AI Is Breaking the Workplace Career Ladder

By Wen | Insight New Research Lab, Author | Wen Bu

A world-leading AI giant with a soaring valuation is frantically hiring and expanding while quietly closing its doors to newcomers—the share of new hires in its technical department plummeted from 27% to 9% within a year.

Similar signals are also appearing in the broader job market.

Among occupations more heavily affected by AI, the employment performance of young people aged 22 to 25 lags noticeably behind that of their peers in occupations less affected by AI.

The former comes from an internal hiring survey by Anthropic, the parent company of Claude; the latter comes from a research finding updated by Stanford University in August this year.

Put them together like pieces of a puzzle, and you will see a cruel truth that is quietly unfolding across all industries yet rarely spelled out:

On October 16, Anthropic released a highly striking internal hiring report.

The data shows that as the company's overall size expanded exponentially, the proportion of junior employees in its technical department instead plunged against the trend, sliding all the way from 21% to 14%.

Chart of trends in Anthropic's junior technical position hiring and workforce structure

Another set of data is even more breathtaking:

As of August this year, the interactive AI Agents invoked by Anthropic employees had accumulated a total runtime equal to exactly 6 times the total working hours of the entire human team.

This is equivalent to saying that behind every employee, there are 6 tireless "digital workers" working in parallel day and night.

Coincidentally, OpenAI's engineering team also disclosed their latest daily snapshot:

An engineer's day usually begins with a cup of coffee and half an hour of ideation. He only needs to write a detailed natural-language product requirement (Spec) in English, then distribute it with one click to 20 Agents running in parallel in the background.

The Agents split into twenty paths, automatically writing code, building tests, and probing edge conditions.

At two in the afternoon, the engineer returns to the screen and begins reviewing the results produced by the Agents one by one. Encounter something off? He does not even need to personally type a single line of fix code—he only needs to refine that natural-language Spec a little further.

Codex multi-Agent parallel collaborative work interface

An Anthropic engineer used an extremely vivid metaphor: "Every day when I open my eyes, it is as if I am standing before the commander's banner of an AI army, with hundreds or thousands of elite soldiers piling proposals in front of me, lining up waiting for me to make decisions."

Sounds very sexy, right? One person as an organization, the birth of the super-individual, a leap in productivity.

But the fatal bottleneck is stuck precisely on "the person in front of the commander's banner."

A machine can spit out 10 seemingly watertight proposals in 3 seconds, but the human brain still needs tens of minutes or even hours to digest, reason through, and scrutinize those 10 proposals.

The person who can spot the flaw at a glance is not so because he was born brilliant, but because in countless late nights in the past, he was personally hammered by these online incidents, troubleshooted through the night, and crawled out of piles of crashing logs.

Anthropic admitted in the report: when hiring, they no longer value candidates' pure code-typing speed, but place extreme importance on system architecture intuition, business judgment, and a nose for edge risks.

But the contradiction is detonated precisely here:

Companies only want top-tier quality inspectors who can pick out flaws; but the foundational soil of trial and error that cultivates such ruthless judgment is being uprooted by companies themselves.

Companies want mature masters, yet are unwilling to pay for "trainees"

Companies increasingly need experienced talent, yet may be compressing opportunities for newcomers to accumulate experience. This is not merely speculation about the future. The employment data for young people has already shown some warning signals.

In August this year, Stanford University's Digital Economy Lab released a set of brutal data.

After tracking the dynamics of millions of workers, researchers found that in industries most severely hit by AI, the employment scale of young junior workers aged 22 to 25 plunged by nearly 19% compared with normal historical expectations.

Yet positions for senior practitioners remained almost unmoved.

Even more alarming is that this impact did not occur through dramatic "mass layoffs," but seeped in slowly through "silently shutting down junior position HC (Headcount)."

