access and repair, customer explanation, and verification. For each information step, ask whether an approved tool is available, what source it uses, how errors are caught, and whether checking it saves time. For diagnostic or hands-on steps, identify what must be observed and what confirms the result. This inventory can reveal where workflows might change and where experience is needed to evaluate output; it cannot predict job loss or give a personal probability. Map repeated information tasks, their measurements and judgment, and who is accountable for diagnosis and repair. If a tool is introduced, compare the task before and after, including verification time and new documentation or explanation. Without evidence of local use, treat examples as possibilities to investigate, not descriptions of the trade today. O*NET names the task bundle and BLS gives an occupation-level outlook; neither replaces observing a workplace.

Sources: Heating, Air Conditioning, and Refrigeration Mechanics and Installers; Heating, Air Conditioning, and Refrigeration Mechanics and Installers: Occupational Outlook Handbook; Generative AI and Jobs: A 2025 Update; The OECD AI Exposure Measure

Can AI in an HVAC system be confused with AI replacing a technician?

No. AI inside an HVAC system, AI used as a technician’s work aid, and a system that replaces a technician are different claims. The first concerns how a building regulates equipment; the second concerns information support; the third would require a broader change in service tasks, responsibility, and staffing. A building-control project can show that one kind of automation is being evaluated. It cannot by itself show that service technicians are being replaced or that employers are reducing their numbers. Slipstream’s June 30, 2026 page, “Join our AI-enabled HVAC controls field study,” says the project is co-funded by Focus on Energy and WPPI. It aims to evaluate energy savings, cost-effectiveness, and operations and maintenance improvement, and was seeking demonstration sites when the page appeared. The page describes AI control that autonomously optimizes a building’s HVAC energy consumption, alongside “AI-driven insights.” This is evidence of a planned field evaluation of building controls. It does not report completed results, technician staffing changes, or an employment effect. It supports a narrow statement about what researchers intended to evaluate, not a conclusion about adoption across the trade. The word “autonomously” needs its object. Here, the stated function is optimizing energy use in the HVAC system. That is control of equipment operation, not an autonomous service visit: diagnosing a fault, making a repair, verifying the result, and explaining it to a customer. Slipstream’s page does not say its controls perform those service tasks. A controller’s ability to adjust system operation does not show that it can replace an installer or repair technician. Software assistance is another possibility. A tool might help locate a passage in technical documentation or organize service notes. Those are plausible examples of information support, not results reported by Slipstream and not proof that HVAC technicians commonly use such tools. Assistance with one step does not establish that software can confirm a diagnosis, determine whether a repair is safe, or verify the system afterward. The workflow still needs verification and accountability. A tool can change one task without removing the service role. The U.S. Bureau of Labor Statistics makes a related boundary clear in “Artificial Intelligence (AI) exposure categories.” Its relative occupational categories do not measure employment growth, probability of adoption, worker replacement, wage effects, or productivity. BLS also says they do not distinguish automation from augmentation. Even a category in the downloadable table would not tell a reader whether a particular employer will buy HVAC controls, whether they will reduce service calls, or how a local technician’s role will change. The measure compares exposure signals across occupations; it does not record a workplace decision. Whether controls change technician work requires evidence beyond a product description. Integration, reliability, cost-effectiveness, and responsibility for exceptions and maintenance all need evaluation. These are practical questions, not findings established by Slipstream. Its stated study aims include cost-effectiveness and operations and maintenance effects, which underscores that the project is evaluating these matters. Without completed results, the advertised functions or benefits should not be treated as demonstrated outcomes. For a technician or someone considering the trade, look for local, observable evidence: whether nearby employers use these controls, which duties they assign to them, what work remains with technicians, and whether staffing, service hours, or required skills have changed. Ask a local employer, instructor, or technician, distinguishing a demonstration from routine paid work. A controls deployment could change operating conditions or create different support tasks; it does not establish that fewer technicians will be needed. Until field results and workplace evidence show more, AI in HVAC equipment is a change to examine, not proof of technician replacement.

Sources: Join our AI-enabled HVAC controls field study; Artificial Intelligence (AI) exposure categories

Why can a risk label beside growth mislead?

