I HAD TO SIT DOWN, I POINTED A GROK BOT AT OTHER PEOPLE'S LIQUIDATION LEVELS AND TOLD IT TO WAIT THERE 8.0 SOL turned into 255.9 in a single session. +3,199%. 36 sweeps. Hit rate 64%. Average hold 176 seconds. Book latency 2 milliseconds. Nothing entered by hand. The map is the product. Long liquidations stacked below spot, short liquidations piled above, and the bot parks at the edge of the cluster instead of guessing direction. $8,740,000 of forced selling swept past it today. It caught 36 of those wicks. The list reads like a casualty report. Short at $109.92 for $16,400. Short at $109.70 for $369,200. Long at $102.00 for $420,700. Every one of them a stop that had to fill at any price. Best single sweep paid 32.1 SOL. Average position lasted under three minutes. It runs 12x leverage and finished the session flat, holding one short at $101.79 that is up 28.38 SOL. The 24 hour price is down 2.25%. The account is up 3,199%. Direction was never the trade. It made nothing predicting the market. It made everything waiting where people had to sell. 36 sweeps. $8,740,000 liquidated. 255.9 SOL harvested.

NO WAY, I GAVE A GROK BOT AGENT $2,000 AND TOLD IT TO BUILD A CITY WHERE EVERY TOWER IS A BET Sixteen days later the skyline is worth $164,300. +8,115%. 14 towers standing, 9 demolished. 1,200 events in the trace. Fully autonomous, relay 39 milliseconds, nothing approved by me. Tower height is capital at risk. That is the whole interface. The tallest one on the board carries $19,400 and glows because it is up 83.4% unrealised. Districts split the risk: macro $77,682, energy $35,785, tech $33,365, culture down 12.7% and still standing. Nine got razed. The trace calls it demolition and gives a reason. One line just says the thesis cracked, minus $2,396, and moves on. It caps itself. When macro hit 17% of book the agent capped the district and stopped pouring. It widened its own spread from 4.7 cents to 6.8 rather than chase a fill. Deepest cut across sixteen days was 8.8%. The stat I keep going back to is quarter Kelly. It never committed more than $2,080 to a single floor. A skyline does not fall because one tower was wrong. It falls because one tower was too tall. 14 built. 9 razed. $164,300 standing.

The most advanced agent stack of 2026 is being used to order dinner DoorDash sits in the chapter list, one slot after coding, in a 21-minute walkthrough of 11 use cases. Look at what shares that list. Email. Calendar. Browser tasks. Meeting summaries. Six chores and one engineering task. Then look at what runs them. Each agent holds its own cloud computer, with a browser, a file system and a terminal. It stays awake when the laptop shuts, and it stays signed into accounts because the session lives on the machine instead of in a chat window. Infrastructure built for autonomy, aimed at an inbox. The chores were never hard. Each one takes 8 minutes and comes back tomorrow, and that is the only thing on the list that compounds. Benchmarks measure code because code is measurable. Nobody scores an agent on the 40 tasks that never reached a ticket.

Trading educator, $10,000 and 22 hours with Grok Bot: "Can it actually trade?" Everyone is testing agents on code they can review. Almost no one is testing them on money they can lose. 21:18, 18 chapters, 2 separate challenges run back to back. $10,000 handed to an agent with its own cloud computer. 22 hours of unattended runtime, no one watching the screen. 1 question the title answers and the thumbnail does not. 18 chapters, and the P&L lands in only one of them. Full run below. Chapter 1 is where he sets the rules before

GTM lead, SpaceX: "Grok Bot for GTM" Everyone is benchmarking agents on pull requests. Almost no one is benchmarking them on pipeline. A full session recorded on Zoom, screen and laptop in frame. The coding demos own the timeline. The GTM demo is the one revenue teams will clone first. 1 agent stack, pointed at the side of the business that gets measured in dollars. Coding agents got the audience. GTM agents get the budget. Full session below. Watch which department she points it at

AI tools reviewer: "I gave it one hour and three real tests and it impressed me" Everyone's talking about what Grok Bot can do. Almost no one is asking who owns the agent after it does it. 9 out of 10 people rent the whole stack and never check whose keys are in it. In a 12 minute test, a reviewer runs xAI's $200/month AI team through 3 real jobs, then spends the second half on the question that changed his answer. 3 minutes 24 seconds in, the comparison starts. $200 a month against a server he owns. 2 minutes 22 seconds later he has the open source version deployed in 1 click. The architecture block at 9:12 is the part nobody covers. Relay, DB, Redis. Three pieces, and you can see exactly which one holds your data. 6:44 gives the VPS specs. Not a range, actual numbers. Both agents run 24/7. Only one of them runs on hardware you can unplug. More value than a $500 course. Save this. Watch 3:24 before 0:26, because the pricing decides whether the security test even matters to

the most impressive thing this Grok Bot desk did last night was nothing, for 41 minutes straight no orders. no panic. no message to me it had 2 open positions, $2,810 behind them, and a stop it was allowed to move on its own it moved it 6 times. every move was closer, never further 01:12 → closed 0.4 sol for +$0.32, then went flat and stayed flat 02:14 → sent one telegram at -1.4% drawdown 03:30 → reviewed both, held, confidence 0.58 227 grok calls, $1.16, one night every parameter sits on the screen. steal it everyone shows you the winning night. this is the boring one that keeps the account alive

6 Grok agents ran a hedge fund research desk overnight with no human awake. 34 workers, 23:30 to 06:00, total spend $6.44. They pulled 7,000 documents, parsed 5,000, scored 1,744, flagged 202. By morning, 26 made the brief. One agent reads filings. One reads earnings. One watches insiders. A chief agent reviews every flag and either confirms it or bins it. The findings are the boring kind that move stocks. Auditor changed mid-quarter. Going concern language added. Cluster buy, 3 insiders. An analyst team does this by noon. It cost less than a coffee.

