The Republic is a journal published by The Palantir Foundation, featuring commentary and essays on technology, national security policy, and international affairs.

A New Paradigm for Technical Talent 

Marvin Beerbohm, Automotive Industry (mural, Detroit Public Library), 1940. Smithsonian American Art Museum.

A New Paradigm for Technical Talent 

The inability to accurately assess the health of U.S. AI talent pipelines has significant implications for national security and future U.S.-led innovation.

By

Dr. Diana Gehlhaus is a Principal Federal Policy Strategist at the MITRE Corporation. Dr. Gehlhaus’ work focuses on the intersection of AI, economics, workforce development, and organizational change.

AI Workforce Policy as an Economic and National Security Imperative

A procurement request triggers a chain of events. Within seconds, one AI agent begins drafting requirements while a second queries suppliers, a third negotiates pricing, and a fourth flags regulatory risks in real time. In minutes, a contract is ready for human approval. This is not a distant scenario. Variants of such agent-to-agent workflows are already emerging across industries from finance to entertainment, energy, and defense.

The launch of ChatGPT in November 2022 was only the beginning of a new industrial age. Agentic workflows with humans above the loop, supervising and orchestrating agent-led tasks, may soon be normal business. The next few years could see the rise of the Agent Economy, defined by agent teams running entire processes, conducting transactions, and executing the design and development of entirely new goods and services. It could spark entirely new industries, regulatory frameworks, and reimagined industrial organization.

This creates a high-stakes imperative for leading and competing in AI with major implications. Who leads in AI is likely to hold outsized economic power, geopolitical influence, and strategic decision advantage in conflict. That makes the U.S. AI workforce, and its competitive posture, foundational to future economic growth and national security.

The Challenge of Defining the U.S. AI Workforce

Understanding the U.S. AI workforce requires a clear definition and clean data. However, five years into the AI revolution, there remains no formal or consensus-based definition. Instead, the existing literature on the U.S. AI workforce provides a range of definitions that are subjective, imprecise, and range widely. This is partially due to interpretation and partially due to limitations in available workforce data.

One of the first efforts to define the AI workforce was in 2019, when the RAND Corporation assessed the then Department of Defense’s AI posture.1 The research found no consensus on what AI was, let alone what constituted AI talent or workforce data. Another early effort to categorize AI talent came from the National Security Commission on AI (NSCAI) in its Final Report.2 It documented similar challenges and provided broad occupational categories.

The Center for Security and Emerging Technology (CSET) was among the first to comprehensively describe and characterize the U.S. AI workforce in 2021.3 It defined AI talent as those involved in the design, development, and deployment of AI. Since then, efforts from the National Academy of Sciences, Stanford HAI, Macro Polo, and others have focused largely on AI researchers as a key driver of innovation. For example, Stanford’s HAI compiles an annual index of AI human capital metrics encompassing supply and demand from government and proprietary sources, while Marco Polo considers publications and citations in AI research.4,5

All efforts to understand the U.S. AI workforce use existing occupational taxonomies which are outdated, lack sufficient granularity, and are slow to change. The result is that there is no consensus on how to clearly define and identify AI talent. For example, in an attempt to create an AI occupational series to better measure and assess this workforce in the government, the Office of Personnel Management (OPM) declined, citing it was too intertwined across existing series to be practical.6

The Evolution of the U.S. AI Workforce

The imperative for economic competitiveness and national security only makes the need to clearly define and identify AI talent more urgent. You cannot secure and strengthen a workforce that you cannot measure.

While the AI policy research community has not aligned on what constitutes the AI workforce, the AI workforce has not stood still. Since the start of the current AI revolution, what constitutes AI talent has shifted. That is, the knowledge, skills, abilities, and tasks that are required to design, develop, and deploy AI capabilities has and continues to evolve.

Moreover, today’s AI workforce spans both technical and non-technical roles, reflecting a shift in importance from research to deployment as AI capabilities diffuse across the economy. Now, those involved in the architecture, deployment, and enablement of AI tools, platforms, and systems are just as important as AI researchers.

