Crane

A 17 year old in Mexico City runs 28 YouTube Shorts channels in the brainrot niche from his bedroom. His parents think he is wasting his afternoon watching cartoons. His third highest earning channel cleared $32,400 last month. He has not opened it manually in 11 weeks. The system takes 10 minutes per video and produces 4 a day per channel. The math compounds across 28 channels, which is the only reason a brainrot operation makes sense at all. He opens Claude and types one prompt. Give him a 45 second Shorts pitch with a concept absurd enough to beat yesterday's upload. A capybara reviewing 17th century Dutch tulip mania. A vending machine giving advice on heartbreak. A subway surfer narrating the fall of Rome. Claude returns 7 scenes with timing, dialogue, and visual descriptions. He pastes the breakdown into InVideo AI. The generator produces voice, captions, cuts, vertical export. From prompt to publishable Short in 5 minutes. He uploads. Walks away. The audience is teenagers in 4 continents who loop the same video 9 times back to back. The retention curve looks like a flatline. They never tap the skip button. CPM in this niche is around $9. He has watched the analytics graph long enough to know the more absurd the concept, the longer the retention. He stopped resisting that 6 months ago and embraced it. The top performing video on his fourth channel did 14.2 million views in a week. A dinosaur explaining inflation to a turtle while skateboarding inside an Apple Store. Made $12,800 by itself. Total monthly cost across the 28 channels. $66. InVideo subscription, Claude API, a scheduler that posts to the right channel at the right local timezone peak. No editor. No voice talent. No thumbnail designer. He hires nobody. He fires nobody. His mom thinks he is watching YouTube on his laptop for 5 hours a day. She is right. He is checking analytics across 28 dashboards, killing channels that fall under 100K views per video and spinning up replacements within an hour. The brainrot niche is unsentimental. He has buried 14 channels in 14 months and birthed 42. His dad asked him over breakfast why his AirPods Pro arrived from Amazon last week. He said he saved up. He has 312,000 pesos in a CitiBanamex account in his name and 14,400 USDT on Bitso his cousin in Monterrey helped him set up. His abuela thinks he has been quiet lately because of school stress. She makes him pozole on Sundays and tells him to take breaks. He says he will. He took a 6 minute break this morning to publish 4 videos. His older brother is studying for the medical school entrance exam and earns 2,800 pesos a week tutoring at a cram school in Polanco. The brother asked him last weekend if he was thinking about a career yet. He said he was figuring it out. He cleared more last month than his brother will earn in 6 years of residency. Across his 28 channels right now, 173 Shorts are queued for the next 24 hours. Capybara CEOs. Confused subway surfers. Talking traffic cones. Vending machines philosophizing about death. He has watched maybe 4 of them in full. He does not need to. The teenagers do

A 19 year old girl in Vienna sells Airbnb style websites for small hospitality operators, closes about thirty of them a month, and clears fifteen thousand euros doing it. She has never touched code. She has never opened Figma. She has never used Photoshop for anything except sending memes to her group chat. She opens Claude Sonnet 4.6 and types one sentence. Give me a fully responsive booking page site for a family run pension with a sticky nav, a hero with a find your stay call to action, a grid of listing cards with photo, rating, price, and a favorite heart, plus a footer with a Google Maps embed and a contact form. Claude sends the prompt back cleaned up and then ships the site in a single response. She previews it, tweaks two colors, swaps the hero photo for something on brand, and hands it over. Cost to her per delivery is under three euros in compute. Turnaround from first message to live URL is under ninety minutes. She charges five hundred euros per site. She does not do revisions after the second round. She does not host the site herself. The client owns the code, the domain, the hosting account, and everything downstream. She books the next call. Her father runs a fourteen room pension across the courtyard from Schönbrunn that has been quietly losing walk in business to the Airbnb listings on the same street for three years. He paid a Vienna web agency eighteen thousand euros in twenty twenty two for the site that still sits on his own domain. He got a quote last spring to redo it for twelve thousand more. He is on the phone with the agency's account manager. He is disputing an invoice line for what they called content strategy consulting. He is waiting on a signed statement of work for a two page update. His daughter builds the same class of site for guests of the same class of pension in ninety minutes for five hundred euros and has already delivered twenty two of them this month with a week still to go. Her father thinks she is helping a friend with her fashion blog. Save this and ship one before your street fills with one prompt booking pages.

@CodeByPoonam yeah the pace is picking up fast. staying curious is the only sane response

@Cranefomo AI is getting really good

We are cooked. Real vs AI with Gemini Omni is insanely impressive 👇

@DataChaz 62k stars in this timeframe usually means someone hit a real nerve. capcut adding watermarks and paywalling exports made this inevitable. moment the community stops tolerating rent seeking is the moment open source wins

CapCut CEO watching a dev recreate his video editor into a 100% free, watermark-free app, and there's nothing he can do about it

@tom_doerr open source 2d to 3d suite designed for cinema is a category nobody was serving. bookmarking the repo

Generates 3D from 2D videos using AI-powered depth mapping and custom stereo logic github.com/VisionDepth/Vi…

