Olivia
Online
Corteiz
@cashcorteiz
Wall Street's biggest players: their money, their mistakes, and how they're betting on AI.
Joined January 2019
27 Following
9 Followers
20 Posts
@quietoraclex Only if every agent runs Opus. In the Flows workshop Anthropic put cheap Haiku agents on the scanning and kept the expensive model for review.
Creator of Claude Code:
"All this engineering stuff that I've learned over the years, so much of it, I just have to throw out"
"Now I just have like armies of agents… it's like a tree of like thousands of agents"
He doesn't use plan mode anymore. About half his engineering now happens from his phone
In 18 minutes he and Claude Code's Head of Product show exactly how Anthropic: → turns every Claude mistake into a CLAUDE.md rule or a skill so Claude can "just run forever" → has routines fix bug reports nobody answered in 5 hours → red-teamed auto mode until they trusted it more than reading every permission prompt
This beats most $1000-$2000 Claude Code courses
Watch it, then read the full guide on writing skills below ↓
A new talk from inside Anthropic here every day.
How to write your own skill, and how to check whether an external skill helps you at all? There are millions of skills available, hardly any was ever tested, and most were vibe-written. The whole skills industry is subjective.
Very proud of this work. https://t.co/bdssLaGpie
Anthropic engineer:
"There's a lot of tweets about how many SaaS companies could have been skills… honestly it's not entirely false"
"If you want to be useful in the industry now, create skills"
Prompt engineering had its run. She says we've moved on to skill engineering
In 21 minutes she shows exactly how Anthropic's Claude Code team:
→ turns any repeated workflow into a skill Claude only loads when it needs it
→ makes a skill run on Sonnet even when you're on Opus
→ runs evals to check your code does better with the skill than without it
This beats any $1000-$2000 course on Claude Code I've seen
Watch it ↓
REA is the most dangerous repo on GitHub right now
It gives your AI agent a real disassembler, so it reads any app down to the binary
Every claim comes back with the address behind it
I broke down the exact 8 steps: https://t.co/eaeZ4OFJrH
Anthropic engineers:
"Obviously, the agents are way smarter than me these days"
"People were walking around with their laptops slightly open so that their Claude Code wouldn't die"
Their best tip sounds backwards. Be less specific at the start and let Claude design the team itself
In 24 minutes they show exactly how Anthropic:
→ lets Claude plan and run its own team of up to 1,000 agents
→ caught 88% of a day's real market events with a Flow vs 52% with one agent
→ puts cheap Haiku agents on the scan and Fable on the review
This beats any $1000-$2000 course on agents I've seen
Watch it, then read the full guide below ↓
REA is the most dangerous repo on GitHub right now
It gives your AI agent a real disassembler, so it reads any app down to the binary
Every claim comes back with the address behind it
I broke down the exact 8 steps: https://t.co/eaeZ4OFJrH
Anthropic engineer on Claude Fable:
"Things that would have taken me weeks I could do in hours"
"How can you not laugh, but also how can you not cry, honestly?"
He loved writing code by hand. Now he says he can't go back
In 18 minutes he shows exactly how Anthropic:
→ cut 80% of Claude Code's system prompt for smarter models
→ makes Claude find the unknowns you forgot to tell it
→ gets Claude to ask you 40 questions about your spec
This beats any $1000-$2000 course on agents I've seen
Watch it, then read the full guide below ↓
https://t.co/X1bBruM26X
Claude Code engineer:
"I used to get a lot of joy from performance engineering... I think Claude is much better at that than I am now"
"70%, 80% of my work happens on Claude Tag now"
In 22 minutes the team behind Claude Code shows exactly how Anthropic runs agents in the cloud, reviews code with a swarm of Claudes and lets Claude build its own harnesses
This beats any $1000-$2000 course on agents I've seen
Watch it, then read the full guide below ↓
REA is the most dangerous repo on GitHub right now
It gives your AI agent a real disassembler, so it reads any app down to the binary
Every claim comes back with the address behind it
I broke down the exact 8 steps: https://t.co/eaeZ4OFJrH
Anthropic engineer:
"Coding agents shouldn't wait for you to press enter to get started"
"You're a single /schedule command away from creating your very first routine"
In 22 minutes she shows exactly how Anthropic builds agents that wake up on their own, catch what breaks and open pull requests while your laptop is closed
Weekly PRs for Claude Code are up 200% this year
This beats any $1000-$2000 course on agents I've seen
Watch it, then read the full guide to building an agent factory below ↓
Stop letting unfinished Opus 5.5 runs eat your budget
Set up the harness before your next long task
Save progress, verify results, reuse the workflow
The full 7-layer setup below ↓ https://t.co/rz78goHxfw
Creator of Claude Code:
"Who writes 100% of their code using AI?"
