You're More Capable. Why Do You Have Less Bargaining Power?

The tool that makes you better at your job also makes you easier to replace.

What happens to a skill on the day everyone else can borrow it?

If you work at a desk, you may already know the first half of the answer. The report that took a day now takes a morning. The code you would have typed out arrives as a draft you correct. You clear more work, faster, with fewer mistakes, and by any measure your manager uses, you are better at your job than you were two years ago.

The second half is harder to put into words. The raise conversation does not go the way the output numbers say it should. The posting for your old role asks for less experience than you had when you got it. You may suspect that someone above you has worked out that a newer, cheaper hire with the same tool would get most of the way to what you do.

Both halves are true at once. You are paid partly for what you produce. What you can bargain for depends on something else: how much your employer would lose if you left, how quickly it could find someone to do what you do, and what you could do instead. A tool can raise your output while it shrinks the cost of replacing you, and when everybody gets the same tool, the second can move faster than the first.

This is also why the standard advice feels oddly hollow. Learn the tools, everyone says, and it is good advice: someone who refuses them will fall behind someone who uses them. But advice that everyone can follow rarely makes anyone scarce for long. Once the whole team can turn out the report in a morning, a report in a morning is what the job requires. The time saved could show up as a shorter day or a raise. It can also show up as another report. People will use the same tools with very different skill, and that difference counts. Learning the tools by itself keeps you in the running without guaranteeing a lead over the person who learned them last month.

The Tool Learned From the Best

The clearest evidence comes from a call center. When a large software company rolled out an AI assistant to 5,172 customer-support agents in stages, the agents resolved about 15 percent more issues per hour on average, according to a study by Erik Brynjolfsson, Danielle Li and Lindsey Raymond in the Quarterly Journal of Economics in 2025. The gains were not shared evenly. Less experienced and lower-skilled agents got both faster and better. The most experienced and highest-skilled agents gained a little speed and lost a little quality.

The assistant had been trained on records of past conversations, labeled with how each one turned out and with whether the agent handling it counted as a top performer. In an earlier version of the paper, the authors offered suggestive evidence that it was spreading the practices of the more able agents to everyone else. On that reading, what a good agent had learned over years, such as which phrasing calms an angry customer, was reaching a new hire through the screen.

For the best agent on the floor, the picture is different from the company’s. She knows everything she knew before, and the company is better off. What has shrunk is the distance between her and the person hired last month, and that distance was part of what made her expensive to lose. The study did not track pay, so it cannot say whether anyone’s wages moved. A narrower gap makes her cheaper to replace, which is how it would weaken her bargaining position.

If the assistant did spread her practices, her advantage went through a process of assisted commoditization: her skill became part of the floor everyone now stands on. The company goes on benefiting from what she knows while depending less on her for it. In September, the economists Masao Fukui, Emi Nakamura and Jón Steinsson published a working paper with a model of this mechanism in general form: technology that standardizes work raises productivity but makes workers easier to swap, and the easier swap weakens their bargaining power and can pull wages down.

When Buyers Need Fewer of You

A second study looked at a market where prices move faster than a payroll does. After ChatGPT and the first popular image generators arrived, freelancers in the most exposed occupations on a large online platform got fewer jobs and earned less, according to research published in Organization Science in 2024. The authors also found suggestive evidence that the best-established freelancers were hit hardest rather than spared. They are careful to say this is the short run, on one platform, and that the result on skill is not settled.

This is a different mechanism from the call center. There, a cheaper worker became a plausible stand-in for an expensive one. Here, some buyers may simply have stopped hiring, writing the copy or making the image themselves. Being the best helps against the first danger for as long as buyers still pay for the difference between best and adequate. Against the second it helps less, because you can outperform every competitor while the whole market for the work shrinks.

What the Sat-Nav Took

London’s cabbies went through the same thing a decade before the chatbots arrived. To drive a black cab in London, you have to pass the Knowledge, which Transport for London describes as memorizing the streets and landmarks within six miles of Charing Cross, including 320 standard routes. It typically takes three to four years, and for generations it made a London cabbie very hard to replace.

Satellite navigation left the cabbie’s skill intact and gave every driver without the Knowledge a serviceable version of it. When app-based private hire spread through London in the 2010s, those drivers could follow a screen to the door. TfL’s licensing figures show licensed taxi drivers falling from 25,538 in 2013-14 to 16,676 in 2024-25, while licensed private-hire drivers rose from 65,656 to 106,468 over the same years. Price competition, the pandemic, vehicle costs and the length of the test all played a part, and the figures cannot say how much belongs to each. The screen’s part was narrower: it took away the cabbie’s scarcity. An official of the Licensed Taxi Drivers’ Association has argued that following a sat-nav does not mean you have learned a route. He is right, and fewer passengers were paying for the difference.

Where Your Leverage Survives

The pattern covers only the part of your value that a tool can hand to someone else, and bargaining power has two sides. One is what your employer would lose if you left. The other is what you could do if you did.

Picture two people at the same law firm. One spends her days turning standard templates into first drafts of ordinary contracts. The other is the person a client calls when a deal is about to fall apart, because she knows the client, knows the lawyer on the other side, and will put her name on the advice. The tool helps both of them. It makes the first lawyer’s best skill available to every junior in the building, and it frees the second to spend more of her time on what nobody else can do for that client. Yet the same tool also cuts the other way for the first lawyer. Work that once needed a firm’s back office can now be run from a small practice of her own. Whether that helps her depends on whether any clients would follow her out the door.

So the useful test is what would happen if you left tomorrow. How quickly could someone else, with the same tool, deliver a result your employer would accept? Would any clients, colleagues or opportunities come with you? Does the tool make it easier for you to work somewhere else, or on your own?

The idea that what you know is too subtle to write down offers less comfort than it seems. The top agents did not have to turn their expertise into a manual for the assistant to learn from their conversations. But the company may still need them to supply what the system learns next. The researchers raised that possibility: fewer original contributions from skilled workers could make future versions less effective at solving new problems. Where unfamiliar cases still require an expert, the company continues to depend on that person even if the assistant spreads what they already know.

The answers differ by job. If your edge was a skill the tool now supplies to anyone, such as routine drafting, standard answers or knowing the usual route, your output can rise while your price falls. If the tool amplifies something it cannot hand over, your position can get stronger, though not every such thing protects you the same way.

A client who would follow you out the door raises both what your employer would lose and what you could do if you left. Responsibility protects you only while no one else your employer would accept is willing to carry it. Knowing exactly how your firm works can make you hard to replace. It can also make leaving harder, because another employer may have little use for it. You can be needed where you are and still have nowhere equally good to go.

If cheaper output makes buyers want much more of it, demand for workers can grow too, giving them more employers or clients to choose from. The next few years will show which of these is happening where. Rising pay alone would not settle it, since wages can rise while workers keep a smaller share of what they produce. The better sign is whether experienced workers in heavily assisted jobs become harder or easier to replace, and whether their options outside the firm widen. If they become harder to replace, the argument is wrong about those jobs.

The best agent in that call center still knows what to say to an angry customer. So, increasingly, does the screen in front of the new hire.