Antid

JAKE MILLER RACKED 200 MAC STUDIOS IN HIS AUSTIN GARAGE AND KILLED A $5,508/YEAR AI SUBSCRIPTION STACK FOR $3 IN ELECTRICITY PER BOX jake was paying $459 a month across claude code max, chatgpt pro, gemini advanced, github copilot and cursor pro, the quality was magic but the burn rate was killing him he did the math on running it locally and bought 200 mac studios over 6 months, racked them in his garage on wire shelves pulling less total power than four nvidia servers claude code, chatgpt and cursor used to cost him $5,508 a year combined, all of them now run free on his local stack with one environment variable ollama added support for the anthropic messages api in january 2026, so claude code itself connects to his garage rig with one line of code, same interface, zero subscription bills apple stores ran out of mac studios in 2026 because guys like jake figured out one rack of consumer hardware kills a five figure cloud contract forever the window is open, follow and bookmark before it closes.

@0xRicker 42 saved skills while everyone else resets to zero daily

> do you understand what just happened > a guy with a $0 cloud bill and a laptop > ran the model locally this morning > turned 42 saved skills into $4,900/month > just one good process while everyone else resets to zero daily > bookmark this. seriously. right now

@vorty279 crewai for the entry, pydantic for type safety, openhands for the full loop - the funded startups run the same tools

Ai engineers make 150k. the agent stack is free everyone is hiring people who can wire an autonomous agent. the band starts around 120 to 150k base and the dirty secret is the same as with cuda. none of it is locked behind a course. the frameworks are public, the docs are free, and the gap between i use chatgpt and i ship agents is about ten hours a week for a few months 5 repos. free. open source. this is the stack that takes you from a single prompt to a multi agent system that researches, writes, and ships on its own, the same tools the funded startups actually run on 1. crewAIInc/crewAI github.com/crewAIInc/crew… the easiest entry into multi agent systems. you define a crew with roles, goals, and tools and it handles how they collaborate. 52.9k stars, independent from langchain, used inside a large share of the fortune 500. start here. the role based model maps to how you would split work between real people 2. pydantic/pydantic-ai github.com/pydantic/pydan… agents with type safe outputs from the team behind pydantic. fastapi style developer experience, 25 plus model providers, schema validation that catches agent logic errors before production. 17.6k stars, updated this week. this is the one you reach for when a wrong output format is expensive and you need guarantees not vibes 3. All-Hands-AI/OpenHands github.com/All-Hands-AI/O… a full autonomous developer environment. it browses, writes code, runs tests, and opens pull requests. 76k stars. read how the sandbox and the agent action loop are built. this is the template every autopilot coding agent now copies, and running it locally teaches more than any tutorial on agent design 4. agno-agi/agno github.com/agno-agi/agno a high performance runtime for multi agent systems with session memory, knowledge, and native mcp tool support. 40k stars, updated yesterday. this is what you reach for when crewai feels too rigid and you want speed plus control over how agents hold state across a long task 5. langchain-ai/langgraph github.com/langchain-ai/l… graph based orchestration for stateful agents. nodes are agents, tools, and checkpoints, edges are the transitions you define explicitly. 34k stars, updated this week. this is the one for long running pipelines with feedback loops, retries, and a human in the loop. langchain itself now tells you to build new agents here the build order is boring on purpose. run crewai first and feel a crew finish a task. wrap one output in pydantic ai so it never breaks downstream. only then touch openhands, agno, and graph based orchestration. clone before you read, and read the architecture before you trust the star count no shipping without a loop. the agent plans, calls a tool, checks the result, tries again. that loop is the whole job and every framework here is just a different way to write it the frameworks are public. the wiring is not. that gap is the whole game the repos are free. the llm calls cost less than a coffee an hour if you run a local model through ollama

@gippp69 ai stops being something you rent and becomes something you build on your own machine, that's the actual shift

THIS 19-YEAR-OLD GIRL OPENED THE HERMES SETUP WIZARD AND SHOWED HOW A $0 INSTALL CAN TURN A $180/MONTH AGENT STACK INTO ONE LOCAL WORKFLOW she is not showing another AI logo. she is showing the boring screen that decides the whole game: provider, auth method, Claude login, Anthropic API key, local model, and where the agent actually runs most people burn money in tiny pieces. $20 on Claude, $30-70 on API calls, $25-50 on automation tools, $99 on an agent wrapper, then another $10-20 every time the workflow breaks and needs a new tool Hermes changes the math. set it up once, connect the model once, run the same type of work through one agent, and let repeated tasks turn into saved skills instead of paying tokens to explain the same process again that is where the savings actually come from. not from “free AI”, but from not rebuilding the workflow 40 times. research reports, pricing scans, customer reviews, competitor maps, client audits, all moving through the same agent layer the setup screen looks boring until you understand what it means. whoever controls the provider, the API, the memory, and the skills folder controls the whole workflow this is why Hermes is interesting. not because it looks cool on a laptop, but because it turns AI from something you rent into something you slowly build on your own machine