A 17 year old kid in Naples uploaded a stitched together Erling Haaland meme reel from his bedroom and went from four hundred followers to two point four million in a single afternoon. He did not shoot. He did not edit. He did not know the trending audio. He had twelve cutout photos of Haaland. He opened Claude, dropped all twelve into the chat, and typed one line. Turn these into a viral funny reel with trending audio. Claude sent back a scene by scene plan. Cutout one goes into an office elevator that just opened, the doors closing on him. Cutout three sits at a school desk in a scandalized pose. Cutout seven is glued to the front of a Vespa going the wrong way down Via Toledo. Each frame had a beat marker, a suggested overlay, and a caption line. The audio was a fifteen second slice of a track that had five million active reels on it that morning and had not yet peaked. The kid followed the plan exactly. He did not skip a beat. He did not swap the song for one he liked better. He rendered the sequence in a free web editor on the family laptop, uploaded to Instagram at four in the afternoon, and went to help his mother carry groceries. By dinner the reel was at eleven million views. By midnight it was at fifty five million. By the time his mother woke him up for school it was at a hundred and fifty five million views, fifteen million likes, and two point two million new followers. His father is a regional sales manager for a Serie A club merchandise distributor in Naples who has spent the last quarter chasing three mid tier Instagram creators for a modest campaign at four thousand euros a post. He is drafting the pitch deck. He is negotiating deliverables with a talent agent. He is waiting on legal to sign off on the licensing clause. His son turned twelve free cutouts and one prompt into more reach than the entire agency roster his father is trying to book, in the time it took the pasta water to boil. His father thinks he is playing FIFA in his room.

A 20 year old in Los Angeles built an AI character named Ada Wren, put her on Instagram with a bio that opens with the words "AI. Not a person," and cleared forty three thousand dollars off her last month. Ada does not exist. She does not eat. She has never rented an apartment. She styles other people's tiny apartments in AI-rendered before and after photos, tags the couch, tags the lamp, tags the rug, and the founder keeps the Amazon Associates commission. The character has one job. She reimagines four hundred square foot LA studios and West Hollywood one bedrooms as spaces someone would actually want to come home to. Every post has a pinned reply that says, "Ada is an AI-generated character. Her creator is a human. Every product is a real Amazon link with an affiliate tag. See disclosure below." Nobody unfollows because of the disclosure. Nobody unfollows because they came for the room, not the girl. Her stack is three tools. Claude Fable 5 writes her voiceovers and captions in a tone she has held for six months without drifting. A one prompt render pipeline outputs the before and after images in her signature warm afternoon light. A Codex Pro script pulls the Amazon product feed every morning, checks which items are in stock, and swaps any dead link before her scheduled reels go live. His father runs a boutique interior design firm in Silver Lake that has been billing clients hourly since nineteen ninety eight. He is driving to a client's house to hang a curtain rod. He is waiting on a two week fabric backorder from a mill in North Carolina. He is arguing with a general contractor about baseboard heights. His son is publishing eleven pieces of styled content a week from a folding chair on the balcony, with three hundred and eighty thousand followers who know they are following an AI, and forty three thousand dollars in affiliate deposits from a niche the FTC actually likes because the disclosure is already stapled to every frame. His father thinks he is doing dropshipping for a friend.

A 22 year old in Changsha told his parents he was working as a runner for his uncle's textile export brokerage in the Yuelu district while he figured out his career after graduation. He is not a runner. He runs a permitted AI livestream commerce operation from a small industrial floor in Changsha North that put 216,000 yuan through his account last month. He is one of the first small operators in Hunan to hold a Douyin AI live commerce permit. The certification took him 5 months to get through his cousin's registered trading company and required a signed compliance agreement, category restrictions, and a monthly audit of his AI host disclosures. Every stream he runs carries the mandatory AI主播 tag pinned to the top of the screen for the entire duration. Every viewer knows they are watching an AI host. The AI host does not pretend otherwise. The comments section is full of viewers who prefer it that way because the AI host actually answers their sizing questions instead of ignoring them like a burned out real streamer at hour 6. His floor holds 8 rigs. Each rig is a vertical touchscreen, a ring light, a 4K camera, a green screen surface, and a small dedicated inference PC that runs the AI avatar. He built the rigs himself from a Shenzhen electronics bulk order for 4,200 yuan each. 8 rigs total 33,600 yuan in hardware. That is the entire capital investment. Each rig runs a different AI persona optimised for a specific product category. Persona 1 sells women's activewear to 28 to 40 year old professionals in tier 2 cities. Persona 2 sells small kitchen appliances to newlyweds. Persona 3 sells stationery to college students. Persona 4 sells eyewear to remote workers. Persona 5 sells camping gear to weekend hikers. Persona 6 sells pet accessories to first time dog owners. Persona 7 sells low tier smartphones to migrant workers. Persona 8 sells traditional tea sets to Malaysian and Singaporean buyers in Simplified Chinese and Bahasa. Claude drafts every persona's opening pitch, product education arc, objection handling script, and closing sequence. The scripts are tuned to whatever the platform's algorithm rewarded in the previous 72 hours. The AI host reads the scripts naturally, adjusts its face and gestures, and responds to comments in real time. The kid does not appear on camera. He does not want to. He watches the 8 dashboards from a chair in the corner and approves anomalies that require human decisions. He signs off maybe 20 decisions across all 8 rigs per day. His revenue mix last month. 148,000 yuan from Douyin commissions across the 8 personas. 68,000 yuan from Taobao Live streams for 3 of the 8 personas whose categories are also permitted there. He does not stream to TikTok because he does not hold an international AI live commerce certification and will not risk it until the platform's disclosure requirements settle down. His mother thinks the runner job is stable and safe. She makes him hunan xiaochao rou on Sundays and asks if his uncle is going to make him a full salaried employee soon. He says the conversation is happening. He has 480,000 yuan in an ICBC account his older cousin in Zhuzhou helped him open under joint custody after his 22nd birthday. His father is a shift supervisor at a Xiangtan electrical equipment factory and asked him last weekend if the runner job included benefits. He said it was still probationary. His older sister works as a livestream host for a mid tier fashion brand in Guangzhou and clears 12,000 yuan a month when her streams hit numbers. She complained at Sunday dinner that she was burned out from 8 hour daily sessions and the platform kept demoting her when she cut back. He said that sounded hard. He runs 8 labeled AI hosts on the same platform without any of them getting tired. She has not asked what her brother does for his uncle. The 8 rigs are running their evening cycle right now while he eats spicy noodles from a stall downstairs. His parents think he is at the brokerage.

