Episode Details

A senior engineer hired at a fast growing startup this year is, in the overwhelming majority of cases, someone who took a junior role somewhere between 2018 and 2021. That fact is inconvenient for the hiring plan most startups are building right now. SignalFire’s 2026 State of Talent Report found entry level hiring down 76 percent at early stage startups compared with 2019, a steeper cut than the 65 percent decline at large technology companies. The easy read is that AI absorbed the routine work and startups saved money. The harder read, the one most 2027 hiring budgets are not stress testing, is that the same startups cutting junior hiring today are the ones who will bid against each other for a shrinking senior pool by 2030.

Start with what AI coding tools actually do well, because the mechanism matters more than the headline. Stanford Digital Economy Lab researchers Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen, working from ADP payroll data collected since ChatGPT’s November 2022 release, draw a distinction between codified knowledge and tacit knowledge. Codified tasks, writing a unit test, scaffolding a database endpoint, fixing a compiler error with a familiar signature, are documented in millions of public repositories, which is exactly the training data large language models learn from. Tacit tasks, judging which of three imperfect migration paths fails least badly at two in the morning six months from now, depend on a specific system’s failure history, which no public repository contains. The Stanford team found employment for workers aged 22 to 25 in the most AI exposed roles down about 11 percent since November 2022, while similarly aged workers in the least exposed roles grew about 10 percent over the same stretch. The gap between those two groups reached 19 percent by June 2026, up from 15 percent a year earlier, and it comes almost entirely from reduced hiring, not from firing.

Company level data confirm the pattern from the other direction. Anthropic reports that Claude now writes between 70 and 90 percent of its own production code, and Boris Cherny, who leads Claude Code, says his personal output has been 100 percent AI written for more than two months. Microsoft’s comparable figure, disclosed by Satya Nadella in April 2025, sits at 20 to 30 percent, a wide gap that says less about model capability than about how aggressively a company has restructured its own workflow around agentic tools. Meanwhile Meta cut 10 percent of its workforce in May 2026 after two years of 19 percent growth, and Oracle, Atlassian, and Snap each announced their own layoffs that same spring. The companies still hiring aggressively, Ramp at 94 percent two year headcount growth, Wiz at 84 percent, Datadog at 68 percent, are adding engineers too, just not new graduates in anything close to the numbers they once did.

Here is the part a hiring plan built for one fiscal year will miss. Engineering seniority is not a credential a company can buy faster once it decides it needs more of it. It is built through three to five years of exposure to a specific system’s failures, mentored by someone who already went through that process. SignalFire’s data show engineering manager span of control widening from roughly 10 to 12 direct reports at large technology companies and running closer to 15 at startups, meaning the people who would otherwise be training junior engineers already have less time per person to do it, even before a company decides to stop hiring juniors altogether. Cut the junior intake for three or four consecutive years, as the industry has now done since 2023, and the senior pool available in 2029 through 2031 is smaller than it would have been, regardless of how good the coding agents get in the meantime.

Two companies are betting against the consensus, and the bet is worth watching. IBM announced it would triple its United States entry level hiring in 2026, a direct wager that a well mentored junior