AI Skills: Pay Premium Today, Baseline Tomorrow?
When a skill is scarce, employers pay extra for it. When it becomes common, the extra pay disappears and the skill becomes a condition of entry. New 2026 survey data suggests AI fluency is being priced somewhere between those two states, and hiring managers who ignore the shift risk both overpaying and under-hiring.
What the surveys show
Payscale’s AI workforce impact preview is based on surveys fielded in August 2026: 500 HR, compensation and talent acquisition leaders at organizations with 100 or more employees, plus 1,000 full-time workers and job seekers. The full report is due in November. The preview reports:
- 58% of employers reward AI skills with pay premiums, either now or within the next 12 months.
- 23% first paid a premium for AI skills but now treat AI fluency as a required baseline.
- 74% plan to invest in AI upskilling in the next 12 months.
ERE’s coverage of the iCIMS and Payscale data adds that Payscale describes a “looming talent crisis,” with 40% of organizations saying they cannot find candidates with the AI fluency they need.
On the candidate side, the iCIMS survey of 1,000 U.S. adults, commissioned by iCIMS, found that 42% say an employer offering AI training would be more attractive than a similar employer that does not, and 14% would accept a lower salary for a job that offers it. It also found that 47% had taken steps to build AI skills in the past six months, and 30% were self-taught in AI, up from 22% a year earlier.
Note who funded what. iCIMS sells recruiting software and commissioned its own survey; Payscale sells compensation data. Neither is neutral, though both publish their sample sizes.
Two different skills hiding under one word
The data point to a split. ERE’s piece reports that using generative AI tools such as ChatGPT or Copilot was the only AI skill reported by more than half of workers. And iCIMS’s Trent Cotton says most people use AI tools in a basic way, while employers want more specialized skills.
So “AI fluency” covers at least two things:
- Tool use, which is becoming universal and, for 23% of employers, already unpaid because it is expected.
- Building and operating AI systems: evaluation, retrieval, fine-tuning, deployment, monitoring. This is scarce, and it is what a premium is most likely to attach to.
If your job description lumps the two together, you will get applicants who are fluent in the first and unqualified for the second, and a salary band that fits neither.
What this means for hiring managers
- Decide which skill you are buying. Write the requirement as tasks: “evaluates model outputs and ships changes based on the results,” not “AI proficiency.” Candidates can respond to that, and screeners can test it.
- Do not pay a premium for the baseline. If everyone on the team uses a coding assistant, that is a working condition. Reserve premium pay for the specialized skill.
- Treat training as part of the offer. The iCIMS figures suggest it moves candidate preference, and ERE reports Cotton’s view that training is a recruiting tool, not just a retention benefit. A small share would trade salary for it, so do not rely on that as a discount.
- Reskill before replacing. Cotton argues employers should reskill and redeploy workers rather than dismiss them for lacking skills. Late-October research on “AI rebounding,” covered in our myth-buster, suggests replacements can be costly to reverse.
What this means for candidates
Basic tool use will not distinguish you for long. Build evidence of the narrower skill: a project where you evaluated a model, built a pipeline, or ran a system in production. Then look for roles where that skill is the job, such as ML Engineer or LLM Engineer.
The takeaway
Expect the premium to attach to specific capabilities and the baseline to rise. Define precisely what you need, pay for scarcity, and train for the rest.