This article is a deep-dive from JudyAI Lab — an AI engineering playbook series with 100+ published guides, 5,000+ weekly readers across 60+ countries, focused on the practical side of running AI agents, trading systems, and content pipelines in production.
8,000 Resumes, a Door That Won’t Open
I read Nikkei Asia’s June 30 report twice: top US university tech grads sent out 8,000 resumes and heard almost nothing back. The numbers TechCrunch pulled together the same week are even colder — before May 2026, US employers had explicitly tagged over 90,000 positions as “eliminated due to AI.”
Put those two numbers together and, from the builder’s seat, it isn’t a surprise. MIT NANDA’s 2025 State of AI in Business report punctures one contradiction: 95% of enterprise GenAI deployments return zero P&L, but that 5% that does see returns almost all use AI for repetitive entry-level work (data processing, code review, content editing), and the P&L gain shows up directly as headcount cuts. That maps straight onto the other side of Nikkei’s 8,000-resume story — junior headcount is gone, so the resumes have nowhere to land.
On the builder side, I’m not surprised. The deeper I go, the clearer it gets why those 8,000 resumes hit a closed door.
The Three Types of Jobs AI Is Actually Eating
I’m not going to throw around scary generalities. From actually deploying agents, I can clearly name which positions are getting locked out of the interview list.
The first is entry-level technical documentation, code review, and data processing. This is what junior data analysts and entry-level engineers used to do every day: catch typos, standardize formats, generate summaries, run basic stats. Deploy an LLM agent now and it hits 70–80% quality, runs 24/7, and costs less than a junior’s monthly salary. A lot of US tech companies have already cut these headcounts — not trimmed, eliminated. That 5% from the MIT NANDA report with real returns almost all use AI to replace this kind of repetitive work, and the P&L shows up straight in the headcount budget.
The second is customer support and content editing. In TechCrunch’s list of 90,000 eliminated positions, support engineers and content editors are the heaviest groups. These jobs used to be the dual-track growth ladder for mid-career juniors — go in for 1–2 years, learn the industry jargon and how to talk to customers, then move up into marketing, product, or customer success. Now AI handles tier-1 ticket triage, FAQ replies, and large-scale copy generation on its own, and that dual-track ladder is broken. This isn’t about whether the news is good or bad — the org chart has already been redrawn.
The third is entry analyst roles — junior finance, data, and market research. Here’s where it gets absurd: I see plenty of US tech companies using AI to screen resumes and run first-round interviews, blocking candidates with the reason “AI already replaced this role.” So AI is acting as the HR gatekeeper while simultaneously killing the entry-level openings. That’s the coldest loop behind the 8,000-resume story.
Five Patterns That Survive
A closed door doesn’t mean total wipeout. From building an AI team, here are five positions I genuinely cannot find a way for an agent to carry alone.
One: AI trainers and prompt engineers. These are new roles, but the bar is way higher than “familiar with ChatGPT” on a resume. What you actually need is someone who can design evaluation criteria, break down agent failure modes, and maintain multi-agent collaboration. I spend 20–30 hours a week doing this myself — far more than writing code.
Two: systems integration. The part of building an AI team that AI can’t replace is wiring different systems (content generation, databases, notifications, review gates) into a stable production line. That requires cross-team communication and judgment about when to let an agent run on its own and when to require human review. It’s a rare mix of “understands tech, understands the business, understands people.”
Three: product positioning and customer problem discovery. AI can spit out 100 solutions, but it can’t decide on its own whether a problem is even worth solving. That judgment requires deep customer understanding — running customer interviews yourself, taking the product through its own failures.
Four: decision ethics and compliance. Anywhere money, law, healthcare, or privacy is involved, a human has to be in the loop — a named human with legal accountability. When building automated trading or decision systems, the common practice is to add emergency stops and human sign-off gates. Not because the tech can’t do it, but because you can’t hand responsibility to an agent.
Five: physical-world operations. Plumbing, F&B, logistics, on-site repairs — AI substitution moves way slower here than people imagine. I keep telling friends around me: don’t underestimate the value of physical labor and on-site experience just because the news is loud.
Three Things Office Workers Should Do
First, treat AI as a colleague to train, not an enemy. Real-world skills are built like this: pick one concrete task (weekly report summary, customer reply categorization, data scraping), design a prompt, run it 10 times, study the failures, refine the prompt. After three months you’ll understand real-world AI use better than anyone in your department who treats AI as just a “chat toy.” That process itself is your differentiation.
Second, build cross-domain skills. Tech plus domain knowledge is always worth more than tech alone. If you’re in marketing, tie AI to customer psychology. If you’re in finance, tie AI to compliance. The surviving positions all sit at the intersection.
Third, document your workflow visibility. Write down seriously what you actually do each week, what your decision chain looks like, and what your output metrics are. Before your boss decides whether to replace you with AI, this is the record they’ll see first. Positions whose workflows are invisible get cut first.
Closing
8,000 resumes isn’t AI’s fault — it’s the “structural disappearance of entry-level roles.” That structural gap won’t be filled by complaining about AI. After building an AI team for over six months, the truth I see is this: AI amplifies the problems each position already had. Repetitive roles disappear fastest; integration, judgment, ethics, and physical roles get more valuable. The number 8,000 isn’t a generation’s failure — it’s a complete reshuffle of the entry-level structure.