Olivia
Online
Antonina(๐,๐งก)๐ธ โก๏ธ๐๐ฆ๐๐ฎ๐ขBAD
@ToniaCryptoKiss
โ๏ธ๐ฆ ๐ @fusyfox ambassador
Airdrop
1
Joined November 2020
0 Following
123 Followers
401 Posts
@hwchase17 meet the team where they are is the only ai distribution strategy that actually works. everything else is a demo
A 15 year old in Shenzhen carries a 3D printed cube the size of a sugar cube on a lanyard around his neck. The cube contains a small AI agent that he built over a school break to stop himself from opening his phone every 4 minutes.
He has not opened his phone for personal tasks in 11 weeks. The cube does it all.
The hardware cost him 184 yuan in parts from the Huaqiangbei electronics market 3 stops from his school. An ESP32 S3 microcontroller. A 1.3 inch round LCD screen. A microphone capsule. A small speaker. A 380mAh LiPo battery good for 9 hours. A button on top. A USB C port on the side. The case he printed himself on the family's Bambu Lab printer over a Saturday morning. Total weight 38 grams.
The cube does not run AI. It does not need to. It connects through his phone's hotspot tether to a Claude API endpoint that handles the actual thinking. The cube just listens, displays, and speaks.
He taps the button. The screen wakes up. He talks to it. The cube transcribes what he said using a local Whisper model that fits inside the ESP32's memory, sends only the text to Claude, gets a structured response back, displays the answer on the screen, and reads it aloud through the speaker if he taps twice.
The agent remembers everything he tells it. It manages his school schedule, drafts replies to his mother's WeChat messages in the polite formal Mandarin she expects, controls the Xiaomi smart switches in his bedroom, tracks his homework assignments by deadline, and quietly orders his favourite milk tea to be ready when he walks past the shop on the way home.
His mother thinks the cube is a fashion accessory. She told her sister it is what the kids wear these days. He has not corrected her.
His father asked him last week what the little box does. He said it is a school project for an electronics elective. His father said he should be proud. The school does not offer that elective.
His older sister is at Tsinghua studying computer science and has 4 AI subscriptions costing her 480 yuan a month combined. She complained at dim sum last Sunday that her Claude usage limits keep getting hit before she finishes her assignments. He said he had no idea what she meant. He pays 18 yuan a month in API costs to the same company through his cube and never hits a limit because the cube only sends what is needed.
His teacher saw the cube at his throat during physics class and asked him to take it off. He explained calmly that it was a hearing aid. The teacher apologised. The cube was at that moment summarising the lesson into structured notes that synced to his Notion when he got home.
His friends at lunch ask him what the cube is. He says it is a portable kitchen timer. They believe him because he is the kind of kid who would carry one.
Two boys in his year offered him 800 yuan each to make them one. He took the orders and ordered the parts the same evening. He has 6 weekend builds queued for next month.
The cube is on his neck right now. It is listening for the word that wakes it up. He has gone almost 5 hours since he last looked at his phone.
His phone is in his locker. It is on silent.
The cube does the rest.
A 17 year old in Frankfurt bought 6 decommissioned HP ProLiant DL380 servers from a Hessen datacenter liquidator for โฌ390 each and stacked them in his parents' garage between the lawnmower and a freezer his mother forgot was on.
He runs a small local AI compute service out of that garage that cleared โฌ11,200 last month.
The 6 servers each have dual Xeon Gold 6248 processors and 256 gigabytes of DDR4 ECC memory. He paid โฌ2,340 for the lot. The same chassis cost โฌ13,000 each when they were new in 2018 and every Fortune 500 had quietly written them off by 2024. He bought 2 Tesla P40 cards for each server at โฌ240 a piece off a German auction site. 48 gigabytes of VRAM per server. Enough to load Llama 3.3 70B at 4 bit quantization with room to spare.
Total build cost across all 6 servers. โฌ5,220. Same compute output as a stack of Mac Mini M4 Pros that would cost โฌ12,000 and offer a quarter of the memory.
