On one global hiring platform, the share of filled jobs that explicitly require AI skills went from 5% in 2022 to 15% in the first half of 2026. It is no longer an engineering matter: 17% of marketing coordinator roles and 15% of recruiter roles now ask for it. What has not moved is the paperwork. Of the companies that added AI requirements to existing roles in the past year, 62% changed neither the title nor the pay.
The work moved before the job description did
The job description is where a company decides what a role is worth, so a job that has changed without it being rewritten is still paid as the old job. That is the situation for many people. Their manager expects them to use AI, and their written duties still describe the work as it was in 2021.
It also explains who is affected. AI requirements show up in 11% to 13% of individual contributor roles and only 8% to 9% of director, vice president and executive roles. The people doing the new work are not the people who rewrite job descriptions or set pay bands, so nothing forces the update.
Why the pencil result matters
The pencil study is easy to laugh at, but it makes a serious point. Economists had read the computer premium as a reward for a skill. The pencil result suggested that the pay difference came from the kind of job a person held, and that the jobs given computers were already the better ones. A tool that spreads through a company does not pay anyone for using it, because an employer pays for what the job produces and how scarce the person doing it is.
AI fits that pattern closely, with one difference. Computers came in with new equipment and often new kinds of jobs. AI is arriving inside old ones, and 53% of companies say they prefer to train their existing staff when skills are missing. The skill is built on the job, by people whose pay was set before it was needed.
What it looks like for one recruiter
Take a recruiter at a mid-sized distributor. Her description says she screens résumés, books interviews and keeps the hiring tracker up to date. Over the past year she has started using an AI assistant to summarize applications, draft outreach and prepare interview questions, and she now handles more open roles than before.
If the assistant saves her five hours a week, that is about 250 hours a year, roughly six working weeks. Nobody has written those six weeks down, so the company has simply received them. She has no new title to point to, and the next recruiter hired will be advertised as an "AI-enabled recruiter" at a higher rate than she earns.
That last point is where I expect this to go. Pay will first reach people who are hired into roles that name AI, because those roles come with a label that can be priced. Staff who picked it up quietly will either get their job rewritten or find that the market pays more for the same skills elsewhere. Companies that rewrite roles early will keep their best AI users; the rest will train them and then lose them.
What to do with it
A manager can start by asking what AI work already happens in the team and writing it into the job. An employee can do the same from the other side. Write down three things you now do with AI that your old description does not cover, with the hours or the extra volume beside each, and bring it to your manager as a proposed new description. It is easier to approve a rewrite that already exists than to answer a request for a raise.