Three Apple boxes arrived in a university dorm room. Two Mac Studios and a Mac Mini, each one labeled with a sticky note. ADMIN. DEV. UI/UX. His roommate took one look and asked why anyone needed four thousand dollars worth of hardware to check if it was going to rain. That was not what the hardware was for. Aviation runs on atmospheric data that most people never hear about. METAR reports. TAF feeds. Sensor readings from thousands of ground stations refreshed every one to three hours, temperature logged to a single decimal point. The system exists because pilots cannot afford to be wrong. It is publicly accessible, completely free, and almost nobody outside a flight deck ever opens it. He opened it. Then he pointed a model at every active station on the planet and gave it one instruction: find the places where the atmosphere has already made up its mind and the official forecast has not caught up yet. The three machines each handle a different layer. One ingests the raw sensor streams. One runs the predictions. One strips out all the noise and produces a single clean output for wherever you happen to be. Data enters on one end. A forecast comes out the other. The whole thing keeps running whether he is in a lecture hall or asleep. At some point it stopped being a side project. Logistics companies, event organizers, production crews, agricultural operations, anyone whose margins collapse the moment conditions shift unexpectedly, started subscribing to a forecast that consistently arrived hours ahead of the national services. The dorm room setup now generates more revenue per month than most entry-level engineering salaries, built on data the government publishes for free and a power draw smaller than an espresso machine. Meteorological agencies spend enormous budgets on orbital infrastructure and compute clusters. A single student undercut them using the same source data they already distribute publicly and largely ignore at the granular level. His roommate eventually asked for a real explanation. The student rotated the second monitor. The roommate went quiet for a moment, then asked how much a desk in the room would cost him. The most reliable signal in the world has been sitting in plain sight the whole time. Most people just never thought to read it.

A 20 year old in Porto told his parents he was doing an unpaid internship at a Foz do Douro real estate agency to build experience for his commercial law degree. He has not visited the agency. He runs a foot traffic analytics service for 9 small Porto businesses that closed €7,200 in monthly subscriptions last month. He started the service because he wanted to open a small specialty coffee bar himself and could not find any data on how many people actually walked past the 3 candidate locations he was considering. He built the first version in a weekend. A Logitech C920 webcam pointed at the sidewalk from his upstairs bedroom window overlooking Rua Cedofeita. A laptop running YOLO11 out of the box with the standard COCO pretrained weights that already know 80 common object classes. ByteTrack assigns a persistent ID to every person and vehicle so the same body walking through the frame gets counted once instead of once per frame. The counts stream into a CSV file that a small Streamlit dashboard reads live. He ran it for 6 weeks on 3 different windows facing 3 different candidate streets. He got specific hourly foot traffic patterns he could not have paid a market research firm to produce for less than €4,000. Then his neighbour, who runs a small vintage clothing shop on Rua Miguel Bombarda, saw the dashboard on his laptop at a coffee shop and asked what it was. The neighbour was paying an intern €400 a month to sit at the front window with a clipboard counting customers on weekends. The kid quoted €800 a month for a fully installed camera plus live dashboard access. She signed the same afternoon. She told 3 other Bombarda gallery district owners at a merchants meeting the following week. They all signed within a month. His current 9 clients pay between €400 and €1,200 per month depending on how many camera feeds they need. A parking lot in Cedofeita pays €400 to know when their peak occupancy hits. A ginjinha bar in Ribeira pays €1,200 for 3 cameras covering their entry, terrace, and interior seating to correlate weather with revenue. A crossfit gym in Boavista pays €900 for entry counting plus a heat map of which equipment gets used. His hardware cost per client is one €47 webcam and one €160 mini PC that runs the inference. He installs each site himself on a weekend. Each site pays for itself in month 2. He keeps €190 in monthly compute costs across the entire operation for cloud dashboard hosting and Claude API calls that generate the weekly written reports each client receives. Every camera looks at the business owner's own property. He drafts and prints the data protection notices required by Portuguese law and installs them on the door frame during setup. Every counted person is aggregated into hourly totals immediately. No individual faces or plate numbers are stored anywhere in his system. His cousin who is finishing law at the University of Coimbra drafted the client contract for two bottles of moscatel a quarter. His parents think the internship is going well. His mother makes him francesinha on Saturdays and asks if his supervisor at the real estate agency is going to write him a recommendation for the law faculty. He says the conversation is happening. He has €38,400 in a Millennium BCP account his older cousin in Braga helped him open under joint custody after his 20th birthday. His father works as a bus driver for STCP and asked him last weekend if the internship was affecting his coursework. He said it was fine. His older sister works as a junior consultant at a Lisbon retail research firm that sells foot traffic reports to grocery chains for €14,000 per market study delivered in 8 weeks. She complained at Sunday dinner that her firm had lost a bid to a competitor she had never heard of. He said that sounded competitive. He had 9 subscriptions rolling from his bedroom. His workstation is counting Wednesday morning foot traffic on Rua das Flores right now while he does his commercial law reading. His parents think he is at the agency.