The catch is the noise. Each DL380 pulls 280 watts at idle and screams at 65 decibels under load. His parents' garage sits 30 metres from the kitchen. His mother thought a neighbour was using a chainsaw the first weekend. He bought industrial sound dampening foam from a recording studio supplier in Mainz and lined the garage walls in 4 hours. The noise dropped to 38 decibels at the kitchen.
Each server runs a hosted endpoint for one specific use case. Server 1 is a 70B legal translation model fine tuned on EU regulatory documents. Server 2 is a coding model. Server 3 is a long context document summariser. Server 4 is image captioning. Server 5 is voice transcription. Server 6 sits as a load balancer and writes hourly logs.
His customers are 14 small businesses that cannot afford OpenAI enterprise contracts but want AI in their workflows. A Frankfurt accounting firm uses the translation endpoint. A Munich law firm uses the document summariser. A Berlin coding bootcamp uses the coding model for their students. Each customer pays โฌ600 a month for unlimited inference on the endpoint they need.
His parents think the garage hum is the freezer. His mother asked him last week why the electricity bill jumped โฌ112. He said his cousin had been charging his bike all month. She believed him because his cousin had a new bike.
His dad thinks he is doing well in his electrical engineering preparatory class. He has not opened a textbook in 4 months. He has โฌ31,400 in a Sparkasse account he opened the day he turned 17.
His older brother works as a junior consultant at a Frankfurt management consulting firm and earns โฌ58,000 a year before tax. The brother complained at family dinner that his firm just signed a โฌ4 million annual contract for an AI tool that does basically what his brother is offering for โฌ600 a month per client. He said it sounded like enterprise pricing. He pulled in 6 times his brother's monthly salary last month from servers that hyperscalers threw away.
The 6 DL380s run all night next to the freezer that has been on since 2019. The Tesla P40s are 8 years old. The chassis are 8 years old. The customers do not care.
His garage is the loudest room in the neighborhood and the most profitable
Who to follow
Blaq๐ค ๐ช๐ป
@ElenduCharles2
Hello I'm Kitoxchange The most reliable vendor for your crypto currency and other services
Serg(๐,๐งก)๐
@sergeybil777
X
Cosem ๐๏ธ@smcdao
@Cosem78
Exploring Web3 | Visual Artist | Pi | WKC| SMCDAO~Believer| GTAN|| KEDOLIK | ASSETCHAIN | XEND FINANCE -RWA| SHALOM | TiwiCat
A 19 year old in Belgrade told his dad he was studying for the mechanical engineering entrance exam. His dad is a mechanic at a state owned bus depot and has been pushing him toward a real trade since he was 14. He has not opened an engineering textbook since November. He has been selling and operating Claude powered automation boxes for independent auto repair shops across 4 Serbian cities.
12 shops on retainer. โฌ2,200 a month each. โฌ26,400 a month gross. He delivered the 12th box last week.
The product is a fanless aluminum mini PC the size of a hardcover book that sits on a shelf in the back office of an auto shop. It runs a Claude agent stack he built around an open weights model and 3 narrow workflows the shop owners actually care about.
Customer intake form replies. A driver fills out a contact form on the shop's website. 60 seconds later they get a personalized SMS with a quote estimate based on the make, model, and described issue. The reply mentions the shop owner's first name and a slot that is actually free this week. Half the time the customer books before the shop owner sees the original lead.
Parts reordering. The agent watches inventory levels through the shop's existing point of sale software, predicts the 7 day forecast against last year's same week orders, and emails 3 wholesale suppliers a structured request for quotes by Tuesday morning. The shop owner picks the best quote with one tap on his phone. The agent confirms and tracks delivery.
Invoice follow up. The agent runs a 3 step sequence on overdue payments. Friendly nudge after day 7. Direct request after day 14. Phone call from the shop owner with a context briefing the agent prepared after day 21. Recovery rate went from 41 percent to 79 percent at the pilot shop.
The shops were paying 2 to 3 people a combined โฌ18,000 to โฌ24,000 a month to do these tasks badly. They pay him โฌ2,200 a month and the box does them well.