"Now it's roughly, I think like 70% of people write 100% at these very small startups"
In 18 minutes, he and Stripe's CEO show exactly how Stripe ships with Claude Code: one engineer merged 600+ AI-written pull requests, 2-3 engineers did 6 months of work in 8 weeks, and new businesses on Stripe doubled
This beats any $1000-$2000 course on AI I've seen
Watch it, then read how one AI startup just hit a $1B run rate below ↓
https://t.co/nByTwj2JPX
Mark Zuckerberg on committing up to $200 billion to AI before he had even seen the model:
"We haven't seen the models yet... we're going to write this check for like $100 billion or $200 billion. I'm like, Alex, are you sure this is going to work?"
Alex is Alexandr Wang, Meta's AI chief. The CFO kept pulling him aside:
"If this doesn't work... it's going to be very, very bad in financial terms. I really need to impress on you that this model has to work"
Zuckerberg on the price:
"To operate a gigawatt a year, call it on the order of $15 billion"
Building one: "$50 or $60 billion"
When people at Meta wanted to slow down:
"No, we're doing this... We're going to burn the boats"
Why he wasn't scared:
"For it not to work out, it's not just that we have to fail. AI overall has to fail"
Yesterday Meta launched the model. Muse Wonder
Full 7-min clip ↓
Karpathy says only 0.2% of people have seen what AI agents can actually do
The man who built Claude Code is in that 0.2%. He ships 10 to 30 pull requests a day and hasn't written a single line of them by hand
When Boris Cherny first showed Claude Code inside Anthropic, it got 2 likes. A year later, 4% of all public commits on GitHub were written by it
Share of his own code written by Claude: 20% → 30% → 100% in 9 months
step 1 → strongest model, max effort. A cheaper one needs more fixing and burns more tokens in the end
step 2 → start in plan mode. 80% of his tasks begin with one line added to the prompt: "Please don't write any code yet"
step 3 → once the plan looks good, auto-accept. He says it gets it right the first time almost every time
step 4 → run agents in parallel. He starts a batch the moment he wakes up and always has about five running
step 5 → Claude reviews the code. At Anthropic it checks 100% of pull requests, then a human looks
Asked if people should still learn to code:
"In a year or two it's not going to matter"
Most people still type one prompt and wait for one answer. He runs five at a time
Full 4-min clip with subtitles ↓
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel.
- Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all.
- Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract.
- Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code.
- And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while.
The funnel is driven by a combination of factors:
1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel.
2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel.
3. Access. Free-tier, paid-tier, internal.
So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.
Put $500 a month into the market for 40 years and it grows to about $1.5 million.
Most people end up with $900K. Where the other $600K goes is the most expensive thing nobody explains to you.
Jack Bogle, the founder of Vanguard, explained it in one sentence. The market returns about 8% a year. Fees, trading and middlemen take about 2%. You get 6%:
"We put up 100% of the capital, we take 100% of the risk, and we get about 30% of the market's long-term return."
Then he gave his rules for regular investors. A few of them:
"Time is your friend." At 7%, money doubles about every 10 years.
"Forget the needle, buy the haystack." Stop hunting for a genius fund manager. Own the whole market at low cost.
"There's no escaping risk." Even a savings account loses money when inflation is higher than the interest.
