Future of work · Updated October 2026
The work is changing. The thinking still counts.
AI is taking on tasks, not whole jobs, and the first changes are showing up in entry-level work. This section shows what the evidence says, which human skills employers name most, and free lessons and labs to practice them.
How to read this page
Evidence labels
Every claim carries a label. Documented means published research or an official document. Reported means a news outlet, survey or official source reports it. Suggestive / early means early or thin evidence, or a study that measured something narrower than the claim. Claimed means the person or company says it, with no independent check.
01 · What’s changing
Three things the evidence actually shows
Headlines about AI and jobs mix very different kinds of numbers. Here are three that come from careful work, each with its label and its limits.
Predict first
Which group of workers do you think AI has affected first: new graduates, or experienced workers in the same job?
Show the answer
New graduates, so far. In U.S. payroll data, employment of 22- to 25-year-olds in the jobs most exposed to AI has fallen well behind their peers in less-exposed jobs, while experienced workers in the same jobs show no comparable gap. Most of it comes from companies hiring fewer young people, not from firing. The researchers call this an early warning sign, not proof that AI caused it. See card 2.
1 · Tasks, not jobs
Most jobs are more likely to be helped than replaced
A job is a bundle of tasks. The International Labour Organization rated every task in every job and concluded that generative AI’s main effect is likely to augment work (a person does the job with AI help) rather than automate it. Clerical work, such as data entry and scheduling, is the most exposed.
Documented
2 · The entry-level squeeze
Young workers in AI-exposed jobs have fallen behind
Stanford economists tracked millions of U.S. paychecks through June 2026. Employment of 22- to 25-year-olds in the most AI-exposed jobs is 19% below where it would be had it kept pace with less-exposed peers. Older workers in the same jobs show no such gap. The drop is concentrated where AI replaces tasks; where it helps workers, employment is flat or rising.
Documented, early · working paper
The authors call these early signs, not proof of cause, and found no economy-wide job loss. An earlier version (August 2025) reported a smaller gap.
3 · Skills employers name
Analytical thinking tops the list
In the World Economic Forum’s survey of more than 1,000 large employers, seven in ten called analytical thinking essential, the top core skill. Next came resilience, flexibility and agility. Employers expect 39% of workers’ core skills to change by 2030.
Reported · employer survey
This is what managers of big companies say they expect. It is not a measurement of what happens.
02 · Two skills that travel
Judgment and agility
Whatever job you end up in, two kinds of thinking keep showing up: deciding well when you can’t be sure, and learning things nobody has taught you yet. Both can be practiced.
Judgment
Deciding well when you can’t know for sure
- Saying out loud what you are assuming
- Putting a number on how sure you are (calibration)
- Weighing trade-offs when facts are missing
- Imagining the plan already failed and asking why (a pre-mortem)
- Spotting what a fix will break next (a second-order effect)
Agility
Learning what nobody has taught you yet
- Borrowing an idea from a different field (transfer)
- Changing a plan when the rules change, and keeping what still works
- Learning something new fast, then teaching it back
- Saying “I don’t know yet,” and finding out
We teach the behaviors. We don’t claim to measure “learning agility” as a trait. The term comes from corporate talent management, and research on measuring it has not kept up with its use. Claimed · early research Nothing on ARQIV scores or certifies a student’s judgment or agility.
03 · Practice it
Labs you can try right now
Hand-authored logic that runs in your browser. Nothing here is AI-generated, nothing is sent anywhere, and you need no account.
Judgment · Solo or class
How Sure Are You?
Answer true-or-false questions, say how sure you are, and see whether your 90% really means 90%.
Grades 7+10–15 min
Judgment + agility · Classroom game
The Black Box
Something is failing and the cause is hidden. Spend a budget of 12 questions, name the cause, then pick a fix that won’t backfire.
Grades 8+30–45 minTeams 1–5
Agility · Solo
Same Bones
Two cases from different worlds share one idea. Find it, then use it to predict a third case.
Grades 7+10–15 min
Review · Solo
Your AI Path
A map of ARQIV’s AI topics with short checks and a Daily Five review, saved only on your device.
Grades 7+5 min a day
Agility · Solo or class
Rule Change
Build a plan, then watch the rules change mid-game. Score points for what you save, not for polish.
Grades 7+10 min
If you have played a lab on this device, a small progress label appears on its card. It is read from this browser only.
04 · Teach it
Decide & Adapt: a six-week unit
The Think-First routine in decision mode. Each 45-minute lesson works with no student devices and grades the thinking, not the final answer.
The 60% Problem
Decide with missing factsWeek 2How Sure Are You?
Match confidence to evidenceWeek 3Kill the Plan
Find the failure before it happensWeek 4Same Bones
Borrow from another fieldWeek 5The Rules Just Changed
Change the plan, keep the pointWeek 6Learn It in a Period
Learn fast, teach it back05 · Prove it
Proof, not points
The Challenge Chart lists twelve things a student can do in front of real people. A teacher signs off. No points, no prizes, no ranking.
Sample challenge
Catch a chatbot being wrong
Catch a chatbot stating something false and prove it with two independent sources.
Sample challenge
Run a pre-mortem
Lead a 10-minute pre-mortem for a real class or club plan and hand over the fixes.
Sample challenge
Defend a decision
Give a 3-minute decision defense to a panel of adults, and say what would change your mind.