His first shop was his uncle's. The uncle had been losing customers to a chain that had a slick online booking system. The kid installed the box on a Saturday in March. By July his uncle's monthly take was up 34 percent. The uncle told 4 other shop owners over rakija that month. All 4 bought.
The hardware costs him โฌ420 per box. He charges โฌ1,800 setup plus โฌ2,200 monthly. He breaks even on hardware in week 1 and prints the rest. Power per box is around โฌ4 a month. No staff. No office. He drives the boxes around in his mom's Yugo.
His dad still thinks he is studying. His mom asks every Sunday whether he is eating enough at the engineering library. He says yes. She makes ฤevapi and packs leftovers for him to take to his fake study sessions. He eats them in his bedroom while testing the next shop's agent configuration.
His older brother is a junior mechanic at the same state bus depot as their dad. Earns 95,000 dinars a month before tax. The brother complained last Sunday that his depot manager had refused his request for a tablet to use for repair documentation because the depot had no budget. He said it sounded frustrating. The 12 boxes in 4 Serbian cities run on equipment he paid for with last month's revenue.
He has 7 new shops booked for installation in May.
His dad bought him a new wrench set last weekend as a study break gift.
The wrench set is still in the box.
@samlambert 3x throughput once working set exceeds ram is the part most aurora users dont budget for until prod is on fire. the cliff is real and nobody documents it. testimonials like this hit harder than any benchmark deck
@leerob congrats. anyone who shipped your level of dx work is going to make composer noticeably better fast
An 18 year old in Sofia told his parents he was taking a gap year to study for the German university entrance exam. He has not opened a German textbook in 9 months. He has been running an air gapped AI inference service for regulated Bulgarian law firms, dental practices, and accountancies out of his kitchen.
The infrastructure is 4 Mac Mini M4 Pro boxes stacked between the coffee machine and a bowl of clementines. Each has 64 gigabytes of unified memory. Linked together through exo, an open source clustering project that turns the 4 machines into one 256 gigabyte pool. He paid โฌ9,400 for the entire setup using his savings from a summer job at a Bansko ski rental.
The cluster runs Qwen 2.5 235B at 4 bit quantization. About 14 tokens per second. Fast enough for the work his clients send. His apartment electricity bill went up โฌ19 a month.
His pitch is the one thing the cloud cannot legally offer his clients. The data never leaves equipment he physically controls. When the Sofia bar association asks his client where their confidential filings get processed, the answer is in a closet in Sofia, on hardware that is not connected to any third party server, owned by an individual contractor under a signed processing agreement. Audit trail clean. GDPR clean. Attorney client privilege intact.
His first client was a law firm that handles intellectual property disputes. The senior partner had been told by 4 cloud AI vendors that he could not use their service because his case files involved NDAs. The kid set up a private endpoint, signed the data processing agreement the partner's lawyer drafted, and charged โฌ3,200 a month for unlimited inference.
The firm now runs every contract review, every discovery search, and every translation between Bulgarian and English through his endpoint.
The partner introduced him to a 4 dentist practice in Plovdiv. Same agreement. โฌ3,200 a month. The dentist uses it for clinical note summarisation under Bulgarian medical confidentiality law. Then her accountant brother in Varna. Then a small notary in central Sofia. Then a tax advisor near the airport.
By month 6 he has 6 active clients. All on โฌ3,200 monthly retainers. โฌ19,200 a month gross. Power and internet under โฌ70 combined.
His mom thinks he is at the library every day from 9 to 5 preparing for the TestDaF. She makes him banitsa and ayran every morning and asks how the verb tables are coming along. He says they are tough but he is making progress.
His dad asked him last weekend if he wanted help going through Munich apartment listings because the family had been saving for his stay there. He said he was still waiting on his application status. He has no application. He has 86,000 leva in a Postbank account he opened the day he turned 18.
His uncle is the senior partner at one of the law firms paying him. The uncle does not know the kid is the vendor. The kid set up the contract through an LLC his cousin in Plovdiv let him register under her name for a 4 percent revenue share. The uncle thinks the vendor is a small German firm that does private AI hosting for European regulated industries. The kid emails his uncle in formal German once a month from the LLC's address.
