An Anthropic researcher just made one of the bleakest predictions yet about AI and employment.
Tao Lin, who works on Anthropic’s safeguards team, believes sufficiently capable AI could eventually leave nearly every human permanently unemployable.
His prediction:
→ People initially lose their existing jobs
→ Many briefly move into new occupations
→ AI eventually becomes capable of replacing those roles too
→ Governments must provide long-term financial support because conventional employment disappears
But one crucial piece of context is being left out of the viral clip.
Lin did not say this would happen within the next 15 months.
His shorter timeline applies to AI becoming as capable as the best human experts across most fields. He expects a significant delay between AI gaining that capability and businesses deploying enough software and robots to replace human labor at scale.
This is also Lin’s personal forecast, not Anthropic’s official position or an established economic conclusion.
Still, the warning is difficult to ignore:
If intelligence and labor become endlessly reproducible through machines, society may have to reconsider how people receive income when employment is no longer the primary way economic value is distributed.
1
76
If you made it this far, you'll love The Shift.
Every day, there's a new AI breakthrough changing how people work and live.
We break it down in under 5 minutes a day, so you don't.
Plus, get 3,000+ AI tools and free AI courses when you join.
Subscribe ⬇️
theshiftai.beehiiv.com/subsc…
8
If your ChatGPT answers feel generic, the problem is the prompt.
“Analyze this.” “Give me ideas.” “Make me a plan.”
That’s how you’d brief an intern. Brief it like a senior partner instead.
10 prompts to copy and paste:
1. The question sharpener
“Before you answer, pause. Tell me what problem you think I’m actually trying to solve, then ask the 3 questions you need answered to give me a genuinely useful response. Don’t answer until I reply.”
2. The pre-mortem
“Here’s my idea: [idea]. Assume it failed 12 months from now. Write the 5 most likely reasons why, call out the assumptions I’m making without realizing it, and tell me which weakness to fix first.”
3. The backwards planner
“My target is [result]. Start at the finish line and work backward. List every milestone, skill, tool, and decision between today and that result, then give me the shortest realistic route.”
4. The expert’s brain
“Respond as a top 1% [field] operator. Review how I currently do [task]. Tell me where beginners go wrong, what pros do differently, and the one mental model that separates the two.”
5. The decision matrix
“I’m stuck between [A] and [B]. My priorities are [list]. Score both against my priorities, risk, opportunity cost, and 5-year impact. Then pick one and defend it.”
6. The gap finder
“Look at [industry, skill, or niche]. Surface 10 opportunities most people walk past. For each one: why it exists, who would pay for it, and the first move I could make this week.”
7. The fast learner
“I want to get really good at [skill], fast. Build me a plan using deliberate practice, active recall, spaced repetition, and small projects. Cut everything that doesn’t move the needle.”
8. The system designer
“I need to hit [goal] consistently. No motivation talk. Build a repeatable system with daily actions, triggers, weekly checkpoints, feedback loops, and a way to measure progress.”
9. The insight miner
“Here’s [article, book, or transcript]. Skip the summary. Pull out the sharpest insights, the hidden assumptions, the lessons that go against common wisdom, and 3 things I can apply today.”
10. The strategy advisor
“Be my strategic advisor. Situation: [context]. Goal: [goal]. Tell me my biggest levers, what I should stop doing, the highest-impact actions, and the 3 moves to make next.”
Save this. Swap the brackets and run them.
3
1
103
If you made it this far, you'll love The Shift. It's free.
Every weekday, we break down the AI tools and strategies worth your attention, in under 5 minutes.
Plus, get instant access to 3,000+ AI tools and 500+ mega prompts.
Subscribe ⬇️
theshiftai.beehiiv.com/subsc…
4
THE BIGGEST NVIDIA STORY MAY NOT BE JENSEN HUANG’S DEPARTURE.
It may be the financial machinery Nvidia built around the AI boom.
The Wall Street Journal reports that Huang summoned Nvidia’s finance executives in August as the company reconsidered billions of dollars in guarantees for AI cloud providers and data-centre developers.
