Yarchi

@0xMovez brilliant lecture

Anthropic engineer: "We were spending $90 running agents on Opus. Sonnet 5 did the same thing for $20. Same outcomes. Sonnet 5 hits Opus-class coding at 1/4 the cost" In this 1-hour workshop, Anthropic reveals how to build low-cost agentic systems with Sonnet 5. Worth more than a $500 agentic orchestration course. Watch the session, then read the guide below ↓

@rohit4verse goated article man

@0x_fokki how to set it up?

🚨 3 CLAUDE CODE SUB-AGENTS DO WHAT A DEV TEAM CHARGES $5,000 TO SCOPE free. no extra tools. you run them global, they hit every project. > Explorer: finds the exact code across the repo before you burn a prompt hunting for it > Research Documenter: pulls the external docs and API specs you were about to google > Historian: saves markdown checkpoints of every decision, more readable than a GitHub commit will ever be that last one runs as a context engine. it holds why you built something, not just what changed. then scale it. 6 in parallel, a /goal agent voting done true or false separate from the worker, and a full GTM kit lands in 8 minutes. bookmark this before your agent torches your next sprint.

@rewind02 This is the single highest leverage AI setup rn

Andrej Karpathy just showed off his "second brain" and it reveals all the hidden connections the idea: every thought is linked to everything else - decisions, research, observations, past experience instead of starting from zero every time, he stores everything in one place and uses AI to: - find the info he needs - surface hidden connections between ideas - resurface old thoughts exactly when they're relevant - help him make decisions based on his entire accumulated context that's the whole point of a "second brain." it's not just about storing information - it's about turning it into a system that gets smarter and more useful over time most people build a pile of notes Karpathy built an operating system for his own thinking so how do you build a "Second Brain" for Claude Fable 5? 👇

This is the single highest-leverage AI setup you can build, and I don't get why more people aren't doing it Andrej Karpathy posted the idea and it hit millions of views: stop using AI to write code, use it to build a wiki of everything you know. What people actually use it for: > Research that compounds. Drop in every article, paper, and transcript on a topic. Ask across all of it at once. Setup: a raw folder for sources, a wiki folder for the pages Claude writes, and one instruction to ingest. > Never repeating yourself. Your business context, your projects, your constraints all live in the vault. Every new chat starts with Claude already knowing them. > Decisions with a paper trail. Log what you tried, what worked, what burned, and why. Six months later, ask why you made a call and get the actual reasoning back. > Finding what you already knew. That article you saved, that idea you had, that problem you already solved. It surfaces on question, not on memory. > Writing from your own material. Ask it to draft something and it pulls from your notes and your voice, not the internet's average. Everything stays in plain markdown, so it's yours and it ports to any model. Five minutes to set up, and you never start from a blank chat again. Bookmark this

@ridark_eth goated tool, i've been using it for the last few months

Paying developers by the hour is the old way. $15,000 in one month, and the team was Claude agents. Not one agent. A pipeline. One plans the work, one writes the code, one tests it, one checks for breaks. Each agent runs its piece. They share context. Nothing waits on a human in the middle. The billing stack matters. Basic tasks route to cheaper models. Complex tasks go to the capable ones. That gap cuts API costs significantly. RuFlow handles the routing automatically. 60 agents can run at once. It has 14,100 stars on GitHub and costs nothing to run. The developer pairs it with Obsidian for notes and task structure. Prompts go in, output comes out, the pipeline handles the rest. Month one on this setup: $15,000. That number is from one person's workflow, not a case study. The Claude subscription stays. The $200 platform fees go. The difference is who controls the stack. Open source means you fork it, adjust it, run it on your own infra. No vendor owns the ceiling.