The Mac Minis hum next to the toaster. The lawyers across town review contracts using a model that lives 12 minutes from his uncle's office.
The uncle has never noticed his nephew's hosting service was 4 silver boxes next to a fruit bowl.
@simplifyinAI framing it as a research tool not financial advice is the right move. the multi agent debate structure is useful well beyond trading
@TheRundownAI claude as a slack coworker is the one that quietly changes how teams operate. async agents inside the same channel as the humans removes the context switch that ate everyones day. the screenshot of cadence picker in a real thread sells it better than any landing page
@threepointone brutal prioritization is the only skill that scales when everything else does. agents make the cost of doing the wrong thing way higher
A 17 year old in Warsaw told his parents he was building a portfolio website to apply to design school. He has been running an AI Reels studio out of his bedroom for 11 weeks serving 28 small businesses across Poland who think his agency has a team of 6.
The agency is him and a Lenovo Legion laptop he won at a school programming competition.
The pitch is simple. He charges a Polish business 1,200 zloty for 10 short Reels a month. The business sends him 1 photo. He sends back 10 finished vertical 60 second Reels by the end of the same day.
The first photo is the only photo he ever needs. Krea takes the raw shot and generates 12 styled variations of it under different lighting, angles, weather, and time of day. The bakery's cinnamon roll on a marble counter at golden hour. The same cinnamon roll on a rustic wooden board at blue hour. The same one in a flat lay with espresso cups and a 1940s Vogue.
Runway turns each variation into a 60 second clip. Slow push in. Slow pull out. Subtle steam rising. Hand entering frame and taking a bite.
Claude writes 10 different captions in Polish for the same business, each tuned to a different hook style. One curious. One nostalgic. One contrarian. One asking a question. He never writes a caption himself.
His first client was a real estate agent in Mokotow who had 340 followers on Instagram and a 1 bedroom apartment that had sat on the market for 3 months. The agent sent one phone photo of the kitchen. The kid sent back 10 Reels by 11pm. The agent posted 1 a day for 10 days. The apartment got 47 viewing requests in the first week and sold above asking the following Tuesday.
The agent paid him 1,200 zloty and asked who else he worked with. He said he was building out his client list this quarter. The agent introduced him to her broker. The broker introduced him to 4 other agencies. Within 4 weeks he had 14 real estate clients.
Then the restaurants noticed. A pierogi place near the Old Town hired him because the Mokotow agent had posted about his work. Then a barber. Then a flower shop. Then a small Warsaw fashion label trying to launch on Instagram before opening their first store.
He charges 1,200 zloty per client per month. 28 clients. 33,600 zloty a month. Roughly $8,400. He spends about 280 zloty on tool subscriptions across Krea, Runway, and Claude API. The rest is profit.
His mom thinks he is freelancing for a friend's design studio after school. She makes him pierogi z miฤsem on Wednesdays and asks if his portfolio is coming together. He says it is looking good.
His dad asked him last weekend why his bank app shows incoming transfers from a real estate agency. He said he was helping them with their website. His dad nodded. He has not opened a website builder in his life.
His older sister works as a junior designer at an actual design agency. Salary 4,800 zloty a month before tax. She complained at dinner last week that her clients never approve anything on the first try. He said that sounds hard. He has not had a client send a revision request in 11 weeks because they pay for finished Reels, not concepts.
The Lenovo runs all night. Krea and Runway take turns. Claude writes captions while the renders queue up. The kid sleeps next to the fan.
By morning 10 more Reels for 3 different clients are waiting to be uploaded.
He has not held a camera in his hand all year.
@MatthewBerman repo is fresh but the direction is right. practical repeatable agent workflows is the next layer above prompt libraries. curious how discover handles privacy when reading chat threads with sensitive context
An 18 year old in Lisbon told his mother he was going to a coding bootcamp every weekday morning. He has not attended a single session. He has been at his desk in their tiny apartment running the largest open weight model nobody else in the city can fit on a single machine.