A 21 year old in Toronto told his parents he was finishing his computer engineering thesis at the University of Toronto library after class. He has not opened his thesis document since March. He runs an AI product scanning service for 14 independent convenience stores and small groceries across the Greater Toronto Area that closed $28,400 in installations and subscriptions last month. He does not sell to Loblaws. He does not sell to Sobeys. He sells to the family run halal grocery in Thorncliffe Park, the Filipino sari sari corner store in Mississauga, the Sri Lankan spice shop on Kennedy Road. Small independents that operate on 4 to 8 percent margins and cannot afford a $60,000 Toshiba point of sale rollout. The pitch is not about firing anyone. Every store he works with runs 1 to 3 staff who cannot be replaced. The pitch is about the 90 second peak. Weekday 5pm to 7pm. Saturday afternoon. The moments when 6 customers stack up at the register while the single cashier processes one order, and the last customer in line puts the basket down and leaves. That customer never comes back. His system runs a cheap Logitech USB camera pointed at a small marble tile installed in the counter. Customers place their items on the tile. A YOLO model he fine tuned on 4 months of North American convenience store products identifies every item in about 800 milliseconds and displays a running total on a 10 inch tablet. The customer taps to pay with tap enabled cards. The cashier handles the customer who actually needs a cashier. Store owners buy the system for $4,200 fully installed. He charges a $180 monthly subscription for continuous product database updates because North American convenience store SKUs change constantly. New Rockstar flavour. Twix repackaging. A new energy drink brand. Every SKU change gets added to the database within 48 hours. The database is the moat. His stack maintains a shared product catalog of 47,800 items across the 14 stores. Every store benefits from every other store's additions. A new store gets accurate scanning on day 1 because the library is already trained. His stack costs him $190 a month in Claude API plus a small AWS bill. He handles physical installation himself on weekends with a hand truck. His first paying store was a family owned Vietnamese grocery in Chinatown. The owner had lost his weekend evening rush 3 weeks running to a chain competitor two blocks away with newer registers. The kid installed the system on a Saturday morning. By the following Saturday the store cleared its highest daily sales in 2 years. The owner told 4 other store owners at a distributor meeting in Etobicoke. His parents think the thesis is progressing. His mother makes him bánh cuốn on Sundays and asks if his advisor is going to write a strong reference for his masters application. He says the conversation is happening. He has $43,200 in a Scotiabank account his uncle in Hamilton helped him open under joint custody after his 21st birthday. His father drives a route for a Peel Region school bus service and asked him last weekend if the thesis was almost done. He said he was on the final chapter. He has not written a chapter. His older sister works as a retail systems analyst at a Loblaws corporate office and earns $68,000 a year before tax. She complained at Sunday dinner that her team had a stalled project to test AI self checkout at Shoppers Drug Mart pilot locations for the last 18 months because of vendor procurement paperwork. He said that sounded frustrating. He had 14 stores running the same functional capability from his laptop. His sister has not asked him what he does after class. The installation truck rolls out to a Tamil grocery in Scarborough tomorrow morning at 8am. His parents think he is at the library.

A 17 year old in Lyon told his parents he was working extra hours at his uncle's bakery to save for a university exchange in the fall. He has not touched a baking sheet. He runs 12 YouTube Shorts channels covering the same content in 12 different languages that cleared 68,400 euros last month. Every channel posts personal finance for people under 25. The same 3 minute daily rundown of a specific money concept. Compound interest on a first savings account. Renter insurance basics. How ETFs work. What a payslip actually breaks down. The concept never changes across the channels. The language does. He picks one topic every morning. Claude writes the base script in English in 4 minutes. Then Claude translates and culturally adapts it for 11 more languages. French. Spanish. Portuguese Brazilian. German. Italian. Dutch. Polish. Turkish. Arabic Levantine. Indonesian. Hindi. The rent example changes from London to Rio to Istanbul depending on the language. The bank names change. The salary figures change to match local averages. ElevenLabs voices each of the 12 scripts in the target language. He uses a different synthetic voice for each channel because audiences bond with a face and a voice they trust. CapCut assembles each channel's short in a template branded to the language and the region. A Python script uploads all 12 at the local peak time. Warsaw at 8pm CET. Jakarta at 6pm WIB. Rio at 8pm BRT. The whole daily production run takes him 90 minutes. Most Shorts factories test 12 different niches to see which one works. He tests one niche across 12 languages to see which markets need it most. The Indonesian channel is the biggest. 480,000 subscribers. 4.2 million monthly views. The Turkish channel doubled twice this year. The Hindi channel is smaller but has the highest engagement rate because personal finance content in Hindi has almost no supply below the millennial age band. The German channel monetises the highest CPM. Monthly revenue by language for last month. Indonesian 14,200 euros. Turkish 11,600. German 9,800. Portuguese Brazilian 8,400. Spanish 7,200. Hindi 4,900. The other 6 languages 12,300 combined. His parents think the bakery job is going well. His mother makes him tarte aux pralines on Sundays and asks if his uncle is going to give him a raise for the summer. He says the conversation is happening. He has 96,400 euros in a BNP Paribas account his older cousin in Grenoble helped him open under joint custody after his 17th birthday. His father is a driver for a regional postal service and asked him last weekend if he was going to sit for the baccalauréat général next spring. He said he was preparing. He has not opened a textbook in 5 months. His older brother works as a junior financial analyst at a Lyon insurance firm and earns 2,400 euros a month before tax. He complained at Sunday dinner that his firm was spending 4,200 euros a month on a content agency to make explainer videos in 3 languages for their customer education portal. The kid said that sounded expensive. He was producing the same volume across 4 times as many languages for the cost of Claude, ElevenLabs, and electricity. His brother has not asked him what he does after school. The Warsaw channel is uploading right now while he eats his mother's leftover ratatouille. The topic is why 22 year old workers should max their PPK contributions before the tax year ends. His mother thinks he is at the bakery.