This shift is reshaping work inside organizations, with more individuals moving into roles involved with the deployment and integration of AI capabilities in their domain or functional area (AI+X). While frontier talent remains concentrated in leading labs and firms, most organizations now require in-house AI+X capabilities to effectively integrate, deploy, and scale AI systems.

Attempts to be more proactive to define the AI workforce, at least the technical non-PhD core, are already happening in the Department of War (DoW). The January 2026 DoW AI Strategy called for the creation of service-level AI Talent Plans that includes both military and civilian personnel.7 The U.S. Army took this one step further, creating a uniformed AI/ML career field that defines roles and responsibilities within a broader AI talent ecosystem.8

The Current U.S. AI Workforce

It is time to clearly define and measure the full U.S. AI workforce in a way that enables rigorous analysis and informs targeted policy. The roles and responsibilities involved in the design, development, and deployment of AI are now interconnected. Today’s AI workforce can be divided into three core human groups:

1. Researchers (design). This group consists primarily of PhD-level electrical engineers and computer scientists who drive foundational AI advances. Often referred to as top-tier AI talent, they are highly visible due to scarcity and intense competition and represent the most traditional and easiest-to-measure segment of the AI workforce. However, they are only a small share of the total workforce and are increasingly bifurcated in quality, with a limited subset commanding outsized influence and compensation.

2. Developers (develop and deploy; AI+X). This is the largest and most diverse segment, comprising software developers, data scientists, engineers, cloud architects, technical product managers, and others who build, deploy, and maintain AI systems. These professionals combine AI expertise with domain knowledge (AI+X) and are responsible for activities such as system integration, infrastructure deployment, testing, security, and product management. The group is difficult to define using traditional occupation-based taxonomies due to variation across industries and roles, and evolving skill demands are expanding entry pathways beyond traditional STEM credentials.

3. Enablers (integrate and scale; technical and non-technical). This group includes both technical and non-technical professionals—such as IT specialists, UX designers, acquisition professionals, legal experts, organizational psychologists, and trainers—who facilitate the adoption and scaling of AI within organizations. Distinguished by their role in implementation and organizational integration, enablers ensure AI systems are usable, compliant, and effectively embedded into workflows, making them essential.

Today’s AI workforce is not a monolithic group of researchers. Arguably, the hardware and software engineers designing and building the AI stack for compute, storage, and systems integration are as important as AI researchers. It is not even solely technical talent. It is multi-disciplinary, with roles and responsibilities across interconnected categories. What constitutes the current AI workforce suggests a movement from “talent pipeline” as a singular idea to an interconnected system that comprises multiple talent pipelines.

The AI workforce and associated workforce dynamics will likely be determined by how these three groups interact, not just their individual supply. The fact that AI tools are perhaps affecting technical talent could provide value insights for how and the pace at with AI will spark the natural cycle of creation destruction, the question most top of mind to economists.

For example, emerging trends in this workforce could also shed valuable insight into AI and the future of work more generally. Already there is evidence of two related yet divergent trends worth monitoring: (1) acute skill-biased technological change that bifurcates technical talent and (2) the democratization of technology skills through GenAI tools. Skill-biased technological change may be far more nuanced in the future than what emerged from the Sputnik era STEM revolution. This is evidenced by the strong demand for an elite group of top-tier professionals alongside weakening demand generally for technical talent.9 Simultaneously, with low/no code tools, anyone with aptitude can learn to code and build agents. Both trends call into question the definition and types of “technical talent” and what the talent pipeline(s) should include. There is a real risk of labor market demand segmenting in new ways that current STEM-based education and workforce policy are not equipped to address.