When Zhipu released GLM 5.2 on June 13, he had been saving for the maxed out Mac Studio for 7 months. 512 gigabytes of unified memory. โฌ9,200 with the discount his cousin who works at the local FNAC pulled for him. He installed it on a Sunday. By Tuesday he was running the lightest 2 bit quantization at around 220 gigabytes, the only single machine setup in his postcode that actually fits a 744 billion parameter mixture of experts model.
He skipped the Ollama route. There was no official tag yet. He pulled the Unsloth GGUF through llama cpp like the rest of the early adopters and tuned the layer offloads for 16 hours straight.
Then he stopped sleeping at night.
Indie developers submit their repository to him before bed. They pay โฌ40 per repo. By morning GLM 5.2 has read every file, mapped the architecture, identified the 3 worst pieces of technical debt, refactored them, run the existing test suite to verify nothing broke, and opened a pull request against the main branch.
The 1 million token context window matters more than the benchmark numbers. The model holds the entire codebase in attention at once. He stopped feeding files and started feeding whole repositories. The output reads like a thoughtful senior engineer left a note at 4am.
First customer was a friend's older brother who runs a small Shopify plugin business. Stripe integration was a mess. GLM rewrote it overnight. The pull request added 412 lines, removed 1,847, passed every test, and earned him โฌ40.
By week 6 he had 73 active customers. Most submit a repo once a week. Some submit nightly. He charges premium tiers for repos above 200,000 tokens because the inference time on a 2 bit quant on a single machine is real. โฌ40 base. โฌ110 large repo. โฌ260 enterprise on monorepos.
His revenue last month was โฌ11,200. Power cost him โฌ43. The Mac Studio sips current quietly even under sustained load.
His mom keeps asking what time his bootcamp finishes today. He says 5pm. She makes pataniscas for dinner and asks how the cohort is going. He says he is enjoying the algorithms module.
His older sister is a backend engineer at a Lisbon fintech. She asked him last weekend if she could pay him to refactor a service her team had been avoiding for 18 months. He said he could not, his bootcamp was busy. He refactored it that night and emailed her the diff Monday morning from an account she did not recognise. She paid the anonymous developer โฌ600 and forwarded the diff to her tech lead. She still does not know it was her brother.
Zhipu open sourced a model under MIT license that even the buyers of high end NVIDIA hardware cannot run as cheaply as he runs it on one quiet Mac Studio. The only thing it needed was someone willing to point it at the right work.
His bootcamp is supposed to graduate him in 3 months.
His sister got promoted last week
@pelaseyed splitting workloads across two open models is the right call. one for text one for vision and you pay a tenth of what you would on a single closed api. the moat people thought was the model was just the bundling all along
A 17 year old in Athens told his parents he was getting better at his summer cafe job. He had been fired from the cafe job 5 weeks earlier. He spent those 5 weeks building a software product that the same cafe was now paying him $500 a month to use.
He had noticed something while bussing tables. Every owner he passed talked about Google reviews. Wanted more. Did not have time to ask for them. Did not know how to chase the ones they had honestly earned. Customers ate, paid, left, never reviewed.
He opened Emergent on a Tuesday evening and typed one prompt. A simple SaaS for small restaurants to collect Google reviews automatically through text messages 2 hours after the bill. Owner dashboard. Per location login. Templated message. Direct link to their Google business listing. Analytics on how many sent, opened, converted.
Emergent built the whole thing in 47 minutes. Auth, dashboard, message queue, billing stub, the lot. He spent the next 2 days rewriting the visual layer to feel like Mercury. Black background. Dark grey cards. Tight serif numbers on the main metrics. Mint accent on the action buttons. Anyone who walks past an Apple Store recognises the aesthetic and trusts it.
Then he wired Zapier into the Square installation at the cafe that had fired him. When a payment closes, a webhook fires, his app waits 2 hours, then sends the customer a one tap link from a local Greek number he rented for 4 euros a month.
He showed the cafe owner the analytics page. 38 reviews collected in 11 days. Cafe rating moved from 4.1 to 4.6. The owner paid him $500 the same afternoon and asked if his cousin's taverna in Glyfada could get one too.
That was client 2.
