STEM Career Projections: Which Fields Still Need Humans in 2035

When parents think about STEM as a safe career path, they’re working from an assumption that’s roughly ten years out of date. In 2012 or 2015, “go into STEM” was fairly solid general advice: the fields were growing, pay was high, and automation was a manufacturing problem, not a knowledge worker problem. The picture in 2026 is more complicated. Some STEM subfields are growing faster than ever. Others face significant automation pressure — specifically the parts that involve pattern recognition, data analysis, and structured code generation, which are exactly what AI tools have gotten good at. A parent helping a 14-year-old think about STEM careers needs a more precise tool than “STEM is good.” What’s good, specifically, and why?

Key Takeaways

• The U.S. Bureau of Labor Statistics projects STEM occupations overall will grow 10.4% from 2022 to 2032, roughly twice the rate for all occupations — but growth is highly uneven across subfields.
• AI automation risk within STEM is task-level, not job-level: software engineering jobs aren’t disappearing, but the routine coding tasks within them are being absorbed by AI tools, shifting what engineers are paid to do.
• The fastest-growing STEM subfields through 2035 are: information security (32% projected growth), data science (35%), and environmental engineering (6%, but from a large base and growing steadily).
• The STEM skills with the broadest cross-field transfer are systems thinking, quantitative reasoning, and the ability to specify problems precisely — all of which become more valuable as AI handles more routine analysis.
• “STEM” in the AI era increasingly means working with AI tools as a native instrument, not competing against them — the relevant skill is knowing what AI can’t do.

BLS Projections by STEM Subfield

The Bureau of Labor Statistics publishes occupational projections every two years in the Occupational Outlook Handbook (bls.gov/ooh). The 2022–2032 projections, the most recent complete set, show wide variation across STEM fields.

STEM Field	2022 Employment	Projected Growth 2022–2032	Median Annual Wage (2023)	AI Automation Risk
Information Security Analysts	168,900	+32%	$120,360	Low (adversarial, creative)
Data Scientists	168,900	+35%	$108,020	Moderate (tools absorb routine analysis)
Software Developers	1,847,900	+25%	$130,160	Moderate (AI assists but doesn’t replace architecture)
Civil Engineers	329,200	+5%	$95,890	Low (physical world, regulatory, contextual)
Biomedical Engineers	22,700	+10%	$99,550	Low (cross-domain, regulatory)
Environmental Engineers	57,800	+6%	$96,530	Low (fieldwork, site-specific, regulatory)
Statisticians	44,800	+31%	$104,860	Moderate (AI handles routine stats, humans handle design)
Nuclear Engineers	18,900	+1%	$125,290	Low (safety-critical, regulatory)
Computer Hardware Engineers	74,000	+5%	$138,080	Low (physical design, silicon constraints)
Actuaries	28,600	+23%	$120,000	Moderate (AI speeds analysis; judgment layer stays human)

Sources: BLS Occupational Outlook Handbook, 2022–2032 projections; BLS Occupational Employment and Wage Statistics, May 2023.

What AI Actually Automates in STEM — and What It Doesn’t

The AI automation risk column in that table requires unpacking. The risk isn’t that AI will replace engineers; it’s that AI will absorb specific tasks within engineering jobs, concentrating human value in the remaining tasks — which tend to be harder, higher-stakes, and more poorly defined.

In software engineering: AI tools (GitHub Copilot, Claude Code, Cursor) now handle a substantial fraction of routine code generation, boilerplate, unit testing, and documentation. The research consensus is that this increases developer productivity rather than reducing headcount — each developer can produce more — but it shifts what engineers are paid for. The routine, unambiguous coding task is commoditizing. Architecture decisions, cross-system integration, understanding what a business actually needs, debugging unexpected failures in production — these remain stubbornly hard for AI and are consequently where engineering value is concentrating. For more on this shift, see our detailed analysis of how AI is changing the software engineering job.

In data science: Similar pattern. AI tools handle routine data cleaning, visualization, and model training on standard datasets competently. What remains human: problem formulation (deciding what to measure and why), data collection design, interpretation of results in business context, communication to non-technical stakeholders. A data scientist who only knows how to run models is in a more precarious position than one who knows how to frame the question the model should answer.

In civil and environmental engineering: The physical world creates a natural AI buffer. A structural engineer assessing soil stability on a specific site, navigating local regulatory requirements, managing contractor relationships, and making liability-bearing decisions operates in a context that is too specific, too consequential, and too embedded in physical reality for current AI to handle. This doesn’t mean AI adds nothing — design software, simulation tools, and code-checking automation are all improving — but the core judgment layer stays human.

In information security: Uniquely, cybersecurity is adversarial — defenders are constantly adapting to attacker behavior, which is itself adaptive. This creates a moving target that AI can assist with (pattern detection, anomaly flagging, log analysis) but cannot fully automate because the attack surface evolves in response to defenses. Human creativity and adaptive thinking are structurally necessary, not just temporarily necessary.

Which STEM Skills Transfer Across Fields in the AI Era

This is the question that matters most for a 13-year-old deciding what to invest in learning. The answer from both the career research and the AI task-analysis literature is consistent:

Systems thinking. The ability to model how components interact, how changes propagate through a system, and how to identify leverage points is broadly valuable and poorly replicated by current AI. A civil engineer thinking about traffic flow, a software architect thinking about microservice dependencies, an epidemiologist modeling disease transmission — all are using the same underlying cognitive skill.

