---
id: "lib-2026-10-22-why-firms-stopped-hiring-juniors-10-million"
type: "library"
title: "Why firms stopped hiring juniors: 10 million postings analysed"
title_original: "[고수수첩] 기업들은 왜 신입을 안 뽑을까? 채용공고 천만 건을 분석했더니"
lang: "KO"
org: "MBC PD수첩 (Underscore)"
source_type: "Podcast"
date: "2026-10-22"
week: "2026-W43"
url: "https://nosov.ai/library/lib-2026-10-22-why-firms-stopped-hiring-juniors-10-million/"
original_url: "https://www.youtube.com/watch?v=d9ueNeLnYog"
tier: "C"
tier_name: "Weak signal"
tier_reason: "MBC PD Note interview with Underscore's Kang Tae-young that mostly restates the written job-postings analysis filed separately; the added survey finding is interesting, but its method is not shown here."
summary_basis: "transcript"
original_minutes: "13"
tags: ["junior hiring", "job postings", "ai access inequality", "survey"]
---

# Why firms stopped hiring juniors: 10 million postings analysed

Original title: [고수수첩] 기업들은 왜 신입을 안 뽑을까? 채용공고 천만 건을 분석했더니

Tier C (Weak signal): MBC PD Note interview with Underscore's Kang Tae-young that mostly restates the written job-postings analysis filed separately; the added survey finding is interesting, but its method is not shown here.

Summary based on the full transcript.

## In short

A 13-minute YouTube interview by MBC PD Note (published 22 October 2026), a companion to the broadcast "AI Dilemma Part 2: The Case of the Missing New Hires". Data analyst Kang Tae-young of Underscore walks through the job-postings study: new-graduate postings fell from 2022 in white-collar jobs, and after ChatGPT employers demanded more experience for more AI-exposed jobs. He then adds a survey of 1,700 adults suggesting that people who use only free AI models underrate it, while those shown paid-model output see more risk to experienced workers.

## Key numbers

- ~10 million: job postings from one Korean platform over about seven years, as described in the interview
- 1,700: adults in a nationally representative survey (age, sex and region quotas)
- 46%: survey respondents who use AI frequently
- 8%: survey respondents who pay for an AI subscription

## What it reports

- About 10 million postings from one large platform over roughly seven years, with date, title, job tags, required experience and new-graduate eligibility.
- New-graduate share rose until 2022, then fell in white-collar groups while blue-collar stayed flat. The turn came around mid-2022, before ChatGPT, overlapping rate rises and the post-pandemic hiring correction, which makes attribution hard; the gap kept widening through 2023-2025.
- Employment insurance data: under-29 employment in information and communications peaked in late 2022 and turned down, while under-29s in manufacturing did not and older ICT workers kept rising.
- Required experience barely differed across jobs until November 2022, then rose more for jobs with higher AI exposure. (The written analysis gives the figures; the interview does not.)
- Survey: about 46% of 1,700 respondents use AI often, but only 8% pay for it.
- Respondents shown free-model output rated AI lower than a control group shown nothing; those shown paid-model output rated it higher.
- On wage outlook, the control and free-output groups said longer careers mean more safety; the paid-output group said longer careers mean more risk, with the gap widening by seniority.
- Kang argues access to top models is becoming a channel for inequality: those who can pay see the threat early and adapt.

## How they got it

Spoken summary by the analyst; the postings method and figures are in the separately filed written report. The survey's design (sample of 1,700 with quotas, randomised exposure to example outputs) is described only briefly. Transcript from Korean auto-captions.

## Limits

- Restates a source covered by its own digest; read the written analysis for method and numbers.
- Survey details (questions, sample source, effect sizes, significance) are not given in the interview.
- Same data limits as the main study: one platform, postings rather than hires, not peer reviewed.

## Why it matters for companies

The survey point matters for leaders judging AI from free tools or from staff who use them: perceived capability depends on which model people have seen. Senior staff who have only seen weaker models may underestimate how their own work will change. Giving people hands-on access to current paid models is a cheap way to align expectations before workforce decisions.

## Key passages in translation

> 근데 그 46%는 되게 고무적이지만 이 AI를 유료 구독하는 비율은 이 8%밖에 되지 않습니다.

That 46% is encouraging, but the share paying for an AI subscription is only 8%.

> 오히려 무료 AI를 경험한 집단에서는 이제 통제군보다 더 낮게 나타나는 겁니다.

In fact, the group that saw free AI output rated it lower than the control group.

> 오늘날의 유료 AI 성능이 이제 한 달 뒤, 두 달 뒤에 무료 AI 성능이 되죠.

Today's paid AI performance becomes free AI performance a month or two later.

Original: https://www.youtube.com/watch?v=d9ueNeLnYog