Then the cousin's friend who runs a souvlaki place near the metro became client 3. The dentist next door asked if he could do it for medical practices. Same code. Different copy in the templated text. $500.
By week 9 he had 10 paying owners. $5,000 every month. The app does the work. He has not written a line of code since the first build.
The analytics page sells itself at every meeting. He shares his screen. Owner sees reviews climb in real time. Owner signs the same afternoon.
His parents still think he is at the cafe earning 5 euros an hour. His mom packs him a fasolada thermos every morning. He eats it on a bench at the park near home before he opens his laptop on the same bench because he never goes anywhere.
His dad asked him last Sunday if the cafe gave him a raise. He said yes. His dad said good, save it for university.
He has โฌ18,300 in his bank account from 9 weeks. He has not told anyone.
The cafe that fired him is now his number one referral source. The owner sends him every shop owner he knows. Each one becomes another $500 line on his sheet.
His phone buzzes with review notifications every few minutes during dinner service. He silenced them. He just watches the dashboard while his mother passes him salt.
The numbers go up while he eats.
@alifcoder the funnel is always the same. free guide today small discord next week 997 dollar cohort the month after. you can see the next three posts before they get written
@Scobleizer the optimism take only holds if washington uses the pause to actually fix something. otherwise its just lost months turned into a quote tweet. would love to see a concrete plan come out of this not just the usual hard road framing
A 19 year old in Bangalore told his college dean he was taking a semester off to deal with a family thing. The family thing was that his gaming PC had started replacing $3,000 a month of company SaaS contracts.
He found a 14 billion parameter open weights model on GitHub last winter. 23,000 stars. Trained by a small Chinese lab nobody had heard of 4 months earlier. It runs on his RTX 3060 with 12 gigabytes of VRAM. 4 bit quantization. About 18 tokens per second. Quiet enough that his roommate did not notice he had stopped using the cloud.
His setup is a Lenovo gaming tower he bought refurbished for $640. The model sits in a Docker container. He exposes it through a tiny Python wrapper that mimics the OpenAI chat completions endpoint. From the outside it looks like he is calling a paid API. From inside his bedroom it is a card he can touch with his hand.
That was the entire technical breakthrough.
Then he sat down and asked what a business would pay $200 a month for that he could solve in 50 lines of code.
He built a code review service for small Indian dev shops. Push a PR. His webhook fires. The model reads the diff against the repo history, flags 4 categories of issue, and posts comments back to GitHub in 14 seconds. Same shape of output that GitHub Copilot Code Review ships. Same shape of output that a senior engineer ships at standup. He charges $290 a month. Unlimited reviews.
First client paid in March. He sent the invoice from his phone during lunch at the canteen. A 9 person agency in Pune. They had been quoted $1,400 a month for the same workflow from a US vendor. They paid him $290 and did not ask why it was cheap.
Then he built a second product. A meeting transcript cleaner for sales teams. Same engine. Different prompt. $190 a month.
Then a third. Customer support tier 1 auto reply. $410 a month per inbox.
Three products. One $640 box. The model never touches the cloud.
By month 6 he was running 41 paying customers. Combined MRR $11,200. His electricity bill went from 2,100 rupees to 3,400. That was his only new variable cost. No API tab. No usage warnings. No 4am bills from a runaway loop.
His mom thinks he is freelance designing logos again like in high school. He lets her think it.
His dad asked how the family thing was going. He said better. His dad said good, your studies will still be there.
The dean kept his seat open for next semester. His friend who picked up his old textbooks asked in January if he wanted them back. He said keep them.
A senior at his old college works at a startup that pays $480 a month per developer seat for the same code review tool he sells for $290 unlimited per team. The senior thinks he is heading for a strong career path. He has 41 customers under his bed.
The 14 billion parameter model still sits on his GPU. Still warm. Still writing reviews while he sleeps.
His gaming PC has not played a game in 5 months.
@DataChaz parsing has been the silent killer of every rag pipeline ive built. you do everything else right and lose 40 percent of the page at the first step. shifting to screenshots indexed as vision is the kind of fix that should have happened a year ago
A 21 year old in suburban Madrid told his parents he was looking for a job. He has not applied to one in 8 months.
