Conversation starter: are AI jobs booming or overhyped?

Key findings

Opinions vary widely about how much AI will transform the labor market, ranging from an immediate job apocalypse for some roles to the slow diffusion of another tool that may change our work in the long run, but with limited impact in the next few years. In our view, it is still too early for labor market data to show roles being eliminated by AI. Many trends, such as early career unemployment, were underway before LLMs came on the scene. That said, some hiring trends are starting to emerge: there are new AI jobs and specializations, and companies are paying a premium for AI skills.

AI jobs: brand new or just specialized?

AI jobs are on the rise, though they still represent a small portion of all job postings. They more than doubled from 2023-2024 (up 123%, machine learning roles excluded) and are set to increase another 53% in 2025. In job postings on Glassdoor, we see two distinct types of AI jobs: those that are AI-exclusive (those that would not exist without AI) and those that are AI-tuned (occupations that specialize in AI but overlap substantially with similar, non-AI roles.) Generally, in this research, we discuss AI in the context of transformer-based large language models (LLMs), though other AI jobs exist as teams continue to build and deploy other types of AI, like recommendation algorithms. 

We treat machine learning roles as separate, since by definition a machine learning engineer is working with AI though not necessarily with LLMs. Additionally, it has become much more common to list AI in machine learning titles: 46% of ML job postings did in 2025, up from 15% in 2022.

AI-exclusive jobs would not exist in a world without LLMs. These include researchers that develop new models and roles specifically attached to training these models. AI research roles are hyper-specialized and there are not a great deal of them. The market for AI researchers is like the market for professional athletes: there is a limited pool of seasoned talent, and a small number of employers compete over them with mind-boggling contracts. 

On the other end of the spectrum, AI training jobs often require specific expertise that has nothing to do with AI. For example, one recent job posting required an advanced degree in history to train LLMs on methods for historical research and to test the validity of AI-generated historical claims. These jobs are often contract or gig positions, and only a few dozen companies hire for them. These roles may be a flash in the pan even in a world where AI transforms the labor market, but they fit in the category of roles that would not exist without LLMs.

AI-tuned jobs existed before AI, and even those working directly with AI in these roles have overlapping skills, experience, and responsibilities to similar roles that do not work in AI. In other words, for these roles AI is a new specialization but not a new job. Numerically, these roles dominate AI-exclusive roles in number, representing 92% of new AI-specific roles so far in 2025.

These roles can be technical, but they do not need to be. Software engineers who deploy LLMs have similar skills and responsibilities to other engineers, but have specific experience and responsibility working with LLMs. For non-technical roles, companies are actively advertising roles for AI-focused attorneys, AI-focused recruiters, AI-focused product managers, and more. Similarly, the fundamentals of these roles are not different from their non-AI counterparts, though they do represent a new specialization. A corporation’s legal team may now include attorneys for AI, alongside the labor or patent attorneys with different specializations.

This figure shows that AI specific roles have increased dramatically since 2022, though they still represent a small overall portion of the market. Combined, they increased from fewer than 1/1000 job listings to 2/1000 listings. This increase is split between growth in machine learning roles and other AI-specific roles, with AI-exclusive roles making up an extremely small share.

AI pay premiums

It comes as no surprise that AI roles tend to pay more than non-AI roles: jobs with AI in the title advertise base salaries around $121k on average, compared to $64k for all non-AI roles. However, many AI-specific roles are already in higher paying professions - so the real question is how much more AI pays relative to a similar role. We call this the AI premium, and the median AI premium is 25%, though it varies from occupation to occupation.

Many of these premiums are for technical jobs: AI backend engineers earn 45% more than other backend engineers, and AI database administrators earn 38% more than other database administrators. However, not all roles with AI premiums are technical. Job postings for AI attorneys list salaries 38% higher than other attorneys, and roles for recruiters specializing in AI talent earn 30% more than other recruiters.

Additionally, not all technical roles have an AI premium. While nearly half of machine learning engineering job listings have AI in the title, pay in these roles is similar with or without the explicit AI specialization. Similarly, AI data scientist listings have pay ranges similar to other data science listings. One potential explanation is that these roles are already highly specialized and AI-related, so having AI in the title may be redundant and AI premiums are already baked in.

Engineering occupations have wide variability in AI premiums. Machine learning engineers and full stack engineers have lower advertised salaries when AI is in the title, and there is no meaningful AI premium for data engineers. AI-focused roles are associated with sizable premiums for front end engineers (though this is the lowest-paid group of engineers) and application engineers.

Conclusion

In terms of transforming the labor market, the AI hype curve is far ahead of the change curve, but two trends are clear: companies are hiring more talent that specializes in AI, and they are willing to pay a premium for that talent. Most of these AI jobs existed before LLMs came on the scene, and AI-specific knowledge complements the experience and skills these workers already have.

Methodology

We analyzed job postings on Glassdoor from 2023 to July 15, 2025. AI-exclusive roles are job titles containing either terms like “AI Training”, or with “AI” or “Artificial Intelligence” in research occupations. AI-tuned roles are roles with those terms in other job groups. Job listings are grouped with similar roles based on Glassdoor’s internal taxonomy. Pay rates on job listings are either included by the employer or estimated by Glassdoor using available data.