For most of history, people got good at things by doing them. The apprentice made the piece and became a master. The developer wrote the code and became senior. AI is loosening that link in almost every kind of thinking work. I call the distance between what we can now produce with AI and the judgment we develop through the work The Becoming Gap. Here I look at it across time: what the work used to build in us, what AI and agents are changing now, and what we carry forward.
The Becoming Gap Through the Lens of Time Travel by Niketa Jhaveri
Eighteen Hours, Gone in Minutes
I've been making Rangoli since I was thirteen. Every Diwali I take the week off, sit on the floor, and build one piece by hand. A big one takes me eighteen to twenty hours. Then it's swept away in about five minutes. I've known that every time I started. Nobody makes Rangoli to keep it.
Today an AI can generate a Rangoli pattern in a few seconds, and it would probably be more intricate than mine. When I thought about that, I realized the pattern was never really why I did it. What those hours gave me was something else. I learned to see when a color isn't working. My neighbors come by to watch it take shape. And there is the Ganesha I made for my father after he passed away, the one in this video, which I could never have handed to a machine. The pattern is gone in minutes. What the making did to me is still here. This essay is about that, and about what happens to it now that so much work can be done without us.
The Ganesha I made for my father, Diwali 2025. Eighteen hours on the floor, shown in 37 seconds. The video keeps the pattern. It can't show what the hours did.
The Past: When the Work Made the Worker
We use the word masterpiece for any great work now. In the medieval guilds it meant something narrower. It was a test. A craftsman made one piece, his master piece, and showed it to the guild. If it passed, he earned the title of master and the right to open his own workshop and train apprentices. The guild cared less about the object than about what it proved, which was that this person could now be trusted with the craft.
And the work didn't only prove people. It changed them. The example that stays with me is London taxi drivers, who spend years memorizing the city street by street. A 2006 study found that the longer they had been driving, the more gray matter they had in the posterior hippocampus, a part of the brain tied to spatial memory. Bus drivers, who follow fixed routes, showed nothing like it. It's an association, and I wouldn't stretch it to every job, but I think about it often. The people who had to find their own way were changed by it. Nobody planned that. It came with the job. Back in 1983 the researcher Lisanne Bainbridge described what happens when it stops coming with the job. In a paper called Ironies of Automation, she showed that automation leaves people in charge of the rare moments the system can't handle, while taking away the everyday practice that would have prepared them.
I learned design this way. In 1999 I was studying commerce and teaching myself Photoshop, HTML, CSS and Dreamweaver on the side. By 2001 I was a junior web designer, still in college. I built a lot of pages that looked right and didn't work, and each one taught me why. I don't build pages anymore. I lead research into how people use AI. But a lot of the judgment I lean on today comes from those years of getting it wrong myself.
The Present: The Masterpiece Without the Journey
This is what's different about AI. Earlier technology took over parts of physical work and then parts of mental work. Generative AI can do the writing, the coding, the analysis, the research synthesis and the design. So for the first time the masterpiece can show up without the journey that used to make the master. I call this The Becoming Gap: the distance between what people can produce with AI and the judgment they develop through the work. The hard part is that you can't see it. The output looks finished. What's missing was never in the output to begin with. It was in the person.
There's some early evidence, and I want to be careful with it. In January 2026 Anthropic ran a randomized trial with 52 mostly junior software engineers who were learning a new Python library. The ones who used AI scored 50 percent on a comprehension quiz afterward. The ones who coded by hand scored 67 percent. And the AI group finished only about two minutes faster, which wasn't a statistically significant difference. That's one small experiment, so I'm not going to claim it holds everywhere. But in that room, people learned a lot less and saved almost no time.
Agents take it a step further. Now you describe what you want, and the agent writes the code, runs the tests and opens the change for review. The developer goes from writing to approving. The same is true in my own field. A researcher can read a polished synthesis without ever learning to hear the difference between what people say and what they mean. Seeing good work isn't the same as being able to do it. I wrote about a related problem in Deliberation Debt, which is what builds up when we skip the thinking behind a decision. The Becoming Gap runs on a longer clock. It opens when we skip the experiences that make us able to decide in the first place.
The Future: What We Carry Forward
What worries me most is the ladder. Senior people got senior by doing junior work. They fixed small bugs, read other people's code, and went through interview transcripts line by line. Agents can now do much of that. Stanford researchers report that employment for workers aged 22 to 25 in AI exposed occupations is 19 percent below where it would be if it had kept pace with their less exposed peers. They are clear that this is descriptive and not causal, and so am I. But it suggests fewer young people are getting onto the first rungs. A company can take the productivity today and find out years later that nobody was becoming the expert it now needs.
Think about what actually leaves when a person leaves. The agents keep the codebase, the ticket history and the context of the last thousand decisions. The person takes only what they became, which is their judgment, their taste, and that feeling that something is wrong before any test says so. That kind of judgment has always been passed along by working next to someone, the way a journeyman worked next to a master. If a generation skips that, it has less to hand on. Other researchers are studying the same shift and have named it the apprenticeship void and the apprenticeship externality. They're asking what happens to entry level work and the supply of expertise. My question is a little more personal. Who would that work have helped us become?
And there's something the work needs that no tool can supply, which is wanting to do it. Nobody spends eighteen hours on something that will be swept away unless they love making it. AI can't give you that passion and it can't take it from you. What it can do is make it very easy to leave it unused. I'm not saying we should do everything by hand. Sometimes you just need the result, and sometimes you need the experience, and the skill is telling which is which. The rule I'd suggest is simple. If you love doing it, keep it.
Closing the Gap: Still Making Masters
None of this is inevitable. Socrates worried that writing would ruin our memory, and every tool since has moved where we do our becoming without ending it. AI can even help. In that same Anthropic trial, the people who asked the AI to explain its code, or only asked it conceptual questions, kept much more of what they learned. That part is an observation and not a proven cause, but it points somewhere useful. And plenty of friction deserves to go. Nobody builds judgment formatting slides. The real job is telling apart the friction that only slows us down from the friction that makes us better, and building new ways to earn judgment where the old ones are disappearing.
If your organization is bringing in AI agents, here are the three questions I would ask. What did people learn by doing this work themselves? If AI does it now, where will they learn that instead? And when the system fails, who will know enough to catch it, step in, and teach the next person? Under all three is one test, which is whether your people can only perform with the tool or are still capable without it.
I often say that the more AI knows, the more humans matter. The Becoming Gap is the condition I'd attach to that. For generations, the effort of doing the work was also making us, and we never had to think about it. Now we do. We can build a future where people produce more and become less, or one where AI lets us do more and still gives us ways to grow. My Rangoli is gone within minutes every year. What making it gave me has lasted my whole life. The masterpiece is no longer proof of the master. What matters now is whether we are still making masters.
Niketa Jhaveri writes on the evolution of intelligence, human judgment and cultural context in AI and through the lens of time travel. She is the founder of Nivan, an Innovation conference bringing together leaders across AI, technology, design, research and innovation. Reach her at connect@niketajhaveri.com

