kocer

@kocer_eth • 3,174 subscribers

useful AI systems, prompts and workflows dm open

Shorts

SEVEN RTX 3090S IN A WATER TANK FOR AI SERVER it is a private AI server with the power bill moved into your room. not a clean Mac mini. not a quiet box under a monitor. loose vertical GPUs sit inside a transparent tank. bubbles rise through distilled water. ALLIED CONTROL is printed on the side. it looks closer to a lab accident than a normal workstation. but the logic is obvious: seven RTX 3090s = seven 24GB cards. that is the used-market shortcut for people who want local inference without paying cloud tax on every run. put Ollama, llama.cpp, vLLM, Open WebUI, Tailscale, Qwen, DeepSeek, or Llama on top. now the box can handle client files, code agents, scraping jobs, evals, transcription, and boring overnight work. not because it beats frontier cloud models. because it changes the bill shape. no rate limit. no per-token anxiety. no sensitive client context leaving the building. no monthly stack quietly turning into rent. the ugly part is physical. seven 3090s can pull serious power, dump serious heat, and punish lazy cooling. distilled water is the weird visual, not a setup tip. real immersion rigs live or die on coolant chemistry, insulation, pumps, maintenance, and whether the room can handle the heat. local AI PCs are becoming less like gaming builds and more like small private data centers. the early question is not: can it run ChatGPT? it is: what work is repetitive, private, expensive in the cloud, and worth owning in hardware?

SEVEN RTX 3090S IN A WATER TANK FOR AI SERVER it is a private AI server with the power bill moved into your room. not a clean Mac mini. not a quiet box under a monitor. loose vertical GPUs sit inside a transparent tank. bubbles rise through distilled water. ALLIED CONTROL is printed on the side. it looks closer to a lab accident than a normal workstation. but the logic is obvious: seven RTX 3090s = seven 24GB cards. that is the used-market shortcut for people who want local inference without paying cloud tax on every run. put Ollama, llama.cpp, vLLM, Open WebUI, Tailscale, Qwen, DeepSeek, or Llama on top. now the box can handle client files, code agents, scraping jobs, evals, transcription, and boring overnight work. not because it beats frontier cloud models. because it changes the bill shape. no rate limit. no per-token anxiety. no sensitive client context leaving the building. no monthly stack quietly turning into rent. the ugly part is physical. seven 3090s can pull serious power, dump serious heat, and punish lazy cooling. distilled water is the weird visual, not a setup tip. real immersion rigs live or die on coolant chemistry, insulation, pumps, maintenance, and whether the room can handle the heat. local AI PCs are becoming less like gaming builds and more like small private data centers. the early question is not: can it run ChatGPT? it is: what work is repetitive, private, expensive in the cloud, and worth owning in hardware?