And the one most people break, buying when they feel great at the top and selling when they're scared at the bottom:
"Don't do something, just stand there."
Full 12-min clip with subtitles ↓
This CEO shrank his company from 7,400 people to 3,000 using AI. Now he says other tech CEOs aren't being straight about what's coming for your job.
His name is Sebastian Siemiatkowski. He runs Klarna, the "buy now, pay later" app with over 100 million customers.
In 2024, Klarna said its AI assistant was doing the work of 700 customer service agents. Then customers started asking for real people, and Klarna brought humans back into support.
He hasn't slowed down:
"I feel a lot of my tech bros are being slightly... not to the point on this topic. I think there is a massive shift coming to knowledge work."
His example:
"There are 8,000 people in Brussels that work as translators, and I would argue that most of that can be done by AI today already."
At Klarna, revenue per employee went from $400K to over $1 million in two years. He says they avoided layoffs by simply not hiring.
His advice if you're worried:
"Embrace it, learn it, educate yourself."
Full 6-min clip with subtitles ↓
The man who predicted the 2008 crash says the place most people keep their savings is "the worst investment" over time.
His name is Ray Dalio. He built the biggest hedge fund in the world.
He's talking about your bank account:
"People think that that's the safest. It's not. It's the worst investment over a long, long period of time because inflation will eat it away."
Prices rise about 3.5% to 4% a year, he says. A savings account pays you around the same, and then you pay tax on the interest. Your balance goes up, but what it can buy doesn't.
The question he says matters most:
"How many years can I live if money doesn't come in?"
Then he was asked about a 30-year-old with almost no savings:
"Your only asset is yourself."
And with AI replacing more jobs, he says that's getting harder for everyone outside the top:
"If you're in the top 10% of talent, let's say the world's your oyster."
Full 8-min clip with subtitles ↓
https://t.co/XwWKTvVPuY
The man who made the first big bet on OpenAI says that in about 4 years, AI will be able to do two-thirds of all jobs. Doctors and accountants included.
His name is Vinod Khosla. In 2019 he put $50 million into OpenAI at a $1 billion valuation, the largest bet he'd made in 40 years. He says it's now worth "a few hundred billion dollars."
His forecast, in his words:
"...two thirds of all jobs will be capable of being done by an AI. So whether you're a physician, whether you're a radiologist, whether you're accountant, whether you're a chip designer, whether you're a salesperson, AI will do your job better."
Then the money part:
"15 trillion of US GDP is labor, 15 trillion that would mostly go away."
What he thinks it does to prices by 2040:
"Maybe $10,000 will buy much more than you can buy if you have $100,000 income today."
College, he says, is "goodbye." Healthcare outside of procedures like heart surgery becomes "near free."
And his warning:
"You can't leave 80% of the population behind. They will revoke capitalism if that happens."
Full 5-min clip ↓
One man found the biggest Ponzi scheme in history 8 years before it collapsed.
He warned the government 5 times. Nobody stopped it, and thousands of people lost their retirement savings.
His name is Harry Markopolos, a Boston portfolio manager. The fraud belonged to Bernie Madoff, "one of the most powerful men on Wall Street."
When Markopolos first went to the SEC in 2000, he says it was a fraud of under $7 billion. By 2008 it was $50 billion.
He sent his reports unsigned while his team tracked Madoff in secret:
"Each of us feared for our lives."
Under oath in Congress, February 2009:
"I gift wrapped and delivered the largest Ponzi scheme in history to them and somehow they couldn't be bothered to conduct a thorough and proper investigation..."
"If a $50 billion Ponzi scheme doesn't make the SEC's priority list, then I want to know who sets their priorities."
In 1991, Ed Thorp spotted the same fraud and quietly pulled his client out. Markopolos spent nine years trying to get someone to listen.
Full 9-min testimony ↓
Ed Thorp looked at Bernie Madoff's trades in 1991 and found half of them never happened.
Madoff wasn't caught for another 17 years.
Thorp is the math professor who beat Vegas at blackjack. He was asked to review McKinsey's pension plan. One fund paid 1-2% a month, every month, with a record going back to the late '60s.
