Most people assume AI is a rich-get-richer technology, that it helps the already-skilled pull further ahead. The research keeps finding the opposite. Across several careful studies, the biggest winners aren’t the experts. They’re the beginners.
The support-agent study
The clearest evidence comes from a study by Erik Brynjolfsson, Danielle Li, and Lindsey Raymond, published in the Quarterly Journal of Economics. They followed 5,179 customer support agents at a company that rolled out an AI assistant, and measured what happened to productivity.
On average, agents got about 14% more done. But the average hid the interesting part. For the newest, lowest-skilled agents, the gain was 34%. For the most experienced agents, it was close to nothing. The tool did almost nothing for the people who were already good, and a lot for the people who were still learning.
The reason they found is worth understanding, because it explains the whole pattern. The assistant was trained on how the best agents handled calls, and it surfaced those moves to everyone. So a brand-new agent, in effect, got the accumulated know-how of the top performers piped to them in real time. The tool spread expert practice to people who hadn’t earned it yet through years on the phones.
The pattern shows up again and again
The finding holds up beyond that one study. A well-known experiment run with BCG consultants, the same one that gave us the “jagged frontier” idea, found the junior consultants improved about 43% when they used the tool, while the seniors improved around 17%. The people with less to draw on gained the most from having something to draw on.
And an NBER experiment found that AI closed roughly three-quarters of the performance gap between higher- and lower-education workers. The gap that usually separates the credentialed from the not narrowed sharply once both groups had the tool. It acted as a leveler, pulling the bottom up toward the top rather than stretching the top further away.
Three different settings, same shape. When you give people a tool that encodes good practice, the ones who benefit most are the ones who didn’t have that practice yet.
What this means for the experience curve
I keep coming back to what this does to the way careers normally work. The old deal was that you got good slowly, by doing a task badly, getting corrected, and doing it a little better next time, over and over for years. That grind was how a junior turned into a senior. It was also brutal and slow, and a lot of talented people washed out of it before they got good.
A tool that hands you expert practice on day one compresses that curve. You can see how a strong performer would frame the answer before you’ve built the instinct yourself, then work backward to understand why. That’s a different way to get good, and it’s tilted in favor of exactly the people who used to have the hardest time getting a foothold.
If you’re early in your career, this is the most encouraging finding in the whole AI-and-work literature, and it’s the one that gets the least airtime.
How to actually use the edge
The catch is that the tool only compresses the curve if you learn from it. If you copy the output and move on, you got a faster answer and learned nothing. The junior agents in that study didn’t just paste suggestions. Over time, the good ones absorbed the patterns and started producing that quality on their own.
So the move is to treat every answer as a worked example. Ask the tool to explain its reasoning, not just its conclusion. Compare its approach to what you would have done and notice the gap. Keep the parts that teach you something and let the rest go. You’re using it to see how good work is built, then building the habit yourself.
Done that way, the tool is less a crutch and more a very patient mentor who shows you their reasoning on demand. That’s the thing juniors almost never had access to before, and it’s why the curve bends.
There’s a warning tucked inside this, though, and it’s worth stating plainly. The compression only works if you’re building your own judgment underneath the borrowed output. If you skip that part, you get fast without getting good, and the moment you hit a problem the tool handles badly, you have nothing of your own to fall back on. The newcomers who win aren’t the ones who lean hardest on the tool. They’re the ones who use it to learn faster and then quietly stop needing it for the things they’ve absorbed.
Where you have to stand on your own
There’s one place the tool can’t stand in for you, and it’s the place that decides whether the compressed experience actually lands you the job: the interview. Nobody’s prompting a model in the room. You’re being asked to explain your thinking, defend a call, and handle a follow-up you didn’t see coming, in a live conversation.
That’s a skill in its own right, and it responds to practice against something that pushes back rather than nods along. The good news is that it fits the same pattern as everything else here. A newcomer who practices talking through their reasoning, out where someone can probe it, closes the gap with more polished candidates faster than they’d expect. The experience curve compresses there too, if you work it.
The honest read
The fear that AI mostly helps the already-skilled turns out to be backwards. It helps the newcomer most, because the newcomer is the one who didn’t already carry the expert’s knowledge around. For anyone early in their career, that’s a rare piece of good news, and it’s worth taking seriously.
But the edge only exists if you use the tool to learn rather than to skip learning. Study the reasoning, close the gaps, and practice standing on your own where it counts. Do that, and the years it used to take to get good start to look more like months.