Guide · Job architecture

When AI changes the work inside a job

The title is the last thing to change. Re-level on what the work now asks of people.

Short answer: When AI changes the work inside a job, judge the job again on what people now do: How complex the work is, how much judgment it needs and what they are accountable for. Some jobs grow, some shrink and many stay where they are. A new title or a new tool is not a reason to move a level. Changed work is.

Three requests in one week

Picture a 180-person company in Spokane that runs billing and customer support for veterinary clinics. A year ago it rolled out AI tools across the business. In one week, three requests reach the head of people.

The head of support wants the 40 support agents retitled "AI support specialists" and moved up a level. Routine questions are now answered automatically, so the agents spend their days on the hard cases: Billing disputes, angry clinic owners, problems nobody has seen before. Their manager says the job has changed completely.

The finance lead asks for nothing. Yet her three analysts no longer build reconciliations by hand. They review reconciliations a tool drafts, chase the exceptions and explain the results to clients. Nobody has called that a new job, but the work is not what the job description says.

The recruiting team wants to know what to post for the entry-level support role. Most of the work the description lists is now done by software. The last two people hired into it left within six months, saying there was nothing to learn.

Each request is about the same question: When the work inside a job changes, what happens to its level? The company has an answer for one of them so far. It is the one with the loudest manager.

This is a composite of a common situation at companies this size, not a client. The details and figures are illustrative.

Why re-leveling AI-changed jobs is hard

The question sets two good goals against each other. The first is accuracy. A level is meant to describe the size of a job and when the work changes, the level should follow. Leaving people at a level that no longer fits their work underpays them and blurs every comparison the structure is meant to support. The second is stability. Levels anchor pay, careers and fairness across the company. If every change in tools starts a debate about levels, the structure stops meaning anything.

The change is real and fast. In May 2026, 30% of US employees used AI at work at least a few times a week. 15% used it every day [1]. A compensation data provider reported in September 2026 that 61% of its respondents were rewriting job descriptions because of AI. 49% said their salary structures had not kept pace and 48% said their market benchmarks no longer reflected the skills roles now require [3]. The figures are the provider's own.

The direction of change is not one way. In some jobs AI takes the routine part and leaves the harder part, so the job grows. In others it takes the part that needed judgment and leaves the checking, so the job shrinks. Many jobs barely change at all. A study of US payroll data from millions of workers found that employment of 22 to 25 year olds in AI-exposed occupations was 19% below where it would have been had it kept pace with less-exposed peers. The declines were concentrated where AI substitutes for tasks. Where it complements them, employment was flat or rising [2]. The same tool can make one job bigger and another smaller.

And the signals people send are not reliable. Managers ask for new levels where they feel the change. Jobs where the change is quiet, or where nobody asks, are left alone.

Two terms will help. A level is a step in the size of a job, defined by things like scope, complexity, judgment and accountability, that applies across teams. Re-leveling means placing a job at a different level because its work has changed, as distinct from promoting a person into a bigger job.

Five traps when AI changes a job

"Put AI in the title"

What it looks like: A team whose work has changed is given a new title with "AI" in it. The new title signals that the job is different and often comes with a request to move up a level or a pay range.

Why smart teams fall into it: A title is fast, visible and free. It rewards a team that has adapted, helps with hiring and feels modern.

What it costs: A title says nothing about the size of the job. Two teams with "AI" in their titles may be doing work of very different weight. One team without it may be doing more than either. When the title becomes the reason for a new level, the structure starts tracking labels instead of work. It also breaks comparisons. Under US federal guidance on equal pay, job content, not job titles or classifications, determines whether jobs are equal. The comparison looks at the duties people actually perform [5]. A new title does not change that comparison. Changed duties do. Titles also travel. Once one team has an AI title, other teams ask for one. Our guide on mapping job titles to levels covers why titles should follow levels, not lead them.

"It's the same job with new tools"

What it looks like: The company treats AI as a change in tools, like moving to new software. Job descriptions and levels stay as they were.

Why smart teams fall into it: It is true for many jobs and it protects the structure from constant change. Re-leveling also costs money. Nobody wants to open the question for every role.