@Nekt_0 3am grind hits different

Karpathy sat with Lex and explained why 3AM is his best working hour. No meetings. No messages. No errands. No humans pulling the problem out of his head. He said hard work needs days of uninterrupted context. You load the problem into mental RAM, keep it there, and let the answer compound instead of restarting from zero every morning. That was the whole productivity system. Not a calendar hack. Not a morning routine. Just protecting the loop long enough for the work to finish. The article is pointing at the same thing from the AI side. Agents only become useful when they stop acting like one-off chats and start running in loops: take context, act, check, update, repeat. Same pattern, different machine. Karpathy does it with attention. AI agents do it with context windows, tools and memory. Prophet does it with markets. @prophetmarketai turns a future event into a yes/no market, then lets an AI price it and take the other side. You are not waiting for some random person to match your view. You create the market, make the forecast, and the system gives it a live interface. That is the interesting part. Opinions are cheap until they are forced into a structure. Will this happen or not? What price is the AI giving it? Is your read sharper than the market? Deep work needs a clean loop. Conviction needs a clean market. Prophet is not available in the US. Predicting carries risk app.prophetmarket.ai/?ref=nOUbD3yOM…

@gippp69 must have for every ai user

THIS F**KING 7,300 STAR REPO TURNS CLAUDE FABLE 5, CODEX, AND CURSOR INTO ONE AI TEAM THAT KEEPS WORKING WHILE YOU SLEEP 00:01 he opens Omnigent, an open source layer that puts multiple coding agents inside one workflow while sandboxes and policies control exactly what they can touch. setup takes minutes. Fable acts as the supervisor, cheaper models handle execution, and premium tokens are reserved for decisions that actually need judgment. each task gets split across separate agents and git worktrees, then a fresh model reviews the final diff before anything can merge. a $5 daily cap, 150 line approval limit, automated signoff tests, and a 95% pass rate keep the whole team from running wild. bookmark this. one repo, four agents, and roughly 15 to 20% of the usual token cost.

@slash1sol goated lecture bro, thanks for sharing

THE GUY WHO RAISED $55.5M FOR HIS AI COMPANY AND CO-CREATED STANFORD'S DEEP LEARNING CLASS WITH ANDREW NG -- JUST GAVE A 90 MINUTE MASTERCLASS ON WHY FINE TUNING IS A TRAP AND HOW REAL TEAMS BUILD AI AGENTS IN 2026 Kian Katanforoosh -- CEO of Workera. Clients include Samsung, Accenture, Siemens, and the US Air Force. Taught AI to 4 million people and won Stanford's highest teaching award. His pitch: the difference between a toy and a product isn't the model -> it's the architecture around it. > The jagged frontier: AI helps in some tasks and quietly ruins others. BCG consultants who trusted it outside its lane "Fell asleep at the wheel". > Two survival modes: Centaurs (Delegate big blocks) and Cyborgs (Rapid back and forth). > The fine-tuning trap: By the time you finish tuning on your Slack history, the next base model already beats you. Ross Lazerowitz tried it -- got a lazy coworker who kept saying "I'll do it tomorrow" instead of actual work. > Chaining beats one giant prompt: Extract -> Plan -> Write. Every step debuggable and every step testable. > RAG isn't optional: Vector DB + HyDE (generate a hypothetical answer, then search for it). This is what stops your model from hallucinating in production. > MCP is the shift: Anthropic's protocol so agents discover what APIs can do without you hard coding every integration. > The paradigm change: Engineers used to write deterministic code. Now you design "fuzzy" systems, think like a manager, decompose work into roles, keep a human in the loop for the messy parts. Stop trying to build a bigger model. Learn to orchestrate the ones we already have. Watch it, then Bookmark ↓

@andreysuperior this is so smart setup

Do you actually understand what this LLM wiki becomes when you build it for companies instead of yourself. A shared brain. Built once. Reads by every AI forever. Gets better the more you feed it. He heard that and thought about who would pay $1,500 a month to have it built and maintained for them. A law firm associate types a question. Three seconds later the wiki answers. Billing codes, client protocols, matter procedures. No partner interrupted. No time wasted. He set that up in three days. $2,200 once. The firm pays $1,500 every month to keep it current. He runs six of these. Different industries. Same structure. Same Claude. Same retainer. $9,000 a month. Ten hours of work a week. The wiki gets smarter every time he adds to it. The clients get more dependent every month they use it. Nobody cancels a system that knows everything about their business.