@0xMortyx most people open claude, type one thing, close the tab and think that's using ai - it's maybe a tenth of it

Godfather of AI: "If you sleep well tonight, you may not have understood this lecture." The man who built the foundations of modern AI quit his job to tell us he's scared of it. He left Google specifically to warn the world about the thing he created His line on stage says it all: if this lecture lets you sleep tonight, you probably didn't understand it 47 minutes. Honestly the clearest, most unsettling explanation of where AI is actually heading that I've watched all year A few things he refuses to sugarcoat: - AI is already showing abilities nobody programmed into it - on most thinking tasks, it's already past us - "will it surpass us" is the wrong question - the real one is "when" - and you don't get to opt out of which side of that line you land on Now contrast that with how most people use this stuff: open Claude, type one thing, read the reply, close the tab. They genuinely believe that's "using AI." It's maybe a tenth of it So I did the work - watched the whole thing, then matched everything Hinton describes to what Claude can already do right now Watch the lecture first, than read the breakdown's in the post below

@ridark_eth pinterest to pollo to tiktok to shopify in one pipeline, 85% margins and zero inventory is the new agency math

a 21 year old makes $10,500/month. clothing photo. AI model video. fashion brand agency. zero inventory Pinterest → Pollo AI → TikTok → Instagram → Shopify. one pipeline. $153,680 in sales > Pinterest: source aesthetic background & outfits > Pollo AI: converts flat clothing images to realistic model videos (Seedance 2.0) > Reference to Video: inputs clothing photo + room photo. 9:16 vertical video > Social Media: millions of views. automated content factory > Shopify: clothing brands close the sale 56 cents per video. traditional agencies charge $1,500 for a single model shoot. 7 clients = $10,500/month. 85% margins 50 videos for $50. human models were the bottleneck

@Kaffchad strategy shift was dropping the hourly model entirely

@antisadh Jake's strategy shift is impressive, leveraging tech to increase earnings with less time commitment.

JAKE QUIT $2,800/MONTH FREELANCING AND BUILT A SECOND BRAIN WITH HERMES AGENT ON A $999 MACBOOK. NETS $4,400/MONTH ON 15 MINUTE REPORTS jake was grinding $2,800 a month writing landing pages for startups, 50 hours a week, every dollar tied to his time on the keyboard he found hermes agent in a reddit thread, installed it on his macbook air in 30 minutes, fed in every brief and template he had ever written hermes saved each task as a skill on his disk, by week 2 the folder had 18 files, by month 3 it had 47 and every new job finished 40% faster than the last he runs competitor research reports for early stage founders, charges $300 per report, delivers in 15 minutes what research firms charge $2,000 for 15 reports a month plus 6 retainer clients at $400 each, nets $4,400 in pure margin on $2 a month in electricity the agent gets smarter every job while jake sells, most freelancers compete on hours, jake compounded a folder of skills instead the window is open, follow and bookmark before it closes.

@beamnxw exactly, agency sells hours, jake sells reusable skills that compound

@antisadh local memory loops are completely breaking the agency model

@gippp69 the install is the front door, the real value starts when the agent stops forgetting between tasks

THIS CHINESE GUY SET UP HERMES AGENT IN 5 MINUTES, OPENED 200+ MODELS FOR $0, AND SHOWED HOW A FREE LOCAL STACK CAN REPLACE A $200/MONTH AI HABIT he walks through the setup step by step. windows, macos, linux, free api, localhost, and a clean install flow that looks way less complicated than most people assume this is the part people keep missing about Hermes. the install is only the front door. the real value starts after the first task, when the agent begins saving repeated work as reusable skills instead of forgetting everything like a normal chat tab most AI users keep renting the same output every day. Hermes turns pricing scans, review mining, research summaries, and competitor breakdowns into workflows you can run again without rebuilding the process from scratch that is why local agents are getting dangerous. not because they are flashy, but because one laptop, one model stack, and 10-20 saved skills can do the kind of repeatable work people still pay humans $300-$500 to produce watch this first, then look deeper into Hermes. the free setup is not the real story. the real story is owning the workflow, the memory, and the system that gets sharper every time you use it

@LunarResearcher 86M polymarket trades free on github and people still pay for course screenshots of the same data

My girlfriend noticed something Sunday morning. "You're smiling at your laptop." "What?" "You. At 9am. On a Sunday. Smiling." I shrugged. "Just checking the bot." She put her coffee down. "What bot?" "The trading one. Open-source. Anyone can audit the code." She stared at me. "So you code for free?" "No. The code makes money." "Sure it does." I didn't argue. I opened my laptop. One wallet I was tracking cleared $2,800 in a single weekend. Another flipped 156 trades in 48 hours with 78% winrate. One more turned $950 into $14,200 in 17 days. She stopped mid-sip. "That's... from a script?" Exactly. Then I showed her the repos. All free. All public. First: github.com/warproxxx/poly… 86M+ trades on Polymarket. Every outcome since day one. Free to download. Second: github.com/warproxxx/poly… Market making bot. Both sides of the book. Gas optimized. Google Sheets execution. Third: github.com/pselamy/patter… ML + heuristics. I fed 14,000 wallets into Claude. One prompt. 4 minutes. Found 47 traders with 70%+ winrate. Bot mirrors them with 60-second delay. She went quiet for a long time. Then: "Why are you only telling me this now?" "We've been dating nine months." "Exactly. Nine months." I didn't have a good answer. She closed her laptop lid. Looked at me. "You're going to show me everything tonight. From the top. I'm not asking." Profile: @xuanxuan008?r=lunarlunar#w7FJaKQ" target="_blank" rel="nofollow noopener">polymarket.com/@xuanxuan008?r… Girlfriends don't ask about the money. They ask why you waited nine months. Then they take over the conversation.