06 · The evidence
What the evidence says about learning these skills
Some ways of practicing have strong research behind them. Others are popular but thinly tested. We built the labs and lessons on the first kind and label the second.
| Method | What it trains | What the evidence shows | Label |
|---|---|---|---|
| Try the problem before the lesson (productive failure) | Learning agility | Across 53 studies, trying first then being taught beat being taught first, by a moderate amount. It reversed for grades 2–5 and for general thinking skills. | Documented |
| Compare two cases from different fields | Transfer | Students who compared two cases were nearly three times as likely to use the idea later as students who studied the cases one at a time. Repeated in later experiments. | Documented |
| Say how sure you are, then get feedback (calibration) | Judgment | Under an hour of training improved adult forecasters’ accuracy by 6 to 11%. Not yet tested with students. | Documented · adult forecasters |
| Pre-mortem | Judgment | Imagining a failure already happened produced about 30% more reasons. Nobody measured whether decisions got better. | Suggestive |
| Quizzing yourself and spacing out review | Remembering | A review of ten study techniques rated these two the most useful. | Documented |
| “Learning agility” as a measurable trait | — | A corporate idea; measurement tools outran the research. | Claimed · early research |
07 · Counselors and families
Career conversations for an AI world
A 20-minute conversation guide for counselors and advisory teachers, and three things to say at home.
For counselors
Career conversations
Pick a job, sort its tasks, and ask which ones a student would want to be accountable for. Ten printable questions.
Open the counselor guide →
At home
- Ask “which parts of that job would you want to be responsible for?” instead of “will AI take that job?”
- When your child is sure about something, ask “how sure, out of 100?” and check together later.
- Praise changing your mind for a good reason, and tell students that doing it well is a skill.
08 · Where these ideas came from
What we borrowed from Alpha School, and what we left behind
Alpha School, a private network in Texas, gets attention for its software-led mornings. Its learning claims are its own and unaudited Claimed, and reporters have described real problems Reported. Several of its ingredients are well supported on their own. We kept those and refused the rest.
Kept
- Mastery checks before moving on, with “Skip for now” always available
- Feedback with an explanation on every answer
- Spaced review of things you got wrong
- Visible progress, saved only on your device
- Proof in front of real people: the Challenge Chart
Left behind
- Webcams, screen recording or any tracking
- Questions written by AI that nobody checked
- Rewards that work like money
- Pressure: countdowns, daily-use counters, rankings
- Unaudited success claims
Honest limits
What this page leaves out
Read the numbers above with these in mind.
Not covered here
Wages, unions and labor law; countries outside the U.S. for the entry-level data; which specific jobs to choose. We give no salary figures, “safest jobs” lists or predictions. The entry-level finding is one research team’s analysis of one payroll company’s data, revised several times, and the authors say it shows a pattern, not a cause. The skills survey asks employers what they expect. The evidence on teaching judgment comes mostly from adults and from lab studies, not from school classrooms.
Review status
Written by ARQIV. Initial edition, October 2026. These pages have not been classroom-piloted or independently reviewed yet, and every question and scenario is hand-authored and checked against its sources by ARQIV. Tell us about anything wrong or unclear.
Sources
Where this comes from
Open the source before quoting a number.
- Paweł Gmyrek, Janine Berg and David Bescond, "Generative AI and Jobs," ILO Working Paper 96 (August 2023).
- Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," Stanford Digital Economy Lab working paper, revised August 12, 2026 (first version August 2025).
- World Economic Forum, Future of Jobs Report 2025 (January 2025), chapter 3, Skills outlook.
- Steph Brown, “Analytical thinking remains top core skill for employers,” FM Magazine, February 18, 2025.
- Tanmay Sinha and Manu Kapur, "When Problem Solving Followed by Instruction Works: Evidence for Productive Failure," Review of Educational Research 91(5), 2021.
- Jeffrey Loewenstein, Leigh Thompson and Dedre Gentner, "Analogical encoding facilitates knowledge transfer in negotiation," Psychonomic Bulletin & Review 6(4), 1999; see also Gentner, Loewenstein and Thompson, Journal of Educational Psychology 95(2), 2003.
- Welton Chang, Eva Chen, Barbara Mellers and Philip Tetlock, "Developing expert political judgment," Judgment and Decision Making 11(5), 2016, pp. 509–526.
- Mitchell, Russo and Pennington (1989), as reviewed by Jason Collins, "Premortem"; Gary Klein, "Performing a Project Premortem," Harvard Business Review (2007).
- John Dunlosky and colleagues, “Improving Students’ Learning With Effective Learning Techniques,” Psychological Science in the Public Interest 14(1), 2013; Association for Psychological Science press release “Which Study Strategies Make the Grade?”.
- Deeper Signals, "What is learning agility?" (a vendor explainer that summarizes the research, including DeRue, Ashford and Myers 2012).
- ARQIV, “What other schools are trying, and what happened” (Alpha School review, with its own sources).
All links opened and checked October 21, 2026 (accessed October 2026).
AI words on this page
Calibration, Base rate, Generative AI, AI literacy. New entries for pre-mortem, second-order effect, transfer, mastery learning and spaced repetition are being added.
Start here
Practice deciding under uncertainty
Ten true-or-false questions, one confidence number each. About 10 minutes.