I just finished reading an economic research paper about what happens if AI automates half of cognitive work by 2030. The kind with displacement models, equilibrium equations, and phrases like “occupation-switching frictions”. Like a good prepper, by Sunday night I was pricing soldering stations.
Yes, I know how that sounds. Software engineer panic-buys tools because the robots are coming. Very original. Very 2026. But the research is real, and some of them are genuinely uncomfortable if you pay rent by typing on a keyboard… like me.
So I made a survival plan. Not canned beans and a bunker, thanks god! But more like “skills and assets that still work when the economy restructures around you”. Call it paranoid if you want. I call it “paying the mortgage and not getting kicked out”.
What Paper? Ah, The Paper
The paper is “Economic Scenarios for Transformative AI” by Korinek, Jones, Sacher, Cotter & McCrory (Anthropic Institute, September 2026). It models three scenarios through 2030: modest, substantial, and extreme. The survey median aligns with substantial.
But you are right, preppers don’t plan for average weather. Here’s what the extreme scenario looks like by 2030:
- ✅ GDP is 32% larger than without AI. More wealth than ever.
- ❌ 1 in 5 knowledge workers is unemployed (cognitive employment down 21.5%).
- ❌ Cognitive wages (desk work, coding, analysis) fall 11.5% in real terms.
- ✅ Non-cognitive wages (physical work, trades, hands-on) rise 33.6%.
- ❌ Labor’s share of total income drops from 60% to 45%. Capital’s share rises from 40% to 55%.
- ❌ 90% of AI use is full automation (AI does the job alone, not alongside a human).
- ❌ New human tasks created per task automated: zero.
2/7 are good news. 5/7 are bad news. Also, read that last one again.
The economy booms. But the boom flows to compute, infrastructure, and the people who own them. Money doesn’t flow to people exchaging dollar-per-keystroke. The paper’s own math: labor share falls by a quarter while GDP rises by a third, and 0.45 × 1.32 ≈ 0.60. There’s enough wealth to make everyone better off. But, of course, only if redistribution exists. Without it, knowledge workers get poorer while the economy gets richer.
What about timing? Most divergence between scenarios (modest/substantial/extreme) happens after 2027. The labor market looks almost identical across all three scenarios until mid-2027.
Translation: you have roughly 12–18 months before the worst case starts becoming visible. After that, the “occupation-switching frictions” modeled in the research (slow, painful, competitive) work against you.
This Is Not (Totally) Sci-Fi
Before you close this tab thinking I’ve gone full tinfoil hat, some of these trends are already showing up in 2026 data:
- US labor share of income hit 52.8%, whic is the lowest since records began in 1947.
- AI infrastructure is now 14% of total US private fixed investment, up from 3.3% in 2022.
- Entry-level hiring in AI-exposed fields dropped ~14% since 2022.
- Trades already out-earn the average bachelor’s starting salary by $10–20k/year.
- 550k+ construction positions are unfilled. 94% of contractors report hiring difficulty.
- Expert median estimates for AGI have compressed from ~2060 to ~2033 in six years.
The extreme scenario doesn’t look like a dystopian novel after all.
Non-cognitive wages (physical, hands-on work) rise 33.6% while cognitive wages (desk work, coding, analysis) fall 11.5%. That’s not a rounding error. That’s a structural inversion: physical work becomes the better-paid half of the economy.
This is already happening. Electricians earn a median $63k/year with 9% projected growth. Data center electricians pull $84k (15–25% premium). HVAC techs at $61k with 8% growth. And these numbers predate the AI explosion ($755 billion forecasted for 2026).
If you’re a knowledge worker with zero physical-world competence: in the extreme scenario, you’re competing with 17.9% unemployed knowledge workers for shrinking desk roles. If you have any physical skill (installation, maintenance, farming, equipment repair, cooking professionally), you have an alternative income channel that’s actually growing.
You don’t need to quit your day job and become a full-time electrician. You need one physical competence that can generate income when desk work compresses. Think of it as a parachute, not a career change.
“But Won’t Robots Take Physical Jobs Too?”
Well, yeah, I hear you. If the advice is “learn to use your hands”, and humanoid robots are around the corner, aren’t we all just cooked?
Let’s look at where humanoid robots actually are right now.
As of September 2026, there are only seven verified commercial deployments worldwide. Seven. The largest fleet commitment is Agility Robotics’ 1,000 Digit robots on 3-year contracts. Figure 02 spent 11 months at BMW moving sheet metal parts (1,250 operating hours, 90,000 parts, one task). Two robots. Already retired. Tesla Optimus? Zero external sales. ~1,000 units doing “learning and data collection, not productive work” (Musk’s own words on the Q4 2025 earnings call). Current unit cost: $100k–$150k. And it takes two humanoids to match one human worker’s output (JPMorgan, August 2026).
