Open the job ad you posted last year. Odds are it asks for: strong writing skills, research ability, attention to detail in documents, good email communication, maybe a second language.
Now open Anthropic's list of the 50 tasks humanity most often delegates to AI (Economic Index, May 2026):
#1: Search databases and sources for information. #10, #13, #29: Edit and rewrite material. #21: Compose business correspondence. #26: Review documents to respond to requests. #39: Rewrite material into other languages. #40: Adjust formatting and layout.
Your job ad and humanity's delegation list are the same document.
One honest caveat before the panic, and it matters: this dataset measures tasks appearing in conversations โ not jobs, not employment โ and its own methodology says it cannot support job-displacement conclusions. Nobody can tell you from this data that junior roles are dying. What the data does show, unambiguously, is which tasks stopped being scarce. And hiring is the act of buying tasks that used to be scarce.
So the question isn't "should I still hire?" It's "what exactly am I buying now?"
The ad, rewritten
The Job Ad, Rewritten
Same salary line. Different purchase.
What you're actually buying now: the five scarcities
Every line the machine absorbed points at what it didn't. Our read โ practitioner judgment, clearly labelled as such โ on what a salary buys in 2026:
1. Judgment under ambiguity. The top-50 list is full of answering; it contains no deciding. Which customer to call first, which supplier claim smells wrong, when "urgent" isn't โ that's the product now.
2. Verification instinct. Fluent output is infinite; correct output is not. The junior who asks "where did this number come from?" is worth three who format it beautifully. Hire the sceptic.
3. Ownership. A machine cannot be responsible. Someone must sign, follow up, take the call, and carry the outcome. In our industry we'd phrase it harder: a machine can draft the demand letter; it cannot be the person a debtor learns to take seriously.
4. Tool leverage. The best 2026 junior isn't the one competing with the machine โ it's the one operating it: producing what a 2019 team of four produced, then spending the saved hours on points 1โ3.
5. The noticing. No task on the list is "spot what's off." The customer whose orders shrank quietly. The invoice pattern drifting. Every business disaster we've ever been called into was noticeable months earlier โ by a human paying attention.
Three practical hiring changes for owners and CFOs
Change the test, not just the ad. Stop testing writing samples โ you'll be grading a model. Give candidates a flawed AI-generated document and ask them to find what's wrong. The ones who catch the confident error are your shortlist.
Stop paying the language premium. "Trilingual" was worth real money in 2019. The machine translates and drafts in any language now โ pay instead for the judgment to run cross-border relationships, which no tool provides. (What holds for your hires holds for reading counterparties too: flawless English from a foreign supplier or debtor no longer signals sophistication. It signals a subscription.)
Hire earlier for character, later for skills. Skills the machine absorbed depreciated to zero; skills it can't touch โ scepticism, ownership, noticing โ were always character traits wearing a CV. Interview for those explicitly.
The Interview Question Bank, 2026 Edition
One question per scarcity
Data context: Anthropic Economic Index, May 2026 โ tasks, not jobs; judgment is ours.
Frequently asked questions
Should small businesses still hire junior employees in 2026?
Yes โ but for different scarcities. The measured data shows which tasks (lookup, drafting, formatting, translation) stopped being scarce; it explicitly cannot show job displacement. The rational move is hiring for what remained scarce: judgment, verification, ownership, tool leverage and attention.
What skills should I hire for in the AI era?
Judgment under ambiguity, a verification instinct toward fluent-but-unproven output, willingness to own outcomes, the ability to operate AI tools for leverage, and the habit of noticing what nobody asked about.
Is it still worth paying more for multilingual employees?
For the language mechanics alone, largely no โ machine drafting and translation collapsed that premium. What retains value is the judgment to run cross-border relationships, which happens to correlate with language learners but is the actual asset.
How should job interviews change in 2026?
Stop grading writing samples โ you're grading a model. Test instead with flawed AI-generated material to find, prioritisation scenarios to argue, and questions about problems the candidate noticed unprompted.
The bottom line
The salary line didn't change. The purchase did. In 2019, โฌ30,000 bought tasks. In 2026 it buys the five things no delegation list contains โ and the businesses that update the ad first get those people while competitors are still shortlisting for typing.
We rewrote our own job descriptions the same way. The machine drafts our letters beautifully. We hire the people debtors actually answer to. Related reading: Accounts Payable Automation 2026 and Report Inflation.
Recover what's owed. No upfront fees, global reach.
Contact Us โ Free Review