Better understanding the human AI workforce could also help inform the machine “AI-workforce” and associated policy implications. The rise of agents and agent-to-agent workflows are increasingly able to handle multi-step complex tasks without human intervention, which will inevitably have significant implications for technical and non-technical human work. For example, agents can conduct research from multi-modal sources across domains, write, test, and deploy code across environments, analyze data and build reports from cross-cloud and platform connections, detect and circumvent fraud and cyber threats, and more. The nature and degree of human-machine teaming will be important to consider with the augmentation versus automation dimension of this workforce, just as with the broader workforce.

Assessing the AI Workforce: A Skill-Based Framework

Today’s workforce cannot be accurately measured through a set of occupations. Rather, it is increasingly skills-based, technical and non-technical. Unfortunately, existing workforce measurement frameworks and taxonomies have not kept pace.

Federal workforce data systems remain too coarse, occupation-based, and slow-moving to capture the realities of today’s AI workforce. Occupational taxonomies from the Bureau of Labor Statistics (BLS), Census Bureau, Office of Personnel Management (OPM), and Employment and Training Administration (ETA) were designed for standardization, not for rapidly evolving, highly specialized technical roles. As a result, AI and emerging technology workers are often misclassified, inconsistently categorized, or aggregated into vague “miscellaneous” groupings. Even ETA’s O*NET, the gold-standard for task-level insights, cannot adequately capture variation within occupations or across sectors. It is limited in understanding the pace and nature of change for today’s interdisciplinary AI work.

To compensate, analysts increasingly triangulate federal data with proprietary sources like ADP, LinkedIn, Lightcast, and Revelio. While these datasets offer more timely and granular signals, they inherit many of the same structural limitations by mapping back to federal taxonomies while introducing new challenges around access, representativeness, and data quality.

Currently available skills-based taxonomies also do not meet the mark in current form. Many are generally subjective, inconsistently granular, have overlapping categories, and are generally hard to interpret for policy. This also makes understanding the nature of AI-related skills-biased technological change a challenge, limiting the timely ability for educators and policymakers to respond dynamically to labor market needs and realities.

There is a third option to assess today’s AI workforce: Try a hybrid approach. A workforce framework doesn’t need to be black and white, occupation or skills. Why not have a hybrid approach that spans both? For example, take existing investments in emerging technology workforce data collection and build:

  • Dynamic occupation modeling and processes to upgrade federal taxonomies.
  • Industry-driven work roles that have a federal governing board and clearly defined tasks, knowledge, skills, and abilities (KSAs), and qualifications.10  
  • Real-time AI workforce data dashboard that is crowd sourced as an industry consortium.

The cybersecurity workforce provides a clear precedent for modernizing AI workforce measurement through a hybrid, work role–driven framework. Public-private partnerships such as the NICE Framework and the Defense Cyber Workforce Framework demonstrate how industry-aligned roles, standardized tasks, and competency definitions, paired with public-private data platforms like CyberSeek, can produce actionable, real-time insights on workforce supply and demand.11,12,13 Building on this model, a similar approach for data and AI would integrate dynamic occupation updates across federal occupation taxonomies.

Such a hybrid framework would strengthen the competitiveness of the U.S. AI workforce by moving our understanding of this critical group from one that is static and imprecise to one that is dynamic, granular, real-time, and skills based. If the official taxonomies evolve, the research community will help with workforce assessment and gap identification will evolve as well.

Toward a Future-Ready AI Workforce Framework

The need to clearly define and identify the U.S. AI workforce is an economic and national security imperative. Just as AI is evolving, so too is the AI workforce, and our ability to measure the strength of this workforce must also adapt to accurately assess its competitive posture.

However, current assessments of the U.S. AI workforce are imprecise, relying on outdated workforce frameworks designed for the STEM era. AI talent does not fit clearly into current occupational or skills-based taxonomies, or into traditional definitions of STEM.

That there continues to be no consensus on defining the AI workforce (e.g., AI researchers to anyone involved in the AI stack) should set off an alarm and be a call to action. Without a clear understanding of this workforce, we are left with inconsistent assessments and policy design, affecting an array of issues from education to federal hiring. Lumping AI workforce policy into STEM policy is simultaneously too narrow and too broad.