@Suryanshti777 this changes everything

Anthropic just handed Claude Code users a massive unfair advantage. Almost nobody is talking about it. They released an official prompts library packed with ready-to-use prompts for nearly every development workflow. Thousands of developers are still: • Writing prompts from scratch • Getting inconsistent outputs • Repeating the same tasks manually Meanwhile, the people using this library are shipping faster because they're starting from proven templates. The biggest lesson isn't the prompts. It's seeing how Anthropic structures context, instructions, and reasoning. This might be one of the highest-signal free resources for developers in 2026. Bookmark it now. In a few months, this will probably become the default playbook for Claude Code power users.

@slash1sol nice stack. waht's the total cost?

A $800 USED RTX 3090 JUST HUMILIATED NVIDIA'S $4,000 DGX SPARK ON LOCAL LLM SPEED Someone benchmarked 19 GPUs and found that the "AI-ready" flagships get destroyed by used gaming cards. VRAM decides what runs. Bandwidth decides what actually flies. Same 13B model, only the memory bus changed: 936 GB/s -> 19ms per token. 256 GB/s -> 70ms per token. 3.7x speed difference from bandwidth alone. The traps: > NVIDIA DGX Spark, $4,000, 128GB memory but only 273 GB/s bandwidth. > AMD Ryzen AI Max+ 395 mini PC, up to $4,000, 96GB unified memory at 256 GB/s. The actual value picks: > Used RTX 3090 -> $600 to $1,050 -> 24GB at 936 GB/s. > RX 7900 XTX -> $800 to $1,000 -> 24GB at 960 GB/s. The cores were never the bottleneck. The memory bus was, and the marketing has been lying about which spec to look at. The full 19-GPU ranking is in the article. Save this before you overpay for a "supercomputer" that cannot feed itself ↓

@doublenickk waiting for apple to release something like that

APPLE JUST GOT EMBARRASSED BY A $249 TOY We've been told we need massive cloud data centers or premium Mac Minis to run real-time AI. Turns out, that’s just brilliant marketing. A tiny $249 Nvidia Jetson developer kit just absolutely destroyed a Mac Mini at real-time object detection, clocking 33 FPS compared to the Mac's measly 21. The real AI revolution isn't happening in the cloud. It’s on the Edge: Zero Cloud, Zero Subscriptions: 100% of the computation is on-device. Nothing ever leaves the hardware. Insane Local Power: From GPS-free drones running visual SLAM to local coding assistants pumping out Llama 3.2 at 17 tokens/sec, devs are proving that optimization beats raw server size. This is why on-device SLMs (Small Language Models) are the real threat to big tech.

@N01ennn no way. this is gold!

YOU ARE NOT GOING TO BELIEVE WHAT THE OBSIDIAN CEO JUST OPEN-SOURCED kepano dropped his private Claude Code stack. 5 skills. MIT license. 40,000 stars in weeks. zero pitch, zero cloud, zero plugin store the skills teach Claude to read Obsidian files the way a human would, respect wikilinks, edit JSON Canvas without breaking anything, and strip ads off any URL for a clean note "file over app" was the essay. "file over agent" is the workflow now your notes are files. the app is disposable. Claude just learned the format save this before this becomes the standard for every note-taking app on your laptop👇

@insomnia_vip pure alpha

A 22 YEAR OLD CHINESE PROGRAMMER IS BUILDING CINEMATIC ADS WITH CLAUDE INSTEAD OF HIRING A VIDEO TEAM One MCP connection turns Claude into a creative workflow that can generate polished commercial videos from nothing more than a detailed text prompt in a single chat Instead of learning complex editing software or paying expensive freelancers the entire production process becomes writing better prompts and letting the tools handle the execution As AI keeps removing production bottlenecks the biggest opportunity is no longer creating content faster but creating more valuable systems around it People learning these workflows early will have the biggest advantage

@vanvster nice setup actually

my colleague built a $440 ai stack for $0 cursor pro perplexity pro github copilot claude code on free inference then i found the same map shows founders how to unlock up to $400k in api and cloud credits halfway through i realized nearly every major ai lab has a hidden free path сheck the full map below before the offers change