What it costs: For the jobs that really changed, the description now describes someone else's work. The finance analysts in the opening are reviewing, judging and explaining instead of building. Whether that is a bigger job or a different one of the same size is a real question. Saying "same job" answers it without looking. The cost shows up later. People doing harder work at the old level leave for a company that will pay for it. Benchmarks for the job are matched against the old description. And a job description that no longer fits is weak support for any pay decision made against it.

"Move everyone who uses AI up a level"

What it looks like: To reward adoption, the company moves people or teams who use AI heavily up a level, or creates a level above the old one for "AI-enabled" versions of each job.

Why smart teams fall into it: Leaders want to encourage adoption and the people who learned the tools first did create value. A level move feels like recognition.

What it costs: Using a tool is not the same as doing a bigger job. If AI made the work faster but not more complex, the job is the same size, done more efficiently. A level move for tool use also does not last. The same provider found that 23% of employers who once paid a premium for AI skills now treat AI fluency as a baseline expectation [3]. A level given for something that becomes standard is very hard to take back. The better tool for rewarding a valuable skill is a premium on top of the range, with an owner and a review date, not a new level. Our guide on paying for AI skills covers why a premium belongs on top of the band, not in a new one.

"The market data will tell us"

What it looks like: The company waits for salary surveys to show whether AI has changed what a job is worth and plans to adjust levels then.

Why smart teams fall into it: Market data feels objective and it avoids a judgment call the company is not sure it can make.

What it costs: Surveys describe jobs as they were when the survey was designed and match them by description. If your job's content has moved and the survey's has not, the match is wrong in a way the data cannot show. Surveys also lag. A survey published this year was collected last year about work that was defined earlier still. And survey matches tell you what a job pays, not what level it belongs at inside your structure. That is a judgment about your own work and only you can make it. Our guide on job matching for salary benchmarking covers where matches quietly go wrong.

"The entry level will sort itself out"

What it looks like: The routine work that used to fill entry-level jobs is now done by software. The company keeps the entry-level job on paper, hires into it less often and expects people to grow into the next level the way they always have.

Why smart teams fall into it: Nobody decided to remove the entry level. It just stopped being needed in the old form and nobody noticed the career path was missing a step.

What it costs: Entry-level jobs were where people learned the work before they could judge it. Without them, the company has no way to grow its own mid-level people and has to hire them from outside at a higher price. The entry-level people it does hire have little to do and leave. The payroll research on young workers suggests this is not a small effect [2]. A company can still decide to stop hiring at the entry level. It should decide that knowingly, or redesign the entry job around learning and checking work, not leave a job description that describes work nobody does.

What a sound response to AI-changed jobs looks like

A sound response does not re-level every job that uses AI and does not freeze the structure either. It looks at what each changed job now asks of people and places it on that basis. At 50 to 500 people, the responses that hold up share these qualities.

The job is judged on its level criteria, not on its tools. The question is whether the work is more or less complex, needs more or less judgment and carries more or less accountability than before. The same criteria the company uses for every level apply here.

The description comes from the work people actually do. Not from what the manager says changed, but from what people now spend their time on, checked with the people in the job.

Change is recognized in both directions. Some jobs move up a level. Some get smaller. Where a job shrinks, the people in it keep their pay. The change shows up in how the job is defined for future hires, not in a cut.

A level change is kept apart from a skill premium. If the job grew, it moves level. If the person holds a valuable skill the job does not require, that is a premium, reviewed on its own date.

Jobs are reviewed by family, not by request. The company looks at each job family in turn and asks whether its work has moved. That catches the quiet changes as well as the loud ones.

The comparison is checked by group. If the jobs re-leveled are mostly held by one group and the jobs left alone mostly by another, the company should be sure that reflects the work. In the EU, the criteria used to compare jobs must include skills, effort, responsibility and working conditions and must not undervalue relevant soft skills [6]. Jobs where AI took the routine work and left the people work are exactly where that matters.

Benchmark matches are redone when the content changes. A job that moved level is matched again against the description that now fits.

The entry level is a decision. Either the entry job is redesigned so people can learn in it, or the company knowingly hires at a higher level and pays for it.

The calls only you can make

The bar for a new level. Your level criteria set the bar. Whether a job that is noticeably harder but not yet a full level higher stays put, or gets a premium in the meantime, is a policy decision.