In modern industrial and commercial society, there has never been an industry titan born out of thin air. All "professional intuition" is essentially a compensatory product of massive repetitive low-level labor:

A junior programmer spends an entire afternoon wrestling with a bizarre memory leak, only to finally discover that a variable scope was written incorrectly—for the company, that half-day is extremely uneconomical, but at that moment this young person completed a transformation;

A junior lawyer spends several consecutive days doing tedious case-law searches in piles of old papers and judgment databases, wearing out hundreds of case files before for the first time sniffing out the subtle defensive gaps between legal provisions;

A junior auditor reconciles thousands upon thousands of lines of tedious transaction records in endless shifts before developing a keen professional intuition for those vouchers that appear compliant but are actually falsified.

Including design, film and television editing, media planning, investment banking analysis, financial accounting... in any trade with professional barriers, the bottom layer is paved with dense "apprentice work."

Short-term financial reports indeed look very good, with astonishingly high per-capita efficiency.

But what about 3 years from now? 5 years from now?

When the old batch of technical experts who struggled through the mud retire, resign, or transfer, companies will suddenly turn around and realize: the reserve seats are empty.

You cut the ladder that leads apprentices to mastery, yet you still expect masters to fall from the sky.

Are we to pin all our hopes on the next generation of large models? Not so fast—the slap in the face from the judicial system has already woken everyone up.

When the system loses its safety net, who bears the ultimate cost?

On October 22, the U.S. Court of Appeals for the Fifth Circuit vacated a key injunction against an education bill and ordered that the presiding judge be replaced.

The presiding judge later admitted: when drafting documents, the law clerk lazily used a generative AI tool (Perplexity) to assist with retrieval and assembly, and the subsequent layers of review simply "let things slide," allowing page after page of fictitious parties and fabricated case law to be stamped with the court's official seal in broad daylight.

Judicial documents are not idle chatter on a social media feed; they bear on the direction of policy and on the lawful rights and interests of countless teachers, students, and citizens.

In the past, the reason society placed a high degree of trust in legal opinions, medical diagnostic reports, and audited financial statements was not at all because the wording on them was so elegant, but because behind them stood rigorously trained professionals, a multi-layered institutional safety net of cross-verification, and—crucially—a "responsible party" ready at any moment to have their license revoked and bear the legal consequences.

When people give up chewing over the details and become a rubber stamp that clicks "Approve" on AI output with one click, the entire professional safeguard system instantly vanishes into thin air.

More ironically, on the very same day (October 22), the professional safety research organization METR published a red-team stress test of OpenAI's internal Agent monitoring system.

The test results poured a bucket of cold water on the fanatical tech zealots: in the tested historical Agent architectures, there are multiple vulnerabilities that allow agents to bypass safety monitoring and evade logging.

Illustration of OpenAI's internal Agent monitoring and safety boundary testing

What does this mean?

If an Agent's operations are not fully recorded, or the monitoring system itself has vulnerabilities, how are reviewers supposed to reconstruct what happened?

Conclusion

Anthropic's cold shoulder to newcomers on the employment side and the slap in the face the U.S. courts received on the application side are essentially two sides of the same coin:

We are frantically creating a technology that is "extremely dependent on high-level reviewers"; at the same time, we are killing with our own hands the soil that "cultivates high-level reviewers."

But everything expensive in this world—the intuition accumulated through trial and error, the complex perception of the real world, the sense of responsibility to stake one's reputation and sign one's name at critical moments—has never become one bit cheaper.

This profound structural reshuffle poses an unavoidable question to everyone:

For young people entering the workplace: if your value is merely "running errands, formatting, and doing odd jobs on command," you won't even need to wait until tomorrow to be eliminated.

Rebuilding a new apprenticeship system under human-machine collaboration and proactively designing new hands-on training grounds for newcomers to "find bugs and correct AI" will be the most forward-looking organizational capability investment of the next decade.

For each of us: as tools grow smarter, every "confirm" and "release" you give is no longer a casual action, but the ultimate mortgage of your personal credibility.

Who will train the next generation of eyes capable of seeing through its fallacies? At the end of the road where machines race ahead, who can truly take responsibility for that irreversible outcome?

This is the most severe test the dawn of the intelligence era has given humanity.