He clears $35,000 a month from a desk in his childhood bedroom. One laptop. One Claude window. Same Lego set on the shelf he had when he was 12.
He sells AI automations to small businesses. Real estate offices. Dental clinics. Contractors. Anyone who still runs follow ups on sticky notes and spreadsheets. He builds the system that kills the busywork. Charges $2,500 upfront and $300 a month forever. Claude writes every workflow.
The setup never changes.
He points Claude at a niche. Real estate in Valencia. Dental clinics in Bilbao. Small roofing companies in Andalusia. Claude pulls 200 businesses still operating on chaos. By morning he has a fresh lead list with their email, their owner's name, and a one line summary of which part of their operation is leaking time.
He sends a single email. The pitch writes itself off Claude's summary.
3 percent reply. 1 percent close. Out of 200 leads he books 2 calls and lands 2 clients a week.
The build is the same skeleton every time. Lead capture form. Follow up sequence. Booking reminder. Invoicing chain. No show recovery. Claude reads the client's existing process from one Loom video, rewires it into his template, swaps the branding, exports. 20 minutes per build instead of a week.
His first client paid him $2,500 in February. He used the money to buy a second monitor. His parents thought it was for video games.
By month 3 he had 9 clients. By month 5, 15. By month 7 he had 20 active builds in flight and 70 small clients paying him $300 a month whether he opens his laptop or not.
That $300 fee is the engine. 70 clients on retainer is $21,000 a month that lands on the first regardless. The $14,000 on top comes from new builds.
His dad keeps printing job listings and leaving them on his desk. Software jobs in Madrid. Junior backend roles. โฌ1,800 a month before tax. He thanks his dad and adds them to the recycle pile.
His uncle works in HR at a logistics firm. Keeps offering to forward his CV. He says he is still figuring out what he wants. His uncle calls him picky.
His mom asks every Sunday if there are any interviews this week. He says some leads are looking promising. He is not lying. The lead list Claude built last night has 47 hot ones.
What keeps everyone else stuck at 21 is waiting to feel ready. He shipped client 1 before he felt ready. Claude covered the gaps.
His friends from college are sending rรฉsumรฉs. He is deciding which clients to fire.
The job he was looking for pays a quarter of this.
His parents still hang his graduation photo in the hallway and tell relatives he is exploring his options.
He is. He is exploring whether to raise the retainer to $400.
Last Seen Users on Sotwe
new lowerfantasi
Seen from Indonesia
ADRIAN RAMOS ๐ซ๐ณ๐ฎ๐ณ๐ฎ
Seen from Mexico
Abang Askar ๐ฒ๐พ
Seen from Malaysia
moms and milfs
Seen from Turkey
POMPA ZAMANI
Seen from Germany
ๆญฃ่ฃ
่ไธธ
เธเธนเนเธฃเธฑเธเธเธฒเนเธชเธตเธขเธง
Seen from Thailand
Nam 98
Seen from Indonesia
Zona Prohibida
Seen from Ecuador
Thรญch Trung Niรชn Miแปn Bแบฏc
Seen from Japan
Trends for you
Most Popular Users
Elon Musk
@elonmusk
240.7M followers
Barack Obama
@barackobama
119.2M followers
Donald J. Trump
@realdonaldtrump
111.7M followers
Cristiano Ronaldo
@cristiano
111M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
97.7M followers
NASA
@nasa
92.2M followers
Justin Bieber
@justinbieber
91M followers
KATY PERRY
@katyperry
87.9M followers
Taylor Swift
@taylorswift13
81.7M followers
Lady Gaga
@ladygaga
73.2M followers
Virat Kohli
@imvkohli
70.2M followers
Kim Kardashian
@kimkardashian
69.9M followers
YouTube
@youtube
68.7M followers
Bill Gates
@billgates
64M followers
Neymar Jr
@neymarjr
63M followers
The Ellen Show
@theellenshow
62.4M followers
CNN
@cnn
61.9M followers
Selena Gomez
@selenagomez
60.9M followers
X
@x
60.8M followers