A company valued at $183,000,000,000 is run by two siblings, and the sister handles everything except the research. Daniela Amodei is president and co-founder. She left OpenAI with her brother and 5 others at the end of 2020. Her background is not machine learning. It is operations, hiring, and building the company around the science. Watch the commercialisation section. 80% of revenue comes from businesses, and 300,000 companies now sit on the models. The safety positioning is the sales pitch. Enterprises buy the audit trail as much as the capability. She was scaling a research lab before it had a product to sell. Her brother predicts the technology. She prices it.

A billionaire took the mayor's chair in New York for $1 a year and ran the city off the same terminal logic that built his company. Michael Bloomberg had been fired by Salomon Brothers in 1981 with a $10,000,000 payout and used it to build the terminal business. He spent about $74,000,000 of his own money on the first campaign and refused city salary. Watch how the deputy describes the management philosophy. Data first, no closed offices, and the same bullpen layout he used in the newsroom. He sat in an open floor at City Hall, no private office, surrounded by staff at desks. The company he built has been valued in the tens of billions and he owns the overwhelming majority of it. He was pushed out of a Wall Street firm. He came back and rented the city its own data.

Two men could not get a cab in Paris in the winter of 2008, and the company that came out of that night was later worth over $100,000,000,000. Travis Kalanick tells the story on stage while the business is still small enough to explain in one slide. It started as UberCab. Black cars only, one city, a luxury service for people who could pay a premium to skip the wait. Watch what he says about the first version. The idea was not a taxi app. It was a button that made a car appear, and everything else came from following that. Garrett Camp had already sold StumbleUpon to eBay, so the first money came from a founder who had cashed out. San Francisco sent a cease and desist in 2010, and they dropped the word Cab from the name and kept going. Kalanick was pushed out in 2017 after the culture he built on that stage caught up with him. They wanted a ride home. They rebuilt how cities move instead.

On 19 September 2014 the New York Stock Exchange rang its bell for a company that had been rejected by every venture firm in Silicon Valley. Alibaba raised about $25,000,000,000. The largest listing in the history of American markets. Watch who rings the bell. Not the founder. Eight customers, a farmer among them, standing where the executives normally stand. Jack Ma had raised $60,000 from friends in an apartment in Hangzhou fifteen years earlier. Goldman Sachs put in $5,000,000 in 1999 and sold most of it years before this morning for a fraction of what it became worth. SoftBank put in $20,000,000 in 2000. By the opening bell that stake was worth roughly $60,000,000,000. The shares priced at $68 and opened above $92. Yahoo held about 22% going in and cashed out billions on the day. Watch the floor when the price prints. Traders who had never used the product were buying a company that ran no inventory and owned no warehouses. It listed in New York because Hong Kong refused the governance structure the founders demanded. Every American firm that passed in 1999 watched the bell from a screen.

Elon Musk explained how he runs two companies at once, months before the one everyone doubted turned its first profit. January 2013. CEO of Tesla and SpaceX, chairman of SolarCity, with his own money inside all three. Watch how he handles the split-attention question. He is being asked to justify the structure right before it starts printing. Tesla reported that first profitable quarter in May, and the stock ran from roughly $35 to over $190 across the year. He had put nearly his entire PayPal payout into cars and rockets and run it down to almost nothing in 2008. Short sellers had spent 3 years betting the car company would fold. He was defending the arithmetic in January. By May nobody was asking.

In 2013 Elon Musk sat down for a long-form interview with a comedian and the interviewer had to make sure he was correctly naming the companies. Almost no one watches it. The chat show presented by Kevin Pollak in March 2013 had no network and no public relations handler intervening. The year before, Tesla had sent out the Model S and was still several months away from achieving its first profitable quarter. The first private company to achieve this was SpaceX, and it did so by docking with the space station in May 2012. Observe the way he responds to the comedian; the format doesn't allow him to use the scripted remarks that he'd normally employ in a business segment. In the days when podcasts were still something that people did for a friend, he appeared as a guest on one. The interviews became briefer and the number of audience members increased. This one is still too long.

In 1991, while he was running a company with thousands of employees, Bill Gates wrote and operated a program on stage. Very few people watched it. In its first year Visual Basic was just a beige box sitting on the table and one screen behind him. Before typing a single line he pulls the interface into position and then presses Run, after which it compiles in front of the room. See just how little he explains. There is no marketing lead with him and no handoff to an engineer. In the beginning he read the code that his people sent and remembered the licence plates in the car park in order to know who was in the building. Programming was put into the hands of people who had never before compiled anything. The founder continued to handle the compiler himself. It's not common any longer for anyone to do that.