Quantitative reasoning. Not just knowing statistics, but knowing when to use which statistical tool, recognizing when a finding is an artifact of measurement rather than a real effect, and communicating uncertainty honestly. AI tools can run statistical tests faster than any human; they cannot reliably choose the right test or notice when the data collection design makes the result uninterpretable.

Precise problem specification. This is increasingly the core STEM skill of the AI era. AI models are powerful but literal — they do what they are asked to do, not what you want. The ability to specify a problem with precision, identify the assumptions embedded in a formulation, and notice when a result solves the stated problem but not the actual problem is the skill that makes an AI-era STEM professional valuable.

Physical world grounding. Fields that require engagement with physical systems, real materials, real environments, and the messiness of manufacturing or construction have a structural advantage against automation. This is part of why hardware engineering, civil engineering, environmental science, and biomedical engineering face lower automation pressure — the physical world doesn’t run on clean, well-labeled data.

What “STEM” Means in the AI Era

The conventional STEM boundary — science, technology, engineering, mathematics — is becoming less useful as a category. What’s more predictive of career durability is a different distinction:

STEM work that involves clear, specifiable, pattern-based tasks (writing boilerplate code, running standard analyses, generating reports from structured data) faces meaningful automation pressure.

STEM work that involves ambiguous problem framing, physical world constraints, regulatory and ethical judgment, and human communication faces much lower pressure. These are the tasks where AI tools are assistants, not substitutes.

The relevant question for a teenager isn’t “is STEM safe?” but “which parts of STEM work involve judgment, context, and ambiguity that AI handles poorly?” That’s where human contribution concentrates and where compensation follows.

For a broader look at how AI is reshaping career paths, see our overview of the AI talent gap and what it means for kids entering the workforce, and for a perspective on which careers in general will require more human workers, see our future jobs in the AI era guide.

What to Watch For Over the Next 3 Months

If you’re guiding a teenager who is interested in STEM:

Month 1: Help them identify not just the STEM field but the specific type of work within it. A teenager who says “I want to work in tech” should be able to say, after some exploration, “I’m interested in the security side of tech” or “I’m interested in hardware design” — something more specific than the field label.

Month 2: Look for early exposure to the physical vs. digital distinction. If a teenager interested in engineering has only done software projects, find a way to get them working on physical systems — electronics, mechanical design, environmental measurement. Hands-on physical problem-solving develops a different category of skill than screen-based work.

Month 3: Track whether their interest is in using AI tools or in understanding the systems those tools operate within. Both are valid, but the research suggests that understanding the underlying system (the math, the physics, the biology) provides more durable career protection than using the current best tool.

Frequently Asked Questions

Is a computer science degree still worth it given AI advances?

Yes, but the value is shifting toward the layers of CS that AI handles least well: systems architecture, hardware-software interfaces, distributed systems design, and security. A CS graduate who only knows how to write application code is in a weaker position than one who understands how systems work at multiple levels of abstraction. The degree is still worth it; the question is which courses within the degree to take seriously.

Which STEM fields are most resistant to AI automation through 2035?

Based on BLS projections and AI task-analysis research, the most resistant fields are: information security (adversarial, creative), civil and environmental engineering (physical world, regulatory), nuclear engineering (safety-critical, regulated), biomedical engineering (cross-domain expertise, regulatory), and hardware engineering (silicon and physical constraints). Fields with moderate exposure include software development, data science, and statistics — not because jobs disappear but because the human contribution concentrates in the harder, less routine work.

Should my child focus on STEM or on the intersection of STEM and other fields?

Both are legitimate, but the intersection is often more valuable for long-term career resilience. A biomedical engineer who also understands the healthcare system, regulatory environment, and patient communication is harder to replace than a pure specialist. A data scientist who also understands the business domain she’s analyzing produces more value than one who only knows the technical tools. The STEM skills become most durable when they are embedded in real-world context.

What STEM skills should kids develop now that will still matter in 2035?

The most transfer-proof STEM skills are: systems thinking (modeling how parts interact), quantitative reasoning (knowing what numbers mean and when they’re wrong), precise problem specification, physical world engagement (understanding how materials and systems behave outside controlled conditions), and cross-domain communication (being able to translate technical findings to non-technical decision-makers). These are all notably hard for AI to replicate.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. U.S. Bureau of Labor Statistics. (2024). Occupational Outlook Handbook, 2022–2032 Edition. https://www.bls.gov/ooh/

2. U.S. Bureau of Labor Statistics. (2024). STEM Occupations: Past, Present, and Future. https://www.bls.gov/spotlight/2017/science-technology-engineering-and-mathematics-stem-occupations-past-present-and-future/home.htm

3. Acemoglu, D., & Restrepo, P. (2022). “Tasks, Automation, and the Rise in US Wage Inequality.” Econometrica, 90(5), 1973–2016. https://doi.org/10.3982/ECTA19815

4. McKinsey Global Institute. (2023). The Economic Potential of Generative AI: The Next Productivity Frontier. https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-potential-of-generative-ai

5. Autor, D., Levy, F., & Murnane, R. J. (2003). “The Skill Content of Recent Technological Change: An Empirical Exploration.” Quarterly Journal of Economics, 118(4), 1279–1333. https://doi.org/10.1162/003355303322552801

6. National Science Foundation. (2023). Science and Engineering Indicators 2024. https://ncses.nsf.gov/pubs/nsb20241