Peter Madoff wouldn't let him in the door. So he checked the trades against the exchanges:
"Half the trades never happened... Another quarter of the trades had so much volume that the volume couldn't have happened."
He asked a Bear Stearns VP to find who was on the other side of 10 Madoff options trades. There was no trace of Madoff.
He told McKinsey: "This is a fraud." They were making 20% a year with Madoff and 16% everywhere else. If Thorp was wrong, leaving cost them 4 points. If he was right, it saved their jobs.
They pulled out within two months.
The advisor who brought Thorp in heard the whole analysis. He kept raising money for Madoff until the week it blew up in 2008, with his family's money still inside.
Full 8-min clip below
The biggest IPO in history could be six weeks away. Anthropic is reportedly aiming for a valuation of up to $2 trillion.
Two months ago, Ray Dalio was asked if AI is a bubble. Steven Bartlett played him Jeremy Grantham calling it "the biggest investment bubble in American history."
Dalio: "He's right."
Then he explained how it ends:
"Wealth is not the same as money. So you see a lot of people getting wealthy, but you can't spend the wealth. You have to sell the wealth to get money."
On what pricks a bubble:
"There's almost nothing that's easier to produce than stock."
When the market wants stock, companies make more of it. That new supply, plus the need to raise cash, is what pops it.
Asked if he sees the signs now: "Yeah. Yeah. Yeah."
Another sign he named is weak hands buying with leverage:
"It's more like they're crap shooting."
Full 12-min clip ↓
Ed Thorp looked at Bernie Madoff's trades in 1991 and found half of them never happened.
Madoff wasn't caught for another 17 years.
Thorp is the math professor who beat Vegas at blackjack. He was asked to review McKinsey's pension plan. One fund paid 1-2% a month, every month, with a record going back to the late '60s.
Peter Madoff wouldn't let him in the door. So he checked the trades against the exchanges:
"Half the trades never happened... Another quarter of the trades had so much volume that the volume couldn't have happened."
He asked a Bear Stearns VP to find who was on the other side of 10 Madoff options trades. There was no trace of Madoff.
He told McKinsey: "This is a fraud." They were making 20% a year with Madoff and 16% everywhere else. If Thorp was wrong, leaving cost them 4 points. If he was right, it saved their jobs.
They pulled out within two months.
The advisor who brought Thorp in heard the whole analysis. He kept raising money for Madoff until the week it blew up in 2008, with his family's money still inside.
Full 8-min clip below
Ed Thorp looked at Bernie Madoff's trades in 1991 and found half of them never happened.
Madoff wasn't caught for another 17 years.
Thorp is the math professor who beat Vegas at blackjack. He was asked to review McKinsey's pension plan. One fund paid 1-2% a month, every month, with a record going back to the late '60s.
Peter Madoff wouldn't let him in the door. So he checked the trades against the exchanges:
"Half the trades never happened... Another quarter of the trades had so much volume that the volume couldn't have happened."
He asked a Bear Stearns VP to find who was on the other side of 10 Madoff options trades. There was no trace of Madoff.
He told McKinsey: "This is a fraud." They were making 20% a year with Madoff and 16% everywhere else. If Thorp was wrong, leaving cost them 4 points. If he was right, it saved their jobs.
They pulled out within two months.
The advisor who brought Thorp in heard the whole analysis. He kept raising money for Madoff until the week it blew up in 2008, with his family's money still inside.
Full 8-min clip below
Masayoshi Son sold every Nvidia share SoftBank owned, $5.8 billion, to put more money into OpenAI.
On stage in Tokyo he was asked why:
"I wish to have unlimited money... I don't want to sell a single share... I was crying to sell Nvidia share."
Then came the bubble question:
"The people who ask that question is not smart enough. That's my straight answer."
He put numbers on it. $10 trillion of AI capex over 10 years. If AI and robots replace 10% of global GDP, he says that's $20 trillion a year:
"10 years of cumulative investment recouped in six months... Where is the bubble?"
