Joshua Rothman argues the AI jobs apocalypse hasn’t arrived; instead, AI-assisted teams can produce research-grade proposals in a week, causing output quality to converge. The scarce skill becomes judgment and agent orchestration, not drafting. That erodes the link between polished work and perceived talent, pressuring entry‑level roles and forcing managers to evaluate decisions, not deliverables.
What Happened
Joshua Rothman’s “Open Questions” column for The New Yorker, published October 2, 2026, uses a thought experiment about a company launching a new product to explore whether A.I. will still take our jobs. Rothman writes that the jobs apocalypse hasn’t arrived and that something else is happening instead.
His scenario involves you, Frank, and Jill being asked by your boss to draft competing proposals for a new product with a substantial prototype budget. Historically, that meant weeks of labor—calculating market size, surveying competitors, evaluating barriers to entry, estimating manufacturing costs, and working up branding ideas—before anyone wrote a first draft.
In Rothman’s scenario, your team includes Claudette, a recent college graduate who is great at using artificial intelligence. She organizes and records brainstorming sessions, then runs the transcripts through a large language model to generate a research plan. She tweaks it, delegates parts to A.I. agents, and assigns more agents as fact checkers. Then she synthesizes their work.
A first draft, complete with visuals, is done in a week. After the team gives feedback, the revised draft arrives a few days later. Frank’s and Jill’s teams have recent college grads too, and their proposals are also shockingly well researched, with plenty of data and polished writing and images.
Efficiency and quality rose for everyone simultaneously. That creates two sets of problems: one for you, one for your bosses. Managers once read proposals as indirect measures of intelligence, creativity, and ambition—evidence about the people, not just the pitch. That inference becomes far harder when everyone’s work looks this good.
What This Means For You
Your finished output stops being proof of your value. When any competent team can produce a research‑grade proposal in a week, the differentiator moves to judgment: which questions to ask, which claims to verify, which ideas to kill before the boss sees them.
Claudette’s real edge wasn’t prompting. It was orchestration—structuring the brainstorm, splitting the work, building a verification layer, and synthesizing the results. Treat that as a learnable skill, not a personality trait, and start practicing it deliberately on your next small project.
Assume your peers are already doing this. If your team is still drafting by hand while Frank’s and Jill’s are running agent pipelines, you are not competing on talent—you are competing on throughput, and losing.
Keep an audit trail of your thinking. Because a polished deliverable no longer reveals who decided what, document your rejected options, your key edits, and why you overruled the model. When promotion season arrives, that record becomes your evidence.
Watch for your company’s evaluation methods to shift. Expect more structured work samples, process reviews, and reference checks about judgment rather than portfolios of glossy documents. If you manage people, stop asking “who wrote this?”—nobody can answer that now. Ask “who decided this?”
Entry‑level roles face the sharpest pressure and the sharpest upside. The Claudettes of the world gain leverage unusually early, which means junior hiring may concentrate around people who can direct agents rather than just execute tasks.
Expect policy attention to follow. The new U.S. A.I. Czar will likely face questions about displaced entry‑level white‑collar work long before any official unemployment number moves.
Why It Matters
This suggests the change isn’t primarily about tasks disappearing. It is about the sorting mechanism of white‑collar work—the assumption that good output signals good people. When quality converges, that signal degrades.
This could mean a slow squeeze on the rungs that traditionally functioned as talent filters: research, analysis, first drafts, junior consulting. Firms may keep total headcount flat while quietly shrinking the pipelines that once trained future leaders.
The Wall Street AI Jobs Shift to Agent Orchestration Skills coverage described finance roles migrating toward managing agents instead of producing outputs by hand. Rothman’s scenario is that same shift arriving in ordinary offices, minus the job titles.
It also reframes the automation debate. If the apocalypse hasn’t arrived, the risk isn’t sudden mass unemployment—it’s ambiguity about merit, contribution and who deserves the next promotion.
Key Takeaway
- The scarce skill is no longer producing polished work; it is judging, verifying and directing it.
- Orchestrating A.I. agents is transferable and trainable, so start building it before your peers do.
- Managers lose their old proxy for talent and will need new, more explicit evaluation methods.
- Expect quiet pressure on research and drafting headcount even without headline‑grabbing layoffs.
Frequently Asked Questions
Does the column argue A.I. won’t take jobs?
No. Rothman argues the jobs apocalypse hasn’t arrived and that something else is happening instead—a change in how work is produced and evaluated rather than a sudden wave of elimination.
What’s the biggest personal risk here?
Being unable to demonstrate what you personally contributed. When agents do much of the drafting, your documentation of decisions becomes your professional record.
What should a team do first?
Standardize verification, record who owns final judgment on each deliverable, and decide how A.I.-assisted work will be evaluated before review season forces the question.
Leave a Reply