A 17 year old in Valencia told his parents he was tutoring 3 younger neighbours in English so he could help with the electricity bill. He has not tutored anyone. He runs a satellite roof analysis service for Spanish solar installers that closed €14,200 in commissions last month across the Valencia region. He scans rooftops using publicly available satellite imagery from Sentinel-2 and high resolution overhead photography licensed from a Barcelona geospatial data broker for 340 euros a month. He runs the images through a Claude Fable 5 workflow that estimates roof pitch, cardinal orientation, unshaded surface area, tile condition, and rooftop obstacles for every single family home in a target neighbourhood. The output is a scored list of roofs ranked by solar suitability. Each entry includes a probability the household would install if approached with the right pricing tier. He sells the lists to 6 small Valencia and Alicante solar installers. Each installer pays him a 340 euro flat fee per neighbourhood report plus a 220 euro commission on every install that closes from a lead he sourced. Last month the installers closed 47 systems from his lead sheets. He collected 47 times 220 plus 4 neighbourhood packages. Roughly 12,400 in commissions and 1,800 in package fees. The analysis is not solar advice. He sends every installer a signed statement that his work is a preliminary spatial assessment based on public satellite data. The final engineering, permitting, and installation decisions remain with the licensed installer. That language was drafted by his cousin who is finishing law at the University of Valencia in exchange for a case of Estrella Damm every trimester. The workflow runs on his laptop while he eats breakfast. He picks a neighbourhood on Monday. Claude Fable 5 processes 800 to 1,200 rooftops overnight. By Tuesday morning he has a report ready to send. Installers receive the leads on Tuesday afternoon. Sales calls start Wednesday. Signed contracts start landing on Friday. His first client was a small installer in Gandia who had been buying leads from a Madrid marketing agency for 380 euros each with a 3 percent close rate. The kid's leads closed at 18 percent because his analysis screened out roofs that were too shaded, wrongly oriented, or structurally weak from a satellite view before the installer ever picked up the phone. That installer told 2 friends at a trade dinner in Castellón. They both called him the same week. Then a fourth installer heard about him from a supplier and signed on. He has been turning down inquiries from Cordoba and Sevilla installers because his current pipeline is at capacity. His parents think the English tutoring is going well. His mother makes him fideuà on Sundays and asks if the students are improving their pronunciation. He says they are. He has 34,600 euros in a CaixaBank account his older cousin in Palma helped him open under joint custody after his 17th birthday. His father drives a truck for a Mercadona distribution route and asked him last weekend if he was going to apply to the Polytechnic University of Valencia next year. He said he was considering engineering. He is not. His older sister works as a junior consultant at a Barcelona sustainability firm and earns 2,200 euros a month. She complained at Sunday dinner that her firm had spent 3 months preparing a solar suitability report for a single Catalan municipality and still had 4 weeks of edits to go. He said that sounded slow. He had processed 6 municipalities the same month. The satellite images for tomorrow's Alicante neighbourhood are already queued on his laptop. His mother thinks he is at English tutoring.

