The compensation review no one's running

61% of companies have rewritten job descriptions this year to add AI skills, yet 55% of them haven't changed the pay attached to those roles at all, according to Payscale's 2026 Compensation Best Practices Report.

That gap has a name: comp drift. A role gets priced once, against a written job description, then reviewed on a fixed schedule that assumes the job holds still in between. AI breaks that assumption: the content of a role can now shift within a single quarter, while the review built to catch that shift still runs on the old annual clock. The job outgrows the description, and the price tag stays put.

That mismatch is easy to miss, because at the aggregate level, pay looks calm. Employers are projecting a 3.2% merit increase and a 3.5% total salary increase for 2026, the third consecutive year of relative stability, per Mercer's 2026 Compensation Planning Strategy. A steady national average says nothing about whether any individual role is still priced for the job it actually is now.

When AI changes what a role requires faster than compensation systems are built to track, what does that gap actually cost an organization, and what does it take to close it before the cost shows up as an exit, a failed hire, or a fairness problem nobody planned for?

The job outgrew its own description

AI already accounts for 42% of the code developers commit, a share expected to reach 65% by 2027, per Sonar's 2026 State of Code Developer Survey of more than 1,100 professional developers.

Two years ago, "engineer" on Philip Ruffini's team meant someone who wrote code steadily through the day. Today, the same title means someone who directs AI tools to produce most of that code, while spending their own judgment on architecture and on catching what the model gets wrong. "The work moved from output to oversight," says Ruffini, CEO and co-founder of Hire Overseas, who prices and levels every engineering role his company staffs. "Oversight's worth more, and the job description was still describing the 2022 [version of the role]." Enrollment in critical-thinking coursework – the exact skill Ruffini described – grew 101% year over year among software and product learners, as shown in Coursera's 2026 Job Skills Report.

Comp drift isn't confined to tech. 70% of the skills used in most jobs will change by 2030, largely because of AI, according to LinkedIn's Work Change report, drawn from data across the platform's more than 1 billion members.

In 2022, the recruiter role on Lily Stoyanov's own team was linear, junior work, similar to what Ruffini describes: sourcing, screening, and scheduling. By 2026, the same seat was an "AI supervisor," says Stoyanov, CEO of workforce platform TFY (Transformify), which sets pay and levels roles for contractors across 184 countries, "writing and tuning prompts, auditing AI-generated shortlists for bias, judging whether a match is explainable enough to defend to a hiring manager, and owning the outcome when the model is wrong."

Within a single quarter, one SEO specialist's job at Great Lakes Tiny Homes went from running keyword tools and outreach campaigns to auditing AI-generated content and rewriting prompts, while the job description and pay band never moved. "The strategic load had doubled while the title still read junior specialist," says Joe Troyer, the company’s chief marketing officer.

In one residential care company, a caregiver with no compliance background was promoted into a newly created auditor role built around daily checks. "She went from a person doing daily checks to a person overseeing [her team's] daily audits while doing the complex audits herself," says Dr. Joshua McWealth Unamba, the organizational psychologist who runs the company, "and AI is what made that leap possible."

The fairness nobody's pricing in

Ken Herron, co-founder of VCONify, has spent more than 25 years leading marketing teams as a five-time CMO. "Organizations begin asking why two employees with the same title are delivering dramatically different levels of value," he says. "One has embraced AI and fundamentally redesigned how they work. The other is still performing the role much as they did two years ago."

According to Korn Ferry's analysis of its AI Impact Score data, two employees who share a job title can differ more than 20 times as much in how much AI has reshaped their day-to-day work as two employees in completely unrelated jobs do. "Managers recognize the difference almost immediately," Herron says. "The organization usually doesn't, because job descriptions, competency models, and compensation structures typically change far more slowly than the work itself."

That lag shows up directly in Mercer's 2026 Compensation Planning Strategy: 83% of US employers plan to distribute their 2026 salary increase budgets evenly across the organization, rather than directing more toward employees with in-demand skills. Spreading the same percentage increase across nearly everyone means two people doing very different levels of work are likely getting the same raise regardless.