591,697 views

xAI EMPLOYEES LEFT 50 GROK BOT PROMPTS ACROSS A 72-HOUR BUILD AND PUBLIC GUIDES. COPY THEM IN THIS ORDER. at 11:40 p.m., i nearly saved the viral list and moved on. it had 30 prompts attributed to Lauren Tan, Matt Palmer, and Roshan Sadanani during their 72-hour xAI company build. then i opened Lauren's public pstack guides and found the missing engineering layer. i adapted all 50 into one paste-ready sequence and kept the source beside every prompt. source key: [RS] Roshan, [MP] Matt, [LT] Lauren, [pstack] Lauren's public pstack guides. find the market › 01. [RS] mine public posts for agent problems. rank each by buyer clarity, willingness to pay, and speed to launch. › 02. [RS] analyze customer conversations from the last 48 hours. extract repeated pain, urgency, budget signals, and named buyers. › 03. [RS] compress the research into one thesis: customer, painful problem, wedge product, and why it must exist now. › 04. [RS] define who pays, the first job to be done, and the fastest internal test of the product thesis. › 05. [RS] map the two strongest segments by ICP, pain, offer, price range, and five Day 1 interview questions. › 06. [RS] design outreach and onboarding for the first design partner. include the inputs needed to define measurable success. › 07. [MP] find and rank 25 potential customers. identify the best contact path and flag feasibility or integration risks. › 08. [MP] generate 15 company names. score the best five by domain, email format, memorability, and product fit. › 09. [MP] find the fastest path from our current assets to the first $1. define the offer, price, buyer, and payment method. › 10. [RS] find the smallest MVP we can presell. list what must be real before payment and what can wait. design the company › 11. [LT] ask me five questions about the company. then design the first three specialist bots with memory, skills, and initial tasks. › 12. [LT] turn one general bot into a Chief of Staff. it should triage Slack and interrupt me only for money or product decisions. › 13. [LT] turn this generic bot into a specialist. define its role, tone, tools, workflow, escalation rule, and success metric. › 14. [MP] assume zero employees and 72 hours. design the smallest bot team for research, product, engineering, growth, and operations. › 15. [MP] turn Slack into the control plane. create channels for product, engineering, bots, prospects, and customer signups. › 16. [MP] create Operations and Creative Director bots. give each one routine, one metric, and one human escalation rule. › 17. [LT] audit marketplace bots for research, outbound, inbox, engineering, and design. keep only bots that remove friction. › 18. [RS] divide the next 24 hours across three humans and their bots. assign one owner and one definition of done per workstream. › 19. [MP] write a five-line shipping policy for humans and bots. cover speed, direct deployment, rollback, ownership, and review. › 20. [RS] create a Designer bot that delivers a company name, three logo directions, and a landing-page hero within 60 minutes. build and ship › 21. [LT] install an engineering skill pack for tests, PR hygiene, and issue-to-ship loops. remove anything unnecessary for Day 3. › 22. [LT] create a Prototype Engineer that ships a landing page with email signup today. use no database unless it is essential. › 23. [LT] use Grok Bot for research and disposable prototypes. define the exact handoff package for production engineering. › 24. [LT] create an Engineer bot that owns the repository, ships small changes, writes minimal tests, and avoids bikeshedding. › 25. [LT] create the full visual direction: brand voice, hero copy, CTAs, colors, typography, and an implementation-ready brief. › 26. [LT] design the thinnest stack that can accept payment by Day 3. separate what must be built from what can be simulated. › 27. [MP] name the company around its shipping deadline. create the GitHub organization and write the first product README. › 28. [MP] connect the signup form to Notion. define the database fields, follow-up workflow, and confirmation message. › 29. [MP] connect the repository to Vercel. document the path from first preview to production and custom domain. › 30. [RS] map the path from waitlist to first paid customer to subscription. track daily signups, calls, sales, and revenue. understand and plan › 31. [pstack] read this Slack thread. restate the underlying issue in plain English before proposing any solution. › 32. [pstack] trace how works at runtime. show the entry point, data flow, state changes, dependencies, and failure paths. › 33. [pstack] explain why we still use . check Git history, PRs, tickets, docs, Slack, and production evidence. › 34. [pstack] teach me why you implemented it this way. compare the strongest alternative and explain every tradeoff. › 35. [pstack] recall yesterday's work on . combine that context with this new bug report before taking action. › 36. [pstack] recall seven days of related work. explain the current design, identify repeated failures, and propose a replacement. › 37. [pstack] prototype three options for . keep them disposable and compare them with measured evidence. › 38. [pstack] build two UI variants behind a switcher. use /control-app to record each flow and capture review screenshots. › 39. [pstack] /architect before implementation. surface open questions and answer them with small prototypes. › 40. [pstack] turn the approved design into a multi-phase plan. keep every PR small and attach proof to every completed task. verify and improve › 41. [pstack] investigate why background workers time out. separate facts, evidence, unknowns, and ranked hypotheses. › 42. [pstack] architect rate limiting for external webhooks. test open questions with prototypes and wait for my review. › 43. [pstack] plan the StyleX migration as small, verifiable PRs. require visual regression tests and live checks for each PR. › 44. [pstack] reproduce this bug with /control-app. if it exists on main, fix it and return a video showing the working result. › 45. [pstack] use /poteto-mode to build . verify it with /control-app and return screenshots plus a short video. › 46. [pstack] spawn a cloud agent to build with /poteto-mode. require /control-app evidence before reporting completion. › 47. [pstack] improve initial load time. capture a baseline trace, make one targeted fix, then use a swarm to confirm the win. › 48. [pstack] evaluate this skill change. run old and new versions on the same tasks, compare failures, and show the evidence. › 49. [pstack] run /create-verification-skill for this app. build a compact control CLI and a searchable feature map. › 50. [pstack] run /maintain-verification-skill daily. update stale commands, feature paths, and known verification failures. the useful part is not any single prompt. each output becomes the next prompt's context: market becomes offer, offer becomes product, product becomes evidence. that is when Grok Bot stops producing isolated answers and starts carrying a company build forward. bookmark the sequence. the order is the asset.