Every single verified deployment is in a structured, indoor, repetitive environment. Clean factory floors. Warehouse packaging lines. Welding fixtures. Not construction sites, not farms, not someone’s messy building.
Construction sites are a nightmare for humanoids: dust, glare, weather, uneven terrain, layouts that change daily, scaffolding in motion… A Nature paper from 2025 catalogs the problems and concludes that perception systems “optimized for factory conditions often falter when confronted with perpetual variability”. Farming is even worse where battery life is 2–4 hours, soil is soft, there are slopes, insects literally invades robot internals, frost generates water droplets inside the chassis… you name it!
So the physical hedge holds for 2027–2030. Humanoid robotics is roughly 3–5 years behind cognitive AI in deployment maturity. Seven deployments vs millions of people using ChatGPT. The math isn’t close.
When humanoids need to move from clean factory floors into messy reality (farms, construction, field installations, random client sites…), someone needs to design those deployments, collect real-world training data, troubleshoot failures where the environment itself is the variable, and earn trust from the humans working alongside these machines. Every one of those jobs requires a person who has done the physical work and speaks AI.
That person barely exists today. But if you spend 2027–2030 building real physical competence on top of your existing digital skills, you will become that person.
The pattern has precedent. Industrial automation didn’t kill machinists and it didn’t kill software engineers. The winners were the people who could program a PLC and diagnose a jammed conveyor. Same with IoT, where the valuable person is the one who writes the MQTT broker config and the one who wires the sensor and knows why the readings are wrong because the enclosure gets condensation at 4am.
The Prepper Kit (No Bunker Required)
You asked, I delivered: A survival plan. I’m writing it as a list because that’s how my brain works under mild economic panic. Steal whatever’s useful.
1. Twelve months of runway, not six
In the extreme scenario, cognitive unemployment hits 17.9% and switching occupations takes months to years. The Korinek paper models this friction explicitly (it’s slow and painful, not a clean pivot). Six months of savings assumes you’ll find something fast. You might not.
Cut fixed costs. Every subscription, lease, and commitment you drop buys more months of freedom. Think of it as buying time, not saving money.
2. Audit your actual week for AI exposure
Not your job title but your tasks. Anything that boils down to “take this input, apply these rules, produce this output” is high exposure. Writing first-draft code from specs, data analysis, document summarization, template content, financial modeling on structured data… AI already handles these beautifully. Frontier agents score >75% on SWE-Bench Verified for simple coding tasks. That benchmark is now considered saturated.
What’s lower exposure: client trust (somebody needs a face), regulatory sign-off (somebody needs to be sued), physical installation, teaching in-person, creative direction (taste, not execution), cross-cultural negotiation, system architecture under genuine ambiguity.
If more than 60% of your week is high-exposure work, you are in the blast radius. Start migrating now, not when the layoff email lands.
3. Get one physical skill
I really don’t care which one. IoT hardware installation, equipment maintenance, low-voltage wiring, carpentry, farming, professional cooking, on-site photography. Something that requires your body in a specific place doing something a language model cannot do from a data center.
Non-cognitive wages are up 34% in the extreme scenario. The trade labor shortage (550k+ unfilled positions) only gets worse when everyone displaced from desk work tries to pivot at the same time. If you start now, you have a head start over millions of knowledge workers who will discover this insight two years late.
How good to get at that skill? You need enough competence to generate income outside a screen.
4. Turn time-for-money into things-that-earn
Every month you spend building a capital asset (a hosting stack with recurring clients, a content library with distribution, a physical product with repeat buyers, productive land) is a month you’re migrating from the labor side to the capital side. And the capital side is where the money goes in the extreme scenario.
Don’t confuse “running a business” with “owning a capital asset”. A solo consulting practice where you sell hours is still labor with more overhead. A small server with 10 clients paying monthly is a capital asset. The difference: does money flow when you’re asleep?
5. Move from the person who executes to the person who signs off
When 90% of AI use is full automation, the human roles that survive are the ones accountable when it goes wrong.
| Getting automated | What replaces it |
|---|---|
| Write code from spec | Own reliability when it fails at 3am |
| Draft legal documents | Bear liability for what the AI produced |
| Build financial models | Interpret results for humans who need to trust a face |
| Write content | Curate quality, be the filter between AI slop and what readers enjoy |
| Implement infrastructure | Own the uptime SLA. Answer the phone when it breaks. |
The executor-to-owner migration is the single highest-leverage move for most engineers and professionals in high exposure roles.
