The AI Leaders Selling You “Judgment Will Save You” Are Quietly Removing the Only Way to Build It

There’s a reassuring story making the rounds, and every AI leader seems to be telling a version of it. Jensen Huang tells it well: as AI commoditises execution, the scarce, valuable human skills become judgment, asking the sharp question, sensing what’s wrong beneath a polished surface, knowing the purpose behind the task. You won’t lose to AI, the line goes, you’ll lose to the person who uses AI better. It’s become the consensus reassurance every executive repeats from a conference stage, and it’s comforting precisely because it asks nothing uncomfortable of the person saying it.

I think it’s half right and dangerously incomplete, and the missing half is the one that should keep a security leader up at night. Because the same people selling you “judgment is the future” are, with the other hand, dismantling the only machinery that has ever produced it.

Where do you think judgment comes from?

Here’s the question the comforting version never answers. Judgment, the thing everyone now agrees is the scarce and precious skill, is not innate. Nobody is born knowing which alert is a real incident and which is noise, which vendor claim to trust, which line in a report is quietly wrong. That instinct is built, slowly, by doing the unglamorous work. The junior analyst who spends a year grinding through triage learns, without being told, which signals matter and which lie. The graduate who writes a hundred research summaries is quietly absorbing how the organisation actually thinks, which sources are trusted, what the senior people care about versus what they say they care about.

That’s not busywork. It’s the apprenticeship, and it has always been disguised as drudgery. The MIT Technology Review put it precisely: entry-level work is “the economy’s training system.” It’s where junior people learn which numbers can be trusted and how production systems really fail. Strip it out, and you haven’t just removed some cheap labour. You’ve removed the only path anyone has ever walked to become the senior with the judgment you’re now told is irreplaceable.

The story doesn’t survive contact with the hiring data

This isn’t a philosophical worry. It’s happening in the numbers right now, and the numbers are stark.

A Harvard study across 62 million workers found that in firms adopting generative AI, junior employment dropped 9 to 10 percent while senior employment held steady. Stanford found a 16 percent relative decline in employment for 22-to-25-year-olds in the most AI-exposed jobs. Entry-level hiring at the largest tech firms has fallen by more than half in three years. The work that used to train the next generation, the drafting, the triage, the summarising, the routine debugging, is exactly the work being handed to AI first.

And here is the part that exposes the contradiction at the heart of the comforting story. A 2026 Korn Ferry report found that 37 percent of leaders plan to replace entry-level roles with AI outright. In a separate survey, 56 percent of bosses admitted they were lucky to have started their careers before AI, and 43 percent conceded they would never have built the skills they have today if AI had been doing that early work for them. Read that twice. The people preaching that judgment is the irreplaceable human skill know exactly how they built their own, and they are removing that path for everyone coming after them. They know the ladder matters. They’re pulling it up anyway.

The security profession is the canary, not the exception

If you run a security team, this is not someone else’s labour-market story. It is your succession plan, and it’s the sharpest version of the problem, because security judgment is almost entirely tacit.

Think about how your best people got good. Not from a certification. From years in the SOC, working alerts until they could feel when something was off, sitting in incident calls watching seniors make the calls, building the pattern-recognition that lets a veteran glance at a report and say “that’s wrong” without being able to fully explain why. That is the exact kind of judgment Huang is right to call scarce. It is also the exact kind that forms only through the repetitive, low-level work that agentic tools are now absorbing first.

So the uncomfortable question for a CISO is not “will AI replace my analysts.” It’s “if AI does all the Tier 1 work, where does my next Tier 3 come from?” Automate the apprenticeship and you get a profession of architects who have never laid a brick, people nominally in charge of AI-driven systems they never learned to understand from the ground up. The MIT piece names the endgame plainly: firms that automate away the learning stage may improve this quarter’s margins and then find themselves, a decade out, without anyone who understands how their own AI-driven workflows actually behave. For a security function, that isn’t an efficiency problem. It’s a slow-motion loss of the ability to tell when the machine is wrong, which is the entire job.

What to actually do, if you take the warning seriously

The answer is not to refuse AI or to keep juniors doing pointless manual work for its own sake. It’s to be deliberate about preserving the learning that used to happen by accident, because it will no longer happen on its own.

• Treat junior development as infrastructure, not overhead. Your entry-level pipeline is not a cost line to be optimised away the moment AI can do the tasks. It is, in MIT’s phrase, an investment in the future stock of judgment inside your team. Cut it on this year’s margins and you are spending your own seed corn. Decide consciously that some apprenticeship survives the efficiency case.
• Redesign the apprenticeship around questioning the machine, not racing it. The old path, grind the queue until judgment forms, is closing. The new one has to be designed on purpose: pair juniors with the AI tools and spend their time not on doing the triage by hand, but on interrogating the AI’s output, spotting what it missed, understanding why the call matters. The skill you’re cultivating shifts from execution to evaluation, and that has to be taught, not absorbed.
• Make your seniors teach what they can’t quite articulate. The hardest, most valuable judgment in your team lives as instinct in a handful of people’s heads. The window to extract it, to have them mentor, narrate their reasoning, write down why a thing smells wrong, is now, while they’re still here and while there’s still a junior cohort to receive it. Once the ladder’s gone and they retire, that knowledge leaves with them.

Huang and the others are right about the destination: in a world of cheap, fluent, machine-made output, the human who can ask the sharp question and sense the flaw beneath the polish is worth more than ever. What they leave out is that this human is not summoned into existence by a TED talk. They are built, expensively and slowly, by the exact work being automated away first. The comforting story tells you to value judgment. It just forgets to mention that we are busy demolishing the thing that makes it. If you lead a team, that gap is now yours to close, because nobody selling the story is going to close it for you.