84,359 views

500 agents shouldn't be running all day. they should spin up for the exact window a task needs, finish it, and disappear until the next trigger. the bigger the swarm gets, the less any single agent matters. what matters is the system wiring them together. 100 agents can form up to 4,950 possible pairwise connections. 500 = 124,750. when a real signal lands, it spins up the whole workforce: 1 trigger → 100–500 Kimi agents → 5 live data feeds → up to 4,000 steps → verified artifact and that's before you add: sources → claims → memory → tool calls → contradictions → retries so the architecture has to fold all of that activity into one shared state, continuously. more agents just buys you more raw compute. the graph is the only thing standing between 124,750 possible connections and pure noise.

500 agents shouldn't be running all day. they should spin up for the exact window a task needs, finish it, and disappear until the next trigger. the bigger the swarm gets, the less any single agent matters. what matters is the system wiring them together. 100 agents can form up to 4,950 possible pairwise connections. 500 = 124,750. when a real signal lands, it spins up the whole workforce: 1 trigger → 100–500 Kimi agents → 5 live data feeds → up to 4,000 steps → verified artifact and that's before you add: sources → claims → memory → tool calls → contradictions → retries so the architecture has to fold all of that activity into one shared state, continuously. more agents just buys you more raw compute. the graph is the only thing standing between 124,750 possible connections and pure noise.

41,841 views

FIVE LAYERS OF AGENT ENGINEERING, EACH ONE WRAPS THE ONE BELOW IT. IF YOU SKIP LAYER 2, YOUR LAYER 5 WILL LOOK BROKEN WHEN IT IS ACTUALLY JUST STANDING ON NOTHING. for weeks i debated harness vs loop vs graph like they were competing choices. then a stack diagram made the shape obvious. they are not choices. they are floors. 01 | prompt engineering. the message. unit of work: one input. inputs are role, instructions, examples, format. output is a single raw response. 02 | context engineering. the memory. unit of work: what stays in the window. a curator selects, compresses, and drops from query, docs, memory, prior turns, and tool outputs before the prompt runs. 03 | harness engineering. the machine. unit of work: the machine itself. gather (context + prompt) → LLM → tools or sub-agents → verifier → final response. the article calls this the operating environment. 04 | loop engineering. the system. unit of work: the run. goal + success criteria + max iterations + budget + completion check wrap around one harness pass. failed pass appends results to context and retries. 05 | graph engineering. the topology. unit of work: the graph run. goal + nodes + edges + state schema. graph routes to agent nodes, tool nodes, or human approval. a reviewer node with a different model and fresh context checks the final answer. the wrapping is the whole point. layer 5 assumes layer 4 works. layer 4 assumes layer 3 works. skip layer 2 and layer 3's verifier keeps failing without a clear reason. this is why swapping the model is a one-day project and swapping the stack is a quarter. the model is the commodity. the five layers around it are the engineering. full three-layer breakdown of the top of the stack (harness, loop, graph) in the post below.