Once that gap goes unaddressed, pay compression follows, according to Troyer. "I have seen pay compression show up when teammates handle different AI workloads but sit in the same band," he says. "The person doing the heavier lifting starts looking for a new title or leaves, while the lighter role stays overpaid relative to output."

Paul Martin, founder of recruiter-specific job board RecruiterRoles.com, publishes a live Recruiter Salary Report, calculated from the salary ranges employers advertise in recruiter job listings on the board. Those figures reflect what companies are offering to attract candidates. His most recent figure draws on 306 listings across 15 countries in a single 30-day window. "Seniors at $75,000 and managers at $74,856, with directors only $84,000, show pay compression as AI rewrites recruiter jobs faster than pay bands," he says. Managers are supposed to out-earn seniors, and directors are supposed to out-earn both; here, all three sit within roughly $9,000 of each other.

That gap carries a cost that starts with the employee and lands on the employer, according to Herron. "The immediate cost is usually recognition," he says. "The longer-term cost is retention. High performers begin wondering why truly extraordinary contributions aren't reflected in their role, career progression, or compensation."

The discovery costs more than the fix

Lucas Botzen, CEO of Rivermate, lost one of his best employees to a competing offer that recognized AI-enabled analytical work as senior scope; his own band still didn't. "During the exit interview it was very apparent that my employee's internal title and salary represented the type of work he was originally hired to perform," Botzen says, "versus the type of work he had become responsible for over the last year." Engagement had already been declining, he says, because the employee could see the company judging him against expectations that no longer matched what he did every day.

"When someone's real job outgrows what they're recognized and paid for, the gap registers as a chronic, low-grade threat," McWealth Unamba says. "The nervous system reads doing far more than you're acknowledged for as a kind of self-betrayal, and it runs that stress response for months before anyone gives notice."

He nearly lost the caregiver he'd promoted into the auditor role. The moment it became impossible to ignore, he says, was a conversation she asked for herself: she itemized everything she was actually doing, and laid out like that, it was clearly more than one role wearing a single title. "She stayed because she had access and I moved fast," he says, "but the burnout was already running before she spoke up. Most leaders only ever get the resignation version."

Ruffini's discovery came from people who never made it onto the team at all. "We went to fill a role at the rate our band said it was worth and kept losing the strong people," he says, "because the ones who could actually do the AI-steering version of the job knew they were worth more than we'd priced it. Losing a few good candidates back to back is what made us admit the band was describing a job that didn't exist anymore."

A lesson of deferral

Abhishek Shah, founder and CEO of Testlify, an AI-native hiring assessment platform with around 80 employees, hasn't run a meaningful review of his own job architecture since 2022. By his own estimate, that gap alone has let a third of his roles drift. "The job titles haven't changed," he says, "the work inside them has."

Shah's own explanation for why is blunt. "[Deferral is] invisible until it isn't," he says. No one is penalized for letting a job architecture review lapse, Shah says; the flag only comes later, when a retention problem shows up on a dashboard. "By the time the two things get connected," he says, "you've usually already lost the people who noticed the gap first."

"[AI has] changed how I think about it without changing what we actually do about it, which is probably the honest answer most people in this position would give," he says. "The cycle we run is roughly annual. It's not fast enough."

The reason it stays that slow sits in the calendar rather than the workload, he says. "It means decoupling the compensation review from the annual budget cycle, which is where the real friction sits. Right now the two things happen on the same timeline, so any mid-year evidence that a role has drifted gets parked until the next cycle. Shorter feedback loops between what managers are actually observing and what gets surfaced to the compensation committee would help more than any new tool or framework would."

Getting that review moved faster isn't a fight, either, because most people at the company can't actually see the drift well enough to argue about it. "Few team members push back," Shah says. "No blaming them, because they have targeted vision on what they are working on, but I, as a founder, have a bird's-eye view of what works and what doesn't. So it's just a matter of how convincing or result-oriented I can take this conversation."