The time is now to better secure and strengthen the U.S. AI workforce. No longer can AI workforce policy focus solely on top-tier talent if we are serious about economy-wide diffusion. Today’s AI workforce is an ecosystem of talent pipelines that must be considered jointly.

Instead of continuing the current path of imprecise occupational or skills measurement, this essay suggests a third approach: an adaptive hybrid occupation and skills-based AI workforce framework. Such a framework would be aligned with the AI Action Plan, could emphasize the technical and non-technical competencies required to be part of the AI workforce, and allow for more depth and granularly in accurately assessing U.S. AI competitiveness. It could also enable rapid iteration and replicability for other emerging technology workforces.

Policymakers would be well-served to move to a more modern interpretation of the AI workforce. It could provide opportunities to encourage multiple pathways into tomorrow’s high-skill, high-wage jobs and enable generational mobility. And perhaps most importantly, it could improve policy mechanisms for more targeted design and effective intervention, informed by the speed and scale of a new type of skill-biased technological change.

The author’s affiliation with The MITRE Corporation is provided for identification purposes only, and is not intended to convey or imply MITRE’s concurrence with, or support for, the positions, opinions, or viewpoints expressed by the author. Approved for Public Release; Distribution Unlimited. Public Release Case Number 26-1039.

  1. Tarraf, Danielle C., William Shelton, Edward Parker, Brien Alkire, and Diana Gehlhaus. The Department of Defense Posture for Artificial Intelligence: Assessment and Recommendations. Santa Monica, CA: RAND Corporation, 2019. ↩︎
  2. National Security Commission on Artificial Intelligence. Final Report. Washington, DC: NSCAI, 2021. ↩︎
  3. Gehlhaus, Diana, and Santiago Mutis. The U.S. AI Workforce: Understanding the Supply of AI Talent. Washington, DC: Center for Security and Emerging Technology, Georgetown University, 2021. ↩︎
  4. Stanford Institute for Human-Centered Artificial Intelligence. AI Index Report 2026. Stanford, CA: Stanford University, 2026. ↩︎
  5. MacroPolo. The Global AI Talent Tracker. Chicago: Paulson Institute, updated 2025. ↩︎
  6. U.S. Office of Personnel Management. The Artificial Intelligence Classification Policy and Talent Acquisition Guidance — The AI in Government Act of 2020. Washington, DC: OPM, 2024. ↩︎
  7. U.S. Department of Defense. Artificial Intelligence Strategy for the Department of War. Washington, DC: U.S. Department of Defense, January 9, 2026. ↩︎
  8. Cruickshank, Iain, Nicole Curtis, and Chris Eastburg et al. “Making AI Operational: The Army’s New Technical Career Path.” Phalanx 58, no. 4 (Winter 2025): 30–41. ↩︎
  9. Other factors are likely confounded with the impact of AI, but AI could be accelerating this trend. For example: The Economist. “The Tech Jobs Bust Is Real. Don’t Blame AI (Yet).” The Economist, April 13, 2026.  Here “general” includes talent with a limited set of technical skills (e.g., not AI+X) who are not considered exceptional in their craft, lack proven adaptability or initiative to learn new tools or techniques, or lack non-technical skills that allow them to take on multiple roles within an organization. ↩︎
  10. Work roles are skills-based functions or tasks performed by a worker that is more granular than a position or occupation (e.g., an occupation can be associated with multiple work roles). ↩︎
  11. National Initiative for Cybersecurity Education (NICE). “NICE Workforce Framework for Cybersecurity.” National Institute of Standards and Technology. Updated 2026. ↩︎
  12. U.S. Department of Defense, Office of the Chief Information Officer. “DoD Cyber Workforce Framework (DCWF).” Washington, DC: U.S. Department of Defense. ↩︎
  13. CompTIA, Lightcast, and National Initiative for Cybersecurity Education (NICE). “CyberSeek.” Accessed May 2026. ↩︎