AI will force people to find out whether they were paid for effort—or for the value their effort used to hide.
Hourly pay has a logic.
Sometimes the buyer is paying for a person to be present. A nurse on a shift. A lawyer available during a negotiation. A technician responsible for keeping a system running. A manager carrying a problem that might arrive at any time.
Time is not a perfect measure in these jobs. But it is connected to what is being bought: attention, availability, care, responsibility.
In other kinds of knowledge work, time has always been a rougher proxy.
The client is not really buying eight hours of spreadsheet work. They are buying an answer they can trust. They are not buying three days of writing. They are buying a clear argument, a decision, a piece of software, a recommendation, or a result.
The hours are how the work was priced because the answer was expensive to produce.
AI makes that arrangement harder to ignore.
If an experienced person can do in one hour what used to take ten, what exactly should happen? Should the client pay one tenth as much? Should the company expect ten times as much output? Should the person be rewarded for finding the faster way? Or should everyone pretend nothing changed because the old pricing model is familiar?
There is no single answer. But there is a useful distinction.
AI changes the work where time was standing in for expertise. It does not erase the work where time itself is part of the service.
Faster is not the same as more valuable
This matters because the first metric companies reach for is output.
How many reports did the person produce? How many tickets did they close? How many proposals did they draft? How much code did they generate?
Those are easy numbers. They can be badly misleading.
Someone can use AI to produce ten times more material without creating ten times more value. They may create more work for reviewers. They may make weak decisions look polished. They may flood a customer with messages that technically answer the question and practically make the experience worse.
The useful question is not whether the person moved faster.
It is whether the work became more valuable because they moved faster.
Did quality improve? Did risk go down? Did the customer get a better answer? Did the team make a decision earlier? Did an old task disappear? Did the person use the time to solve a harder problem, or simply create more artifacts for someone else to read?
That is the difference between speed and leverage.
Speed means the same work gets done more quickly.
Leverage means the person can create an outcome that would otherwise require more time, more people, more expertise, or more risk.
AI can create either. Companies should not reward them as if they are the same thing.
The awkward incentive
The person who becomes much faster has a reasonable fear.
If I show the company that I can do this in an hour, will it pay me for the hour and quietly raise the expected output for every remaining hour?
That fear creates predictable behaviour. People hide their best workflows. They keep their useful prompts private. They leave enough visible effort in the process to protect the old value signal. They use AI to get ahead personally, but not enough to change the way the team works.
This is often described as resistance to adoption.
Sometimes it is simply a rational response to an unclear bargain.
If a company wants employees to reveal the work that AI can change, it needs to say what happens when they do.
Will the saved time be turned into harder, more interesting work? Will the person be recognised as the owner of a reusable capability? Will they get room to improve the workflow? Will they share in the commercial upside? Or will the reward be an invisible new baseline and a larger workload?
The company does not need to promise that every useful prompt earns a bonus. That would become absurd quickly.
It does need to avoid teaching people that making themselves more effective is a private risk.
Outcome-based pay is not a magic answer
The obvious response is to pay for outcomes instead of time.
That can be better. It can also be dangerous.
An outcome is rarely created by one person. A strong proposal may depend on a brand, a customer relationship, data, legal approval, a team that can deliver, and ten decisions made before the writer opened the document. A software feature may look successful because of timing or distribution rather than the work that built it.
Outcome-based pay can also make people hide uncertainty. If the reward is attached only to a visible result, who wants to say that the model was wrong, the customer was not ready, or the safest decision was to stop?
So the answer is not to replace all salaries with a spreadsheet of AI-attributed outcomes.
The answer is to become more honest about what each role is for.
Some roles are paid for time and availability.
Some are paid for reliable production.
Some are paid for judgment under uncertainty.
Some are paid for the ability to create a system that lets many other people do better work.
AI makes these differences more visible. It does not make them easier to measure.
What becomes scarce
As drafting, analysis, and production become cheaper, several things become easier to see.
The person who can tell whether the work is aimed at the right problem becomes more valuable.
The person who can see the risk a polished answer is hiding becomes more valuable.
The person who can turn a local shortcut into a reliable workflow becomes more valuable.
The person who can get a change adopted by other people becomes more valuable.
The person who can say no to a plausible but wrong idea becomes more valuable.
None of these are mystical qualities. They are forms of judgment, responsibility, and trust. They have always mattered. They were harder to isolate when everyone spent so much time producing the first draft.
That may be the real effect of AI on pay.
It will not eliminate time as a measure. It will make companies confront where time was a convenient way to avoid naming the contribution they actually valued.
The question for an employee is not only, “How can I use AI to work faster?”
It is, “What do I do with the capacity I create, and can I make that contribution legible?”
The question for a company is harder.
If someone creates ten times more value with AI, do you have a way to notice it before they decide to create that value somewhere else?