@Nekt_0 looks interesting

5 AI AGENTS IN A FAKE OFFICE LOOKS LIKE A GAME UNTIL YOU REALIZE THE ACTUAL PRODUCT IS CONTROL. 00:01 the screen looks like The Sims for work: agents at desks, tasks moving through kanban, project status on the side, GitHub sync in the background. The important part is not the cartoon office. It is the interface. Once software starts acting on its own, the question changes from “what can AI do?” to “how do I see what it is doing before it breaks something?” That is also why @prophetmarketai is interesting. Prediction markets usually need another person on the other side. Prophet makes it single-player: you create a yes/no market, the AI prices it, and the AI is the counterparty. No waiting for a fixed market list. No needing someone else to take the other side. If your niche has a real future event people care about, you can create the market around it. That matters for the same reason this agent UI matters. AI is moving from chat into systems with state, pricing, actions, wallets, markets, dashboards and outcomes. The interface becomes the product. Prophet is not available in the US. Predicting carries risk. Try Prophet: app.prophetmarket.ai/?ref=nOUbD3yOM…

@gippp69 it feels kinda illegal

SOMEONE GAVE CLAUDE FABLE 5 A CONTRACT, A $5 DAILY BUDGET, AND A MANAGER. THIS F**KING DANGEROUS SYSTEM CAN CLEAR A WEEK OF BACKLOG WHILE YOU SLEEP. 00:03 he opens Microsoft’s 4,900 star Agent Governance Toolkit, built for policy enforcement, isolated execution, identity controls, and protection against all 10 OWASP agentic risks. setup starts with 3 files. CONTRACT MD sets the limits, boundaries MD defines what the agent can touch, and signoff SH runs every test before anything ships. four roles split the shift. one model reads the logs, Fable picks the highest value task, another writes the code, and a fresh Fable reviews the final diff. a $5 cap stops runaway sessions, commits above 150 lines need approval, and autonomy unlocks only after 20 runs at a 95% pass rate. one failed run sends the agent back to probation. the repo adds the guardrails, but the contract is what turns Claude from a chatbot into an employee.

@0x_fokki this is insane

🚨 Disney spent over $150,000,000 making a Zootopia movie a 23-year-old American student recreated its entire red-carpet premiere on a laptop for $50, and the clip is pulling millions of views. Margot Robbie as Judy Hopps. Dwayne Johnson as a horse in a tuxedo. Shakira as Gazelle. none of it real. these celebrity premieres are the format taking over your feed right now. > Research scores the outliers: views ÷ channel median, over 30 and you copy that format this week > Claude reskins the winning shape onto a new subject and writes the shot list: 20 minutes > CapCut generates every celebrity, every reveal, exports 9:16: 1 hour > Make posts to 3 platforms and reads the view counts back in 48 hours 4 tools. $50/month. Disney pays a studio to premiere one film. this student ships a new one every week. first people to run the format eat the whole trend. the whole factory is in the article above👇

@Nekt_0 no way it's such easy

A $3K LANDING PAGE NOW STARTS WITH CLAUDE, A GITHUB SKILL, AND ONE MCP CONNECTION INSTEAD OF A DESIGNER BRIEF 00:01 he searches UI UX Pro Max on GitHub, opens Claude, installs the skill, connects Magic MCP, and asks for a landing page like it is a normal prompt. the result is not just "AI made a website." the useful part is the stack behind it. Claude gets design rules, UI patterns, component logic, and an execution layer before it starts building. one skill changes the job. no blank chat, no generic template, no guessing what good UI means from a single sentence. Magic MCP gives the model a better way to act. UI UX Pro Max gives it taste scaffolding. Claude turns that into pages like “Design is Everything” and “Launch Your Workflow Into Orbit” without starting from zero. the model is not the edge here. the harness is. a plain Claude prompt gives you a draft. Claude plus skills gives you a workflow. that is why this stops being "Claude killed designers" and starts looking more like every designer having to compete with someone who knows how to wire the tools together.