Judgment Work Is Safe (For Now)

A junior developer writing boilerplate code is more exposed to AI displacement than a junior product manager learning to run customer interviews. That's not obvious until you think about what each role is actually building.

The developer is building output. The PM is building a mental model of how people think — and that mental model is what you eventually get paid to apply without anyone checking your work. One role produces artifacts. The other produces judgment. The difference matters more than job category, more than industry, more than whether you've learned to write prompts.

This is the lens I'd apply to any role right now: what percentage of the work is information processing versus judgment? Processing means taking inputs, applying known rules, producing outputs. Judgment means context, relationships, and accountability attached to you specifically — not transferable to a faster, cheaper system. AI has dramatically lowered the cost of processing. It hasn't replaced judgment yet.

The "yet" is doing real work in that sentence. I'll come back to it.

What's actually contracting

Roles built to move information between formats are contracting. Data entry. Tier 1 support running from scripts. Contract review at smaller firms. Research analyst roles where the deliverable is synthesis, not insight. Finance ops built around assembling reports that now assemble themselves.

What makes this different from prior automation waves: the domain translation function is also going. "I explain what the engineer means to the business" is valuable when that gap is expensive to cross. AI narrows the gap. The middle role that lived in it compresses.

Duolingo made this concrete in early 2024 when they cut a significant portion of their contractor workforce — specifically the roles translating content at scale. The economics shifted. They didn't reverse.

What the "judgment is safe" claim gets wrong

Here's where I think the standard take, including earlier versions of my own, is sloppy: it assumes the judgment/processing line is fixed.

It isn't. In 2020, selecting which ML model to use for a problem was judgment work — expertise, intuition, domain knowledge. AutoML and LLMs that write model evaluation code have eaten a substantial portion of that. Not all of it. But enough that the "data scientist freed from SQL to do strategy" story has a sequel nobody's telling: what happens when the strategy work also gets cheaper to do?

I don't think we're there. But treating judgment as a permanent safe zone misreads the dynamic. The line moves. The assembly/judgment distinction describes where we are right now, not a stable partition of human work.

The pipeline problem nobody mentions

There's a harder problem with "judgment roles are growing": the path to senior judgment-having runs through years of doing assembly work. Legal associates do document review. Finance analysts build models. Junior PMs synthesize research. That's how they build the judgment they'll eventually deploy at senior levels.

If AI eliminates those entry-level assembly roles, you don't automatically get more senior judgment-havers in ten years. You get a training gap. The firms cutting associate intake today are making a bet that AI supervision at the junior level produces the same capability development. That bet might be right. It's a serious, open question that most "here's what's safe" posts ignore entirely.

What to actually do

The advice that sounds good — "push into judgment work" — is genuinely hard to act on if your employer's incentive is to take your freed-up time and add more output requirements, not more strategy time.

So the move is more specific. First: identify the two or three places in your job where your context, relationships, and track record produce outcomes nobody else in the org could produce as well. Those are your judgment nodes. Second: make them legible. Most assembly work is visible; most judgment work isn't. If your manager can't point to what you specifically contribute that AI can't replicate, you're in a bad position regardless of whether your work is "actually" judgment work. Third: if your current role has no judgment nodes and shows no path to developing them, that's the signal. The role isn't becoming more strategic. It's becoming a prompt-checking job before it disappears.

The judgment/assembly frame is real and the people ignoring it are making a mistake. But it's a starting point, not a destination — because the line keeps moving, and the question of whether you're ahead of it or behind it is one you have to keep asking.