Mark Zuckerberg at 21, filmed on a couch in a rented house, one year before saying no to $1,000,000,000 in cash. Almost no one watches the tape. Palo Alto, 2005. The site runs on college campuses only, and you need a university email address to get in. Watch how casually he explains the network next to him. Sean Parker enters the story through Napster and graffiti, not through a term sheet. No office in the frame. A couch, a wall, two guys, a handheld camera, 8 minutes 37 seconds. Yahoo arrived in 2006 with the cash on the table. The board wanted it taken. Meta cleared $164,000,000,000 in revenue last year. Everyone in that room could have cashed out. He kept the couch and the code.

Elon Musk told skeptics that throwing away a $54,000,000 rocket after one flight was a choice, not a law of physics. Almost no one watches the tape. 2011. Four years before anyone landed an orbital booster and flew it again. He builds the whole case out of one comparison. A plane is not thrown away after a single flight, and he asks what a ticket would cost if it were. Listen for the moment the questions turn. He is interrupted mid-argument and answers with arithmetic instead of vision. The propellant inside that rocket ran about $200,000. The rest burns up for no reason anyone could defend. Every established launch provider heard this in 2011 and kept dropping hardware into the ocean. He was pitching a joke back then. The joke landed upright in December 2015.

Ma Huateng shopped Tencent to buyers in 1999 and nobody wanted it. Jack Ma had already been turned down for 30 jobs, one of them at a KFC. The Chinese television archives from that stretch catch both men before anyone had a reason to listen. Studio chairs, no product the audience recognized, no reason for the segment to exist. Jack Ma's first time on camera was not even an interview. A crew found him at night, still sitting on his bicycle, in the middle of something he had not planned to be filmed doing. He had put 7,000 yuan of savings and a family loan into China Yellow Pages, selling web pages in a country that was barely online. Hangzhou Telecom set up a competitor and pushed him out inside 2 years. Ma Huateng started Tencent in Shenzhen in November 1998 with 4 partners. OICQ crossed 1,000,000 users in its first year and earned nothing, while the server bills kept coming. So he went looking for an exit. He approached Sohu. He approached Yahoo China. Both passed. IDG and a fund run by Li Ka-shing's son each took 20% for $1,100,000, which valued the company at about $11,000,000. Then AOL won a trademark case and OICQ became QQ. In May 2001 a South African media group bought 46.5% for $32,000,000. That stake later passed $175,000,000,000, the best venture bet ever placed. Jack Ma started Alibaba in April 1999 in his apartment with 18 people and 500,000 yuan, ran the biggest IPO in history in 2014, attacked China's financial regulators onstage in October 2020, and vanished from public view for 3 months. One could not sell his company. The other could not get hired.

In 1995 Larry Ellison told a room in Paris that the personal computer was finished. He was right about the machine and wrong by 15 years. The idea started as resentment. Windows 95 had just launched into a wall of hype, and Ellison went looking for a way around it. His first pitch was at a White House technology summit. Put a sub-$500 computer on every American schoolchild's desk. Clinton and the executives in the room let it pass. Then Paris. "A PC is a ridiculous device." No hard drive, no local software, nothing to install or break. The machine plugs into the wall for power and into the network for everything else. By 1996 he had Sun, IBM, Apple and Netscape signed on to a shared spec. Scott McNealy stood beside him at Comdex. Trade press called it the end of Wintel. It went nowhere. PC prices kept falling, networks were slow, and a $2,000 multimedia machine you actually owned beat a $500 terminal you did not. Ellison took a seat on Apple's board and pushed a Mac version. Jobs came back, rebuked it in public, and killed it. He tried again in 2000 with a $199 box, unveiled to students in Dallas, 1,100 donated to 23 schools. It sold under 50,000 units and shut down in June 2003. Google shipped the first Chromebook in 2011. No meaningful local storage, everything in the browser, and school districts bought them by the million. Oracle spent the next decade rebuilding itself into a renter of computing power for AI models. On September 10, 2025, Ellison briefly passed Elon Musk as the richest person alive. He lost every machine he built. He won the argument.

Jeff Bezos was worth $10,000,000,000 and still driving the same 1996 Honda Accord he bought before Amazon made a cent. The origin runs through a car too. July 1994. Bezos, 30, the youngest senior VP at a Wall Street firm earning well over $1,000,000 a year, tells his boss David Shaw he wants to quit and sell books online. Shaw walks him through Central Park and urges caution. Don't leave a sure thing for a maybe. Bezos leaves anyway. He flies to Texas, and while his wife MacKenzie drives them 2,000 miles to Seattle, he types Amazon's business plan in the passenger seat. Five years later a TV crew films him in that beat-up Accord, seatbelt on, grinning, as Amazon's market value crosses $10,000,000,000 in a single Friday. He could have bought any car on earth. He kept the one that couldn't tell he was rich. The boss said play it safe. He wrote the plan from the passenger seat instead.