Garry Tan told CNBC in March 2025 that for roughly a quarter of the startups in Y Combinator's current batch, AI had already written 95% of the codebase (CNBC).
"You don't need a team of 50 or 100 engineers... You don't have to raise as much."
That is not a story about productivity software. It is a hiring plan. Founders still building headcount around the old assumption, that team size scales with product surface area, are budgeting for a workforce their competitors no longer carry.
The mistake most founders make when they hear this is assuming the cuts land at the top: fewer VPs, thinner management layers, leaner leadership. The data says the opposite. The roles disappearing fastest sit at the bottom, in the entry-level jobs that used to exist so a 23-year-old could learn the trade by doing its repetitive parts.
The junior tier is the load-bearing wall that broke
A Stanford study led by economist Erik Brynjolfsson, using ADP payroll data, found a 13% relative decline in employment since late 2022 among workers aged 22 to 25 in the most AI-exposed occupations, including software developers and customer service representatives. Employment for older workers in those same job titles rose 6 to 9% over the same period (Fortune). Read that pairing closely: same occupation, same job title, opposite outcome by age. The bundle of tasks junior employees were hired to do, the tasks that justified their salary while they learned the judgment calls seniors get paid for, is precisely what automation stripped out first.
That is the mechanism founders should plan around, not the vaguer claim that AI is simply coming for jobs. The jobs it targets first are structured as supervised execution: implement the ticket, answer the tier-one request, draft the first version of the deck. AI agents now do that work directly, on demand, without a learning curve and without a seat to fill.
The warning gets louder at the macro level. Anthropic CEO Dario Amodei told Axios in May 2025 that AI could eliminate up to 50% of entry-level white-collar jobs and push unemployment to 10 to 20% within one to five years, and said the industry needed to stop "sugarcoating" what was coming (Axios). Whatever you think of that as a societal forecast, as a hiring signal from the CEO of a frontier AI lab it is unambiguous: the entry-level req you are about to open is the one most likely to sit empty of purpose within eighteen months.
What the lean team actually looks like
This is already visible in the numbers, not just the rhetoric. Carta found that average headcount among consumer startups closing a seed round fell to 3.5 employees in 2024, down from 6.4 in 2022 (Carta). Y Combinator's Fall 2025 Request for Startups went further, calling on founders to build "the first 10-person, $100 billion company" and naming revenue per employee as the metric high-agency startups should now optimize for (Inc.). Sam Altman has said there is a running bet among his tech-CEO peers over which year the first one-person, billion-dollar company appears, calling it something that "would have been unimaginable without AI and now will happen" (Fortune).
The clearest working proof isn't a projection. Midjourney has passed $200 million in annual revenue with roughly 40 employees and has never taken outside venture capital, according to founder David Holz (The Information). That is roughly $5 million of revenue per employee, reached without a support desk, a growth team, or a bench of junior engineers, because the company never built those layers to begin with.
What gets more valuable, not less
None of this means fewer humans matter. It means a different, smaller set of humans matter more. The roles that survive and gain leverage are defined by judgment and accountability rather than throughput: the senior engineer who can specify a system and catch what the AI got subtly wrong, the support lead who handles the escalation the bot couldn't close, the operator who owns a customer relationship end to end. Those roles were always expensive to grow junior people into. Now founders can skip the apprenticeship, because AI supplies the execution volume that used to justify keeping a trainee on staff.
- Before opening a req, ask whether the role is defined, repeatable execution or judgment you are not yet willing to automate.
- If it is execution, assign it to an AI agent under a senior owner instead of a junior hire.
- If it is judgment, hire the senior directly. Skip the multi-year apprenticeship you used to need to grow one.
The practical move this week: pull up every open requisition on your hiring plan and run it through that test. Every headcount line you keep should be one where a person, not a policy, is the reason the work gets done well.