6. Don’t count on policy to save you in time
Three US bills introduced in 2026: Sanders’ AI Sovereign Wealth Fund, Casar’s AI Tax and Work Protection Act, Buchanan’s AI Tax Integrity Act. Even the World Bank’s WDR 2026 focuses on AI’s impact on labor share.
None of them are law. Not one. The policy conversation is loud and real. The policy… none of both.
In the extreme scenario, displacement arrives faster than democratic institutions can legislate. Build your own floor first. Vote for redistribution, march for it if you feel like doing some exercise, but do not sit around waiting for it.
7. Know the tripwires
You won’t see a headline saying “EXTREME SCENARIO CONFIRMED”. You’ll see it in leading indicators. Check these quarterly:
| Signal | What it means |
|---|---|
| Entry-level job postings in your field collapse (−30%+ YoY) | Displacement wave started. Junior/mid tiers hit first. Senior follows in 12–18 months. |
| Your manager asks for “managed AI ops” instead of “build us X” | They’re cutting headcount, keeping infrastructure. Value shifts to capital-side. |
| Local trades wages jump 15–20% | Non-cognitive premium is real and accelerating. |
| AI agents score >70% on long-horizon benchmarks (SWE-Bench Pro, Marathon) | The capability gap protecting senior engineers is closing. |
| Labor share drops below 50% (BLS/Eurostat quarterly) | Full swing. Hourly income is structurally devalued. |
| Humanoid unit cost drops below $30k AND deployments move beyond warehouses into construction/field work | The non-cognitive wage premium compresses. Physical hedge window is closing. |
If three or more fire in the same quarter: you’re in it. Act accordingly.
The Traps
These look reasonable but get you killed in the extreme scenario.
“I’ll quit and retrain from scratch”: The entry-level market is where AI hits first. Starting over at zero means competing with displaced professionals AND improving AI agents for the same junior roles. Migrate from where you already stand.
“It won’t happen that fast”: Expert AGI forecasts have slid from 2060 to 2033 in six years. Labor share is already at its postwar low. The trend is visible today. “Hope” is not a strategy.
“I’ll just specialize deeper in this AI framework”: Frameworks churn every 6 months. When 90% of use is automated, the framework is irrelevant. Domain knowledge, liability surface, and physical infrastructure are what resist automation.
“My €200k salary is my safety net”: A fat salary that locks you into one city, one employer, one professional identity is maximally fragile when 1 in 5 of your peers gets displaced. Diversified income at €120k with geographic flexibility and capital assets is how you survive a restructuring. Optimize for resilience and employability, not just compensation.
“I’ll stay 100% cognitive”: Non-cognitive wages surge 34%. Pure knowledge workers with no physical skills are competing with 17.9% unemployment in their category and zero new human tasks being created. One physical competence changes the math entirely.
“I’ll wait and see”: The Korinek paper models occupation-switching frictions as slow and painful. The window to switch before it hurts is now through mid-2027. After that, you’re switching alongside millions of other people who also waited and saw.
The Timeline
| When | What’s happening | What you should have ready |
|---|---|---|
| Now – Dec 2026 | Calm before divergence. Entry-level slowing. AI great at short tasks, poor at multi-hour work. | 12 months runway. Task audit done. Physical skill identified. One capital asset in progress. |
| 2027 H1 | Scenarios begin diverging. Long-horizon AI benchmarks improve. Watch job postings and trades wages. | At least one recurring revenue source operational. Physical skill actively developing. |
| 2027 H2 – 2028 | If extreme: cognitive job postings collapse. Trades booked out months. Labor share drops toward 50%. | ≤30% of income from hourly cognitive work. ≥40% from capital/recurring. ≥30% from physical/presence. |
| 2029 – 2030 | Extreme fully manifest. 17.9% cognitive unemployment. Non-cognitive wages +34%. | Positioned. Or in the 17.9%. |
| 2030 – 2032 | Humanoid costs hit $20–30k. Structured physical work (warehouses, factories) starts getting automated. Unstructured work (construction, field, installs) stays human. | If you stacked physical + digital: you’re the person deploying and fixing the robots. If you only went physical: the next wave is coming for structured tasks. Keep moving up the complexity ladder. |
The Asymmetry
The extreme scenario is… well, extreme, but not inevitable or impossible. The substantial scenario (GDP +8.3%, cognitive employment down 3.9%) is the median expectation, and honestly, that’s manageable.
But if you prepare for the worst and get something milder, you’ve built new income lines, physical competence, and capital assets for free. Nothing wasted. If you get the worst and you’re unprepared, you’re competing with millions of displaced knowledge workers for a shrinking pool of desk jobs while trade labor pays more than it has in a generation.
Preppers bet on asymmetry. The cost of preparation is low. The cost of being wrong is not so low.
Now if you’ll excuse me, I have a soldering station to set up.