An 18 year old in Kunming told his parents he was doing an unpaid data entry internship for a Chinese travel agency after class. There is no travel agency. He runs a viral TikTok comedy channel with 2.3 million followers built entirely around a single recurring AI generated character his parents have never met. The character is a middle aged office worker in a wrinkled white shirt and slightly crooked glasses. His name in the channel captions is Mr. Wu. Mr. Wu has starred in 187 short videos over the last 7 months. He keeps encountering impossible things at work. A dragon in the elevator. A meteor in the parking lot. Aliens in the copy room. A tiny black hole in his lunch bag. Mr. Wu does not exist. He was generated once and locked forever. Most creators trying to make AI shorts run into the same wall on video 2. The character morphs. His nose changes. His shirt collar looks different. The audience notices instantly and stops watching. Continuity kills you before you find an audience. The trick that unlocked the channel was a single canvas. He generated Mr. Wu's front, side, and three quarter reference sheets on one afternoon in his bedroom. Locked the seed. Saved the reference sheets as reusable character assets in Krea's canvas mode. Every new short starts by pulling those assets in as the anchor before any scene generation happens. The pipeline runs on a phone and a laptop. Claude writes each episode as a shot list. 6 to 9 shots per short. Opening frame with Mr. Wu doing something mundane. Impossible event enters. Reaction shot. Escalation. Beat. Punchline. Claude also writes the 4 line caption in the Kunming Mandarin dialect his audience recognises. Nano Banana generates each shot as a still image using the locked character reference. The generation is fast because the character does not need to be reinvented. Only the surroundings change. He picks the strongest still from a batch of 4 per shot. Seedance 2.0 turns each still into 3 seconds of motion with the physical logic he prompts. Mr. Wu ducks. Mr. Wu blinks. Mr. Wu opens the elevator door and finds the dragon. CapCut Pro assembles the cut. He pays $15 a month for the pro tier because it removes the watermark and unlocks the auto captions his audience needs. He publishes at 9pm Kunming time. His total monthly stack across Claude, Krea, Seedance, CapCut, and phone data is 610 yuan. About 84 dollars. Last month the channel cleared 49,200 yuan from TikTok's creator fund, brand sponsorships, and a Chinese noodle brand that paid him to have Mr. Wu discover the impossible under a lid of their instant beef stew. Roughly 6,800 dollars. His mother thinks the data entry internship is going well. She makes him mixian noodles when he visits home for the summer and asks if he is going to apply to Yunnan University's tourism programme. He says the tourism industry is interesting. He has 92,400 yuan in a Bank of China account his uncle in Dali helped him open under joint custody after his 18th birthday. His father runs a small mechanic shop and asked him last weekend if he wanted to learn the trade after graduation. He said he was leaning that way. He is not. His older sister works as a customer service supervisor at an e-commerce warehouse in Chengdu and earns 5,800 yuan a month. She sent him a Mr. Wu video on WeChat last Friday and told him he had to check out this hilarious guy from Yunnan. He said he had seen him before. She has no idea Mr. Wu is her brother's channel. The next Mr. Wu episode is rendering right now while he does his English coursework. Mr. Wu finds a small dinosaur under his office chair. Mr. Wu reacts with perfect Mr. Wu reaction. Mr. Wu goes to lunch. His mother thinks he is at data entry.

An 18 year old in Dallas told his parents he was helping a family friend with a real estate brokerage after school. He has not stepped foot in the brokerage. He runs a personalised direct mail service for pergola and patio contractors across 4 Texas metros that mailed 2,400 postcards last month and closed 47 install jobs at an average ticket of $9,200. His 12 contractor clients pay him $600 for every install he sources. Total November revenue $28,200. The system starts with a legal data feed. He pays $200 a month for clean deed transfer records. Every home sold in Dallas, Fort Worth, Houston, and Austin in the last 12 months lands in his database within 48 hours of close. Then Claude Fable 5 does the actual work. It vision reads the original MLS listing photos and skips the 64 percent of homes that already show a patio cover, a pergola, or a mature shade tree. It skips another 12 percent with no patio at all. The remaining 24 percent are candidates. For each candidate it pulls Google's satellite imagery. It reads the roof pitch, the fence line, and the patio orientation from the overhead view. It runs the hourly solar angle for the property's exact coordinates against the shade the existing structures throw. The output is a table. How many hours of direct sun that specific patio takes each summer day. How many days will hit 97 degrees or higher based on the 30 year climate normals for that address. Then it renders. A louvered pergola drops into the actual satellite photo of the backyard, angled to match the light in the original. He runs the final image through Nano Banana for a last polish. The postcard goes out Wednesday. Front is the rendered image with the exact address printed underneath. Back is the diagnosis. Your patio bakes 9 hours a day in summer. Last Sunday hit 102 by 4pm. Then a QR code that opens a heat report for their specific address on his site with a booking form that routes directly to the contractor covering that zip code. Response rate averages 3.4 percent. Of those responses, 42 percent book a consultation. Of those, 61 percent close. His contractor clients pay him only for closed installs. $600 per close. Printing and mailing cost him $1,680 last month. Software stack $340 across Fable 5 API, the data feed, Nano Banana, and a VPS. His parents think the brokerage internship is going well. His mother makes him chicken fried steak on Sundays and asks if he is going to major in business next fall. He says he is thinking about it. He has $47,400 in a Chase account his uncle in Plano helped him open under joint custody after his 18th birthday. His father runs a small HVAC company and asked him last weekend if he was going to come work with him after graduation. He said he was leaning that way. He is not. His older sister works as a listing agent at an Uptown brokerage and earns $52,000 a year including commissions. She showed him a listing at family dinner and asked if he could recommend a pergola contractor for the buyers. He said he knew a few. He sent her the name of his highest performer. The contractor closed the buyers within 6 days. His sister has no idea her brother was the reason her buyers called that contractor. The system is generating the next batch of 640 postcards for the Wednesday mailing right now while he does his calculus homework. His parents still think he is at the brokerage.