FIVE LAYERS OF AGENT ENGINEERING, EACH ONE WRAPS THE ONE BELOW IT. IF YOU SKIP LAYER 2, YOUR LAYER 5 WILL LOOK BROKEN WHEN IT IS ACTUALLY JUST STANDING ON NOTHING. for weeks i debated harness vs loop vs graph like they were competing choices. then a stack diagram made the shape obvious. they are not choices. they are floors. 01 | prompt engineering. the message. unit of work: one input. inputs are role, instructions, examples, format. output is a single raw response. 02 | context engineering. the memory. unit of work: what stays in the window. a curator selects, compresses, and drops from query, docs, memory, prior turns, and tool outputs before the prompt runs. 03 | harness engineering. the machine. unit of work: the machine itself. gather (context + prompt) → LLM → tools or sub-agents → verifier → final response. the article calls this the operating environment. 04 | loop engineering. the system. unit of work: the run. goal + success criteria + max iterations + budget + completion check wrap around one harness pass. failed pass appends results to context and retries. 05 | graph engineering. the topology. unit of work: the graph run. goal + nodes + edges + state schema. graph routes to agent nodes, tool nodes, or human approval. a reviewer node with a different model and fresh context checks the final answer. the wrapping is the whole point. layer 5 assumes layer 4 works. layer 4 assumes layer 3 works. skip layer 2 and layer 3's verifier keeps failing without a clear reason. this is why swapping the model is a one-day project and swapping the stack is a quarter. the model is the commodity. the five layers around it are the engineering. full three-layer breakdown of the top of the stack (harness, loop, graph) in the post below.

31,198 views

MY GROK BOT / OBSIDIAN MESH SYNCS 20 AGENTS ACROSS A LIVE KNOWLEDGE GRAPH. NOT ONE OF THEM REPORTS TO ME. I FIXED THAT BY HIRING A 21ST BOT WHO DOES NOTHING BUT ROUTE. i watched the mesh light up this morning. vector, canvas, oracle, extract, all 20 of them syncing notes. writing backlinks, healthy status, 190 edges active. it looked like a team. it was a graph. a graph has no chain of command, it just has connections. i had been treating 20 well-wired agents as oversight. they were routing information, not decisions. nobody in that mesh could tell me what mattered first. so i hired one more bot and gave him a different job. not indexing notes, holding the account. 1. Reports-to line in the charter › names me as the human above him, closes the drift where a bot starts acting like the principal instead of the agent. 2. no source, no number › if he can't cite where a figure came from, he leaves it out. the 20-agent mesh never needed this line. › Chief can't run without it. 3. stop list written as verbs › "sending anything to a person" instead of "communications." a verb he can match, not a category he can talk around. 4. Rule 6, approval belongs to the human › he never signs off on another bot's irreversible action, only i do. the mesh has no equivalent rule. › it was never built to approve anything, only to sync. 5. STOP ALL, the kill switch › one message halts every routine and every task he's running, reports what was in flight, and waits. the mesh will keep syncing whether i watch it or not. Chief is the only bot in the account who stops the second i tell him to. that's the difference between a graph that moves information and a bot that's actually accountable for it.

MY GROK BOT / OBSIDIAN MESH SYNCS 20 AGENTS ACROSS A LIVE KNOWLEDGE GRAPH. NOT ONE OF THEM REPORTS TO ME. I FIXED THAT BY HIRING A 21ST BOT WHO DOES NOTHING BUT ROUTE. i watched the mesh light up this morning. vector, canvas, oracle, extract, all 20 of them syncing notes. writing backlinks, healthy status, 190 edges active. it looked like a team. it was a graph. a graph has no chain of command, it just has connections. i had been treating 20 well-wired agents as oversight. they were routing information, not decisions. nobody in that mesh could tell me what mattered first. so