Last month SoftBank sold the largest junk bond in history, $11 billion at up to 9.75%, to keep funding the bet. Its OpenAI stake is heading toward $65 billion.
In 2000 he lost almost $60 billion when the last tech bubble burst.
Full 4-min clip ↓
Masayoshi Son sold every Nvidia share SoftBank owned, $5.8 billion, to put more money into OpenAI.
On stage in Tokyo he was asked why:
"I wish to have unlimited money... I don't want to sell a single share... I was crying to sell Nvidia share."
Then came the bubble question:
"The people who ask that question is not smart enough. That's my straight answer."
He put numbers on it. $10 trillion of AI capex over 10 years. If AI and robots replace 10% of global GDP, he says that's $20 trillion a year:
"10 years of cumulative investment recouped in six months... Where is the bubble?"
Last month SoftBank sold the largest junk bond in history, $11 billion at up to 9.75%, to keep funding the bet. Its OpenAI stake is heading toward $65 billion.
In 2000 he lost almost $60 billion when the last tech bubble burst.
Full 4-min clip ↓
Stanley Druckenmiller went 30 years without a losing year.
Then he walked Morgan Stanley through the two biggest mistakes of his career, and what they did to him physically.
Mistake #1, Nvidia. He'd said publicly he couldn't see himself selling for 2-3 years. Then:
"I couldn't stand success. I'd gone from 150 to 800... I couldn't deal with it and I sold it."
That's $80 split-adjusted. Nvidia is around $229 today.
Mistake #2, the 2000 bubble:
"Everyone knows how I played the Nasdaq melt up in '99. Sold it perfectly in January, then bought the exact top."
Asked what he learned from it, he said "nothing":
"I learned not to do that 20 years before, but I got emotional, which I fight every day."
What it cost him:
"I would literally... throw up once or twice a week, just from anxiety when I'd have a drawdown"
Imposter syndrome for 15 years? "Yes. Maybe longer."
Full 12-min clip ↓
Stanley Druckenmiller went 30 years without a losing year.
Then he walked Morgan Stanley through the two biggest mistakes of his career, and what they did to him physically.
Mistake #1, Nvidia. He'd said publicly he couldn't see himself selling for 2-3 years. Then:
"I couldn't stand success. I'd gone from 150 to 800... I couldn't deal with it and I sold it."
That's $80 split-adjusted. Nvidia is around $229 today.
Mistake #2, the 2000 bubble:
"Everyone knows how I played the Nasdaq melt up in '99. Sold it perfectly in January, then bought the exact top."
Asked what he learned from it, he said "nothing":
"I learned not to do that 20 years before, but I got emotional, which I fight every day."
What it cost him:
"I would literally... throw up once or twice a week, just from anxiety when I'd have a drawdown"
Imposter syndrome for 15 years? "Yes. Maybe longer."
Full 12-min clip ↓
"NVDA gone from $150 to $800 I was long-term in it, and I couldn't deal with it, so I sold. Then it was $1,400 like 5 weeks later, and I was sick," Druckenmiller said.
Most Popular Users
Elon Musk
@elonmusk
241.5M followers
Barack Obama
@barackobama
118.8M followers
Cristiano Ronaldo
@cristiano
114.5M followers
Donald J. Trump
@realdonaldtrump
111.8M followers
Narendra Modi
@narendramodi
107.1M followers
Rihanna
@rihanna
98.7M followers
NASA
@nasa
92.3M followers
Justin Bieber
@justinbieber
91.8M followers
KATY PERRY
@katyperry
90M followers
Taylor Swift
@taylorswift13
84M followers
Lady Gaga
@ladygaga
75.5M followers
Virat Kohli
@imvkohli
73.5M followers
Kim Kardashian
@kimkardashian
70.9M followers
YouTube
@youtube
68.8M followers
Neymar Jr
@neymarjr
66.5M followers
Bill Gates
@billgates
65.3M followers
Selena Gomez
@selenagomez
63.1M followers
The Ellen Show
@theellenshow
62.2M followers
CNN
@cnn
61.7M followers
X
@x
60.7M followers