A 16 year old in Stockholm told his parents he had finally gotten serious about his studies. He has barely opened his school textbooks in 5 months. He has been building an Obsidian vault that reads sources, updates itself, and tells him when its own notes contradict each other. He calls it a living wiki. It earns him about €4,200 a month from professionals who pay him to install a version of it on their machine. The setup is unusual because nothing about it is interactive. He does not write notes. He drops a PDF, a podcast transcript, a research paper, or a long article into a single inbox folder. A Claude agent picks it up within 60 seconds. The agent reads the source, extracts every distinct claim with a confidence level, finds every existing note in the vault that touches the same topic, updates each connected note with the new information, and flags any contradiction with a yellow tag he can review at the end of the day. The result is a knowledge base that grows without him touching it. Every claim has a source. Every contradiction is visible. Every note is up to date because the agent rewrites it every time a new source affects it. He started building the vault in February for his own history class. He was tired of writing essays that contradicted his earlier essays without realising it. Within 6 weeks his teacher noticed his arguments had gotten sharper and his references were more precise. By April his average had moved from 71 percent to 91 percent. Then a friend's mother saw the vault on his laptop during a sleepover and asked him if he could build one for her. She runs a small consultancy that advises Swedish municipalities on energy policy and was drowning in white papers, EU directives, and conflicting briefings from junior analysts. She wanted a vault that would tell her when her own briefings disagreed with each other. He charged her €600 for the installation, the prompt set, and the agent configuration. He charged her €120 a month to keep the agent's prompt library current as Anthropic ships new model versions and new context window features. She told a friend who is a doctor in Uppsala. The doctor told a researcher at Karolinska. The researcher told a lawyer who works on intellectual property disputes. He now has 28 professional clients paying retainers between €80 and €200 a month depending on volume. His parents think he is spending evenings on advanced maths preparation. His mother brought him kanelbullar and tea last night while he watched the agent reorganise a 400 note section on Nordic climate policy in real time. She told him she was proud of his focus. His older brother is doing a Masters at KTH and complained at Sunday dinner that his thesis advisor keeps asking him to find research he wrote about 2 years ago that he cannot locate. He said he could probably help. He installed his brother's old thesis drafts and research notes into a fresh vault yesterday afternoon. The vault surfaced 14 forgotten references and flagged 3 self contradictions in his brother's literature review by morning. His brother thanked him. His brother does not know what he just used. The vault sits on his laptop right now, eating its way through a stack of papers a Karolinska postdoc sent him at lunch. By tomorrow she will have 9 fewer hours of reading to do and a sharper thesis introduction than she wrote alone. He does not consider this work. He considers it cleaning

A 14 year old in Manchester told his mum he had taken on a part time paper round before school. He has not delivered a single paper. He has been running a content system from his bedroom that cleared £21,400 last month across 5 platforms he never opens manually. The system takes about 90 minutes a day of his time. The rest happens while he is at school, asleep, or playing FIFA with his cousin. Every morning at 6am the system scans the last 24 hours of viral content in 3 niches he picked when he set it up. Personal finance for teens. Self improvement for boys his age. Football transfer rumours. It pulls the top 40 posts by saves and shares from each niche, not by likes, because likes are a lazy signal and saves are what makes the algorithm push a post for weeks. Then it generates 10 fresh post ideas ranked by predicted save rate, using a model fine tuned on his own past hits. He opens his phone at 7am, reads the 10 ideas while eating toast, taps the one he likes most. Sometimes he picks the second or third because the top one feels too much like something he already posted. Then 3 different agents take over. One agent writes the post in his voice using the rhythm and slang patterns it learned from 90 of his previous posts. Another agent checks it for grammar, fact errors, and anything that would get flagged by platform policy. A third agent formats and posts the same idea to Twitter, Threads, TikTok, Instagram, and LinkedIn within 4 minutes of each other, each version tuned to the platform's preferred length and structure. He never opens the dashboards. The system uploads, tags, and tracks. After each post goes live, the system tracks saves, shares, and comments. It feeds that data back into the model that scores next morning's 10 ideas. The hit rate has gone from 1 viral post per 12 to 1 per 4 over the last 3 months. 5 million combined views in his first 2 weeks. 100,000 saves across his 5 accounts in his first 2 months. Brand sponsorships started in week 6. The platforms started revenue sharing in week 8 because his retention numbers were comically high. His mum thinks the paper round is exhausting because she sees him eat 3 Weetabix every morning at 6am. He told her the paper round pays £45 a week. He has £21,400 from last month in a Monzo account she does not have visibility into because his uncle in Birmingham helped him set it up under joint custody after his birthday in February. His dad asked him if he wanted help organising a CV for a Saturday job at Tesco. He said he would think about it. He has not thought about it. His older brother is at Salford studying business and works 24 hours a week at a coffee shop near campus making £9.30 an hour. The brother complained at Sunday lunch about how the rota is unfair this term. He said that sounds rough. He pulled in 50 times what his brother will earn this year from a system that runs on his iPad. His system is publishing the next idea right now while he sits in maths class. It will probably do another 80,000 views by lunch. He has not opened the apps in 6 days