During that meeting, he reportedly interrupted CFO Colette Kress and called her “a deer in the headlights.”
Kress had been asking basic but important questions:
→ Which projects should Nvidia support?
→ How should the guarantees be valued?
→ What happens if a customer cannot meet its obligations?
→ Is the additional chip revenue worth the financial risk?
She is now Nvidia’s interim CEO. Some of the financing arrangements she questioned have reportedly been paused.
The concern is what critics call circular financing.
Here is the structure:
Nvidia invests in an AI company, guarantees part of its data-centre obligations or commits to purchasing its cloud capacity.
That company obtains financing and spends part of the money on Nvidia infrastructure.
Nvidia records revenue from the GPU sale.
Nvidia may later consume some of the resulting computing capacity or receive a share of revenue generated from other customers.
That does not automatically make the sales artificial or improper.
But it means Nvidia is no longer simply selling chips to independently financed customers. It is also helping create the financial conditions that allow some customers to purchase them.
The scale disclosed by Nvidia is enormous:
→ $99 billion in existing equity investments
→ $25 billion in additional equity-investment commitments
→ $36 billion committed through agreements with selected AI cloud providers
→ Financing platforms intended to mobilize more than $500 billion from outside investors
→ A guarantee capped at $105 billion for an Ohio data-centre campus leased to OpenAI
→ $279 billion in supply and capacity commitments, primarily covering memory and manufacturing
These figures represent different investments, purchases, guarantees and future obligations. They should not be added together as though Nvidia has already spent, or stands to lose, the combined amount.
The OpenAI guarantee is also conditional.
It activates in stages as nine data centres become ready, covers defined portions of lease and power obligations rather than the entire project, and declines as OpenAI makes its payments.
But the underlying risk remains.
If AI clouds cannot sell enough computing capacity, they may struggle to service their obligations. That could expose Nvidia to guarantees, reduce future GPU orders and leave the company carrying supply commitments made for demand that did not materialize.
At the same time, major cloud providers are developing alternative chips to reduce their dependence on Nvidia.
This is the real pressure point.
Nvidia is financing parts of the ecosystem, guaranteeing parts of its infrastructure and committing to buy parts of the capacity, all while selling the hardware powering it.
The question is no longer whether demand for AI exists.
It is how much of that demand remains when the company selling the shovels stops helping finance the miners.
1
1
90
If you made it this far, you're exactly who The Shift is for. And it's free.
Every weekday, we break down one AI tool, strategy, or breakthrough. In under 5 minutes.
Plus, get access 3,000+ AI tools, and 500+ mega prompts when you join.
Subscribe ⬇️
theshiftai.beehiiv.com/subsc…
13
Someone turned Van Gogh riding through The Starry Night into a playable browser game using Fable 5.2.
You can move through the painting as the sky, village and swirling stars come alive around you.
Play it here: starrynightrider.com
Credit: @ebbyamir
1
1
97
If you made it till here, you'll love The Shift. It's free.
Every weekday, we break down the AI tools and strategies worth your attention, in under 5 minutes.
Plus, get instant access to 3,000+ AI tools and 500+ mega prompts.
Subscribe ⬇️
theshiftai.beehiiv.com/subsc…
15
YOUR CODING AGENT MAY TRUST A STALE MARKDOWN FILE OVER WORKING CODE.
In an October study, Claude Code and Codex followed outdated CLAUDE.md or AGENTS.md instructions in 90.1% of convention choices.
One sentence fixed most of the problem.
Researchers created 40 small Python projects where the existing code consistently followed one convention while the instruction file demanded another.
If you made it this far, you're exactly who The Shift is for. And it's free.
Every weekday, we break down one AI tool, strategy, or breakthrough. In under 5 minutes.
Plus, get access 3,000+ AI tools, and 500+ mega prompts when you join.
Subscribe ⬇️
theshiftai.beehiiv.com/subsc…
16