Testlify isn't unusual in that respect. Only 43% of C-suite and senior HR leaders are highly confident that pay decisions are consistent and fair across comparable roles at their own organization, and 20% say their managers are making those calls on intuition rather than data, according to Korn Ferry's survey of 1,600 such leaders.

The rolling review

Traditional salary surveys update once or twice a year, using data that's already 12 to 18 months old by the time it publishes, according to Ravio's 2026 AI compensation and talent trends analysis. By Ravio's own real-time data, AI and machine learning roles already command a 12% premium over standard engineering pay at the professional level, a gap wide enough that a company benchmarking off last year's survey is pricing the job it used to be, not the one it's hiring for now. "Benchmarking would need real-time task-level data feeds instead of annual title surveys," Troyer says.

Ruffini stopped waiting for the annual pay conversation to check whether a role had drifted. His team now reviews on a rolling basis instead, and treats a repeated hiring struggle as a signal the band is stale rather than proof the recruiting process failed. "Catching drift early is mostly about being willing to reprice the role the moment the work changes," he says, "instead of waiting on the calendar's permission."

Stoyanov re-scoped roles quarterly instead of annually, leveling people against what they're actually accountable for, including "supervising the AI and owning its errors," rather than a document written the year before. She expected resistance and didn't get it. "People felt more comfortable with quarterly re-scoping," she says, "as they could take action faster and understood much better why the changes were required."

Troyer's version is lighter: a quarterly check where each role owner lists the top three tasks that actually move the needle and whether the title and band still match. "It takes one afternoon," he says, "and has already caught two other drifts before anyone reached the exit interview."

Dr. Gleb Tsipursky, CEO of future-of-work consultancy Disaster Avoidance Experts, recommends the same cadence specifically for AI-exposed roles, built around four questions: what work disappeared, what new judgment is required, what risk the employee now owns, and whether the level still matches that responsibility.

Peter Berzin, founder of workforce platform PeopleE3, advises organizations on manager effectiveness. "The middle manager is closest to the work, so the manager feels the disruption first," Berzin says. He tells CHROs to have managers ask sharper questions than whether anyone is unhappy: who on the team would get a meaningfully bigger offer if they looked today, and who has outgrown their title or pay band in the last 12 months. "Most [employees] will not fight for [fair pay]," he says. "They just quietly leave. By the time it reaches your exit data through an interview, the decision was made months ago."

Not every fix is repricing

Ruffini repriced the moment the work changed rather than waiting for a review. Stoyanov re-leveled every role each quarter. Berzin argues a company should raise a top performer's pay before they ever have to ask. All three treat repricing as the natural response once a role has genuinely changed, only the timing is in question.

Suzanne Bali-Courtemanche, director of human resources at Business Sherpa Group, questions the diagnosis itself, before pay ever enters the conversation. "AI shouldn't trigger a series of one-off job evaluations," she says. "It should trigger an organizational review." Companies mistake AI touching a role for the role having changed, without checking whether AI only changed how the work gets done while what the person is actually accountable for stayed the same.

"AI implementation should be viewed as an organizational design exercise, not a job description exercise," she says. "Once that assessment is complete, some roles may change significantly, some only marginally, and others not at all. The extent of change shouldn't be assumed in advance." Run that process honestly, she says, and a lot of roles turn out not to need repricing at all. "If the core accountabilities haven't materially changed," she says, "it doesn't necessarily follow that the role should be reclassified or paid differently."

Amy Spurling, founder and CEO of Compt, spent years as a CFO and COO, sitting on the budget side of pay decisions. She doesn't think higher output by itself earns higher pay, even in a role that's genuinely changed. "AI is a tool, much like a laptop was a tool back in the day," she says. "When people went from notepads and dictation to computers, it made them faster. It didn't make them more expensive." No one raised a typist's pay for typing faster once computers arrived, in her view, and AI doesn't change that logic just because the tool is newer. "One of our team members is doing work that took three people four years ago, and she still has capacity," Spurling says. She calls that an investment in tools and training that lets her team operate at a different level, not proof that any one person's job is now worth more.