What to do with shrunken jobs. Keeping people's pay is the fair floor. Whether you redefine the job for future hires, combine it with another or leave it is your choice.

Whether to keep an entry level. Hiring experienced people is faster. Growing your own is cheaper over time and builds loyalty. The choice depends on how long you plan to need the skills.

How often to look again. Once a year per job family is enough for most companies. Where the work is changing fastest, a shorter cycle may be worth the effort.

Who decides. A manager's view of how much a job has changed is one input. Someone outside the team should make the level call, against the same criteria for every family.

The data and the law

  • How many people use AI at work. In May 2026, about half of US employees (52%) used AI at work at least a few times a year. Three in ten used it several times a week or more. And 15% used it daily [1].
  • Young workers in exposed jobs. A study of high-frequency US payroll data covering millions of workers through June 2026 found employment of workers aged 22 to 25 in AI-exposed occupations 19% below where it would have been had it kept pace with less-exposed peers. Declines were concentrated where AI substitutes for tasks. Where it complements them, employment was flat or rising, especially for experienced workers. The authors describe this as a descriptive indicator, not a causal estimate [2].
  • Job descriptions and structures. A compensation data provider's September 2026 preview of its AI workforce research found that 61% of respondents were rewriting job descriptions because of AI, 49% said salary structures had not kept pace, 48% said market benchmarks did not reflect the skills roles now require and 23% of employers who had paid an AI skill premium now treat AI fluency as a baseline [3]. These are the provider's own figures.
  • What employers pay for AI roles. A survey of more than 4,000 companies in 133 countries, fielded in February 2026, found employers paying an average of about 10% above market for AI roles, mostly between 5% and 15%. 64% of compensation professionals said they were unsure what premium to offer [4].
  • United States, equal work. Under federal guidance on equal pay, job content, not job titles or classifications, determines whether jobs are equal. The comparison considers the duties employees actually perform and differences in responsibility do not depend on titles [5].
  • European Union. Employers must have pay structures that ensure equal pay for equal work or work of equal value (Article 4(1)). Those structures must allow work to be compared on objective, gender-neutral criteria that include skills, effort, responsibility and working conditions and relevant soft skills must not be undervalued (Article 4(4)) [6].

This is general information, not legal advice. Rules and their national versions differ. Check the rules where you employ people.

Our view

We think levels should follow the work, not the tools and not the titles. When AI changes a job, the right question is what the job now asks of the people in it. Some jobs will be bigger, some smaller and many the same. A company that re-levels on that basis keeps its structure meaningful. A company that re-levels on titles or on who asked loudest ends up with a structure nobody can explain.

So we would review job families on a set cycle, judge each changed job against the same level criteria as everything else, keep skill premiums apart from levels, protect pay where jobs shrink and treat the entry level as a deliberate choice. AI is changing work faster than most structures were built to handle. The answer is not a new structure. It is looking at the work more often.

Sources

  1. Gallup, Indicator: Artificial intelligence (US employees' AI use as of May 2026), checked 2026-10-30
  2. Brynjolfsson, Chandar and Chen, Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence, Stanford Digital Economy Lab (revised 12 August 2026; ADP payroll data through June 2026), checked 2026-10-30
  3. Payscale, AI Workforce Impact Report preview, press release (24 September 2026; vendor survey), checked 2026-10-30
  4. Korn Ferry, Only modest pay bumps for AI skills (16 March 2026; survey of more than 4,000 companies in 133 countries, February 2026), checked 2026-10-28
  5. US EEOC, Compliance Manual Section 10: Compensation Discrimination, 10-IV.E.2 (comparison of work), checked 2026-10-30
  6. Directive (EU) 2023/970 on pay transparency, Article 4 (EUR-Lex), checked 2026-10-28

Do AI skills need their own pay bands?

AI skill premiums belong on top of the band, not in a new one. Five traps in paying for AI skills and what a premium that holds up looks like.

How to price a new AI role

Define the work before the title, price from the closest jobs and review in six months. Five traps when AI pay data is thin and moving fast.

Job titles to levels without a revolt

Title revolts start with a public reveal or one negotiated exception. Five ways title mapping goes wrong and what a fair mapping looks like.

See Resources for every guide and the self-check.