@myttle_web3 the model stopped answering in a box and started living inside the workflow, that's the whole shift

this guy is giving a model hands. MiniMax M3 sits behind Hermes. Ollama is the wire. the agent is the body. that sounds small until you watch the screen. the model is not answering in a box anymore. it can sit inside a workflow. open tools. carry context. run the next step. this is why 1M context matters. not because someone wants a longer chat. because agents need memory for the mess: - logs - files - failed runs - screens - tool outputs - the last thing that broke code generation was the demo. task execution is the market.

@bored2boar we are here on the anxiety leg is exactly where most retail stops listening to the people who were right last time

everything happening now was predicted by me way before. remember who told you to sell everything 2 years ago. back then, people called me crazy because they weren't ready to hear the truth. i wonder where those people are now. will tell you when to buy again, wait for it.

@0xMortyx 30 minutes of his real setup is worth more than any $500 vibe-coding course on the timeline

Boris Cherny (Claude Code's creator): "I don't prompt anymore. I write loops, and the loops prompt Claude." In 30 minutes, Boris shows his real daily Claude Code setup Claude + loops + routines + dynamic workflow Worth more than a $500 vibe-coding course

@beamnxw letting the terminal do the clicking, logging, and uploading while you just watch impressions grow is what scale actually looks like

32-YEAR-OLD CONTENT MAKER BUILT A 5-CHANNEL AI VIDEO FACTORY IN 1 AFTERNOON WHILE CMUX MANAGED EVERY AGENT one terminal. Plan mode. Claude Code. YouTube Shorts + automated production + built-in browser. $0 spent on Selenium/Puppeteer. zero code written by hand running a content empire. Reddit sync. AI agents that find trends, script, generate voiceovers, and post videos directly to YouTube Studio the "senior" automation engineer: $1,840 Anthropic bill. 11 prod incidents. 46 chaotic terminal windows open at the same time him: $47 bill. 0 incidents. 5 agents running in a single workspace with vertical splits Tab acceptance rate is not productivity. Managing 46 messy windows is not scaling. True scale is letting your terminal do the clicking, logging, and uploading while you watch the impressions grow

@gippp69 madness at home and netting more than most agencies do with a full team

@antisadh and all this is madness at home

JAKE QUIT $2,800/MONTH FREELANCING AND RUNS HERMES AGENT FROM A $999 MACBOOK. NETS $4,400/MONTH ON 15 MINUTE REPORTS jake was grinding $2,800 a month writing landing pages for startups, 50 hours a week, no leverage, every dollar tied to his time on the keyboard he found hermes agent in a reddit thread, installed it on his macbook air in 30 minutes, ran a competitor research brief and delivered the same report a research firm charges $2,000 for in 15 minutes first week he sent 50 cold messages on upwork offering competitor reports at $300, closed 3 clients, delivered 3 reports in under an hour total by month 3 his hermes folder had 47 saved skills, every new report finishes 40% faster than the last because the agent reuses what worked before he charges $300 per report and runs 15 a month, plus 6 retainer clients paying $400 each. nets $4,400 in pure margin on $2 a month in electricity the window is open, follow and bookmark before it closes.

@N01ennn one review session per week, everything else dispatched overnight from her phone — that's not a content job, that's an operation

A 19 YEAR OLD GIRL MAKES $10,000-15,000/MONTH MAKING VIDEOS FOR IPAD KIDS. CALLS IT EMBARRASSING. THE BILLS DON’T CARE finds viral kids video. copies description. pastes into Flik AI. downloads. uploads. millions of views but one video is still one video. the real money is the system. five channels. Claude as operations manager. Mac Mini runs 24/7 Claude Computer Use controls the screen. Skills load per channel at task init. Dispatch queues overnight from phone. iCloud syncs every output automatically. Google Calendar never misses a publish one review session per week. everything else automated $799 Mac Mini. $55/month tools. month 12: $8,000-12,000/month kids rewatch the same video 50 times. the algorithm rewards it. Claude generates faster than any studio

@gippp69 most agents forget the second you close the tab, hermes writing skills to disk is the difference between a tool and an employee

@insomnia_vip @CyfrinUpdraft blame the hacker after the exploit or learn to read the contract before it ships, only one of those scales

True peace of mind in web3 is just a free @CyfrinUpdraft course away