A low-risk label and a positive employment projection can describe different things. Career HUD’s opened HVAC technician page displays “Minimal Risk” alongside the Bureau of Labor Statistics’ projected growth and annual openings. It is not one. The BLS Occupational Outlook Handbook reports a national employment outlook for heating, air conditioning, and refrigeration mechanics and installers; the page does not present that growth as an AI impact estimate. Career HUD’s label is its own assessment, not a classification issued or endorsed by BLS. Putting the two statements next to each other does not turn them into a BLS finding that HVAC work is protected from AI.

Concise labels have a real use: they can help a reader notice that occupations differ and decide what to inspect next. A comparison can be useful for sorting questions, provided its inputs, boundaries, and date are visible. A single label can compress several questions, including whether a system could perform a task, whether workers currently use it, whether an employer has adopted it, and whether work or staffing has changed. Those are separate stages. An occupation can include tasks that technology may assist while still requiring physical service, judgment, or human accountability; a label alone does not show which tasks are affected, how much time they take, or what happens to demand. Nor does “minimal” specify the comparison group, evidence window, or uncertainty unless the rubric explains those details. A useful label should lead back to task-level evidence before it shapes a career decision.

BLS’s own supplemental AI exposure categories make that boundary explicit. The BLS AI Exposure Categories documentation describes relative categories built from five external measures: three theoretical exposure measures and two that map observed AI interactions to occupational tasks or work activities. BLS says even the latter do not directly establish AI use by workers in a particular occupation. It also says the categories are not forecasts of job loss, productivity, wages, adoption, or employment growth, and they do not distinguish augmentation from automation. The OOH growth projection answers a different question about modeled national labor demand. Neither product, alone or beside the other, establishes a worker’s outcome.

The International Labour Organization’s Generative AI and Jobs: 2025 Update and the OECD AI Exposure Measure help clarify why exposure frameworks should not be read as labor forecasts. The ILO describes a task-level analysis spanning nearly 30,000 tasks, while the OECD measure maps AI capabilities across occupational requirements. These frameworks examine relationships between capabilities and work; they are not observations of HVAC technicians’ tool use, local adoption, or job changes. Their presence in a broader discussion cannot validate Career HUD’s label or supply an HVAC-specific result. They help explain measurement, not settle this occupation’s outlook.

What would change this reading? First, an inspected HVACR row in BLS’s current supplemental table would establish which relative category BLS assigns, though that would remain an exposure signal rather than an employment forecast. Second, credible longitudinal evidence from HVAC workplaces would need to connect actual adoption with changes in tasks, hours, hiring, wages, or staffing. The sources here do not provide that evidence. Until it exists, the defensible conclusion is narrower: BLS projects positive national demand for the occupation over 2025–2035, while the AI-specific workforce outcome remains unresolved. A “minimal risk” label may invite a useful question, but it cannot guarantee safety or quantify an individual’s chance of displacement.

Sources: HVAC Technician Salary & Career Outlook (2026); Artificial Intelligence (AI) exposure categories; Generative AI and Jobs: A 2025 Update; The OECD AI Exposure Measure

What should a worker check next?

The BLS projection is a national starting point, not a decision about your household or a particular employer. If you are considering HVACR, check openings and pay where you can realistically work, then compare the training and licensing route with the time, cost, schedule, and physical demands you can manage. The Occupational Outlook Handbook describes the occupation and its national outlook; it cannot tell you whether a local program fits your circumstances or nearby employers are hiring. If you already work in the trade, write down one recurring task that takes time, such as finding diagnostic information or preparing a service record. Check whether your employer permits a tool for that task, what information it may receive, and how you would verify its output before acting. Treat a trial as a way to learn about that workflow, not evidence that AI is changing the whole occupation. O*NET's HVACR task descriptions place information work alongside inspection, repair, and customer communication, but do not describe your personal workweek. A useful next conversation is with a local training provider, workforce office, or HVAC employer: “What does entry require here, what does the schedule involve, and what should I verify about day-to-day tasks?” If you are mapping current duties, the free task checker can organize change-pressure signals; its result is not a probability of job loss. Revisit your choice if local demand, documented workplace use, or BLS projection assumptions change.

Sources: Heating, Air Conditioning, and Refrigeration Mechanics and Installers: Occupational Outlook Handbook; Occupational Projections and Characteristics; Heating, Air Conditioning, and Refrigeration Mechanics and Installers

Questions readers ask