A 17 year old in Hanoi runs 4 YouTube channels that make nursery rhyme videos for toddlers. Combined revenue last week. $9,840. Combined production time across all 4 channels. 6 hours. The math that nobody who feels too cool to make baby videos has ever calculated is that a 2 year old does not scroll past anything. They tap replay. They tap it again. They watch the same 90 second clip 14 times before lunch. The algorithm interprets that as flawless retention and pushes the channel into every other toddler's feed in the same age bracket on the planet. The least picky audience on the internet is also the most profitable. CPM in toddler nursery rhyme content runs around $8.50. CPM in adult finance commentary is $4. Parents leave the TV running. The kids never skip. His pipeline is dumber than the insight. He types one prompt into Claude. Build him a scene by scene script for a 60 second nursery rhyme about whatever theme he picks that morning. Friendly broccoli teaches a dinosaur to share. Tiny dolphin learns to brush its teeth. Bear cub explains why hand washing matters before dinner. Claude returns 8 scenes with dialogue, repetition hooks, and a chorus that loops twice. He drops the chorus and the verses into Suno. Suno writes the melody and sings the rhyme in a cheerful synthetic child voice within 80 seconds. He picks one of 4 variants. He pastes the scene breakdown into Veo 3. Veo 3 renders each scene in a soft 3D animation style every toddler channel uses because every toddler responds to it. 90 seconds of animation in about 4 minutes. He stitches the audio over the video in a free editor. Uploads. Walks away. Total stack across all 4 channels last month. $27 in Claude API. $19 in Suno credits. $35 in Veo 3. The first channel he started in February. His mom thinks he is doing extra English tutoring online for a private student in Australia. She is right that he is in front of his laptop at strange hours. She is wrong about what is on the screen. His dad asked him last weekend why he had AirPods Pro that the rest of the family does not have. He said he saved up from his tutoring student. He has 312,000,000 dong in a Techcombank savings account and 4,200 USDT on Binance his older cousin set up for him in Saigon. His grandmother who lives downstairs watches one of his channels every afternoon thinking it is a Vietnamese kids show she has not been able to find on TV. She has been singing the broccoli dinosaur sharing song while she cooks. She does not know her grandson is the show. His friends think videos for toddlers are humiliating to make. They are not wrong about the social stigma. He stopped explaining it to them in March. The 4 channels run themselves through the night while he sleeps. The scheduler posts at 6am local time in his target markets. The toddlers in 11 countries wake up. The parents hand them the iPad. The same friendly broccoli teaches a dinosaur to share for the 9th time this week. His older brother is studying for the medical school entrance exam and gets stressed about how he will eventually pay off his loans. He has not opened a textbook in 4 months. He is paying his brother's tuition next semester. His brother does not know yet

@Bytosin_edits pozole on sundays is just love language for take a break mijo. globally undefeated)

@Cranefomo the abuela making pozole every Sunday while he's running 28 channels is the most Nigerian parent energy I've ever seen from Mexico 😭

A 17 year old in Mexico City runs 28 YouTube Shorts channels in the brainrot niche from his bedroom. His parents think he is wasting his afternoon watching cartoons. His third highest earning channel cleared $32,400 last month. He has not opened it manually in 11 weeks. The system takes 10 minutes per video and produces 4 a day per channel. The math compounds across 28 channels, which is the only reason a brainrot operation makes sense at all. He opens Claude and types one prompt. Give him a 45 second Shorts pitch with a concept absurd enough to beat yesterday's upload. A capybara reviewing 17th century Dutch tulip mania. A vending machine giving advice on heartbreak. A subway surfer narrating the fall of Rome. Claude returns 7 scenes with timing, dialogue, and visual descriptions. He pastes the breakdown into InVideo AI. The generator produces voice, captions, cuts, vertical export. From prompt to publishable Short in 5 minutes. He uploads. Walks away. The audience is teenagers in 4 continents who loop the same video 9 times back to back. The retention curve looks like a flatline. They never tap the skip button. CPM in this niche is around $9. He has watched the analytics graph long enough to know the more absurd the concept, the longer the retention. He stopped resisting that 6 months ago and embraced it. The top performing video on his fourth channel did 14.2 million views in a week. A dinosaur explaining inflation to a turtle while skateboarding inside an Apple Store. Made $12,800 by itself. Total monthly cost across the 28 channels. $66. InVideo subscription, Claude API, a scheduler that posts to the right channel at the right local timezone peak. No editor. No voice talent. No thumbnail designer. He hires nobody. He fires nobody. His mom thinks he is watching YouTube on his laptop for 5 hours a day. She is right. He is checking analytics across 28 dashboards, killing channels that fall under 100K views per video and spinning up replacements within an hour. The brainrot niche is unsentimental. He has buried 14 channels in 14 months and birthed 42. His dad asked him over breakfast why his AirPods Pro arrived from Amazon last week. He said he saved up. He has 312,000 pesos in a CitiBanamex account in his name and 14,400 USDT on Bitso his cousin in Monterrey helped him set up. His abuela thinks he has been quiet lately because of school stress. She makes him pozole on Sundays and tells him to take breaks. He says he will. He took a 6 minute break this morning to publish 4 videos. His older brother is studying for the medical school entrance exam and earns 2,800 pesos a week tutoring at a cram school in Polanco. The brother asked him last weekend if he was thinking about a career yet. He said he was figuring it out. He cleared more last month than his brother will earn in 6 years of residency. Across his 28 channels right now, 173 Shorts are queued for the next 24 hours. Capybara CEOs. Confused subway surfers. Talking traffic cones. Vending machines philosophizing about death. He has watched maybe 4 of them in full. He does not need to. The teenagers do