The Most Fluent Generation

I'm optimistic about the employment prospects of young people. For the first time every high school and college student in the country has permissionless access to an era-defining technology. Why should we not assume the same ambitious, talented students that previously polished resumes with extra-curriculars and good grades to impress admissions and hiring officers will readily adopt new and valuable skills directly? AI, among other things, is a tremendous tutor, coach and enabler.

The prevailing narrative runs something like this: AI is coming for entry-level work first. The tasks we used to hand to a junior analyst — summarize this, draft that, pull these numbers — are exactly the tasks a model does in seconds. So the bottom rung of the career ladder disappears, and a generation of young people is locked out before it starts. It is a frightening argument. It is also right about the tasks but wrong about who suffers.

The error is treating young people as the incumbents of entry-level work — the ones with the most to lose when the tasks are automated and the bar is raised. Incumbency cuts the other way: the people most threatened by a new tool are the ones who spent years mastering the old way of doing things and now have to unlearn it. A generation that has never known a workflow without these tools faces no such tax. For today's sixteen to twenty-two-year-olds, AI is not a disruption to a career they've built. It is the water they swim in.

When spreadsheets arrived, each new analyst class at banks and consulting firms was better at utilizing those tools. At BCG, it was a common joke that Managers and Partners couldn't be an Analyst today — a sentiment that has been true at knowledge work firms for decades. New tools don’t lower the ceiling on what a young person can contribute. They raise it, and fastest for those with the least to unlearn.

But the spreadsheet analogy contains the reason this transition is different, and the difference is the whole essay. Spreadsheet fluency was acquired on the job, after the hire, inside firms that already ran on-campus recruiting and already knew how to find twenty-two-year-olds. The tool made the young analyst more valuable to an employer who had already found her. AI fluency is acquired before the hire — at sixteen, at home, or in a dorm room at a school with no on-campus recruiting. Skills now accrue to ambitious people who may be outside every channel built to find them. Spreadsheets raised the value of young workers who were already visible. AI is producing capable young workers whom no one can see.

Here is the part I find genuinely optimistic: As AI raises the supply of capable young people it simultaneously collapses the cost of verifying which ones they are. The first effect makes visibility the binding constraint on this generation. The second makes solving it possible for the first time.

Historically, it has been costly to learn and demonstrate the skills needed to get hired. College is expensive in time and dollars, and a brilliant kid at a lower ranked school with no network faces huge headwinds. For firms, candidate screening was costly, and as a result most firms used blunt proxies like school affiliation.

AI reduces the need for credentialing in two ways. First, it is a tireless, patient tutor that can guide learning in most of the areas important for entry level professional success. That is the equitable argument — students no longer need permission and access to learn. Second, a student can now point at work that exists: an app running in production, a research memo a firm actually used, a small business with paying customers. The credential matters less and the output matters more.

The output is a better signal than the credential, but it is not a clean one. AI generates confusion in attribution and prediction — just because an eighteen-year-old developed an app vibecoding, will they be able to perform well in the workplace? As we see already in high school and college assignments and evaluations, AI can obfuscate as much as clarify true ability. 

I have two thoughts on this. First, an employer with a problem to solve is not a professor evaluating a student. The professor is testing whether you can work unaided; the employer only needs the problem solved. A research memo a firm actually used is not a take-home exam — the market did the verifying. A nineteen-year-old who ships real work with AI is a capable worker in a workplace that also uses AI.

Second, the same technology that creates the problem will solve it. Large employers, who hire young workers at scale, will develop customized processes with their own AI simulating the work to be done, screening for the skills and aptitude they believe will lead to success, and accumulating enough data to iterate over time. The cost of screening per applicant will fall by orders of magnitude, allowing them to open new channels to previously excluded applicants.

But most employers do not have that scale. Small businesses with fewer than 500 employees account for over 40% of private sector employment in the US (60M jobs). These employers do not run campus recruiting and do not maintain applicant tracking systems. Not because they wouldn't value the work — technology skills are becoming more pressing across every industry. Because the cost of finding, screening, and vouching for an unproven nineteen-year-old exceeds the value of the project they'd hand her.

This is non-consumption. These employers are not choosing badly among available options; they have no options. Christensen's name for the innovation that serves non-consumption is market-creating innovation. Airbnb didn't create people with spare rooms; it made them visible and safe to transact with, providing insurance, reputation and scale, and thereby created a market for buyers and sellers. 

The spare capacity today is a generation of young people who are, right now, more able to contribute real knowledge work than any generation at their age in history.  A rapidly growing supply of talent that doesn’t have traditional credentials for employers to evaluate.

The answer is a shared recruiting layer — an organization that diffuses the cost of identifying, evaluating, and matching this talent pool across the employers who could never justify building internal tools. 

The value here is diffuse in a way it wasn't for spare rooms. The value of a room rented is captured in the transaction. The value of a great first job compounds over a career, and almost none of that accrues to whoever made the introduction. That asymmetry is why this market has traditionally been very difficult for venture-backed startups, and also the reason it should be built anyway.

I don't know the precise shape of the organization that closes the gap. I am increasingly convinced it should exist, and when it is built we will look back on the apocalypse narrative the way we now look at the fear of the spreadsheet — as a failure of imagination about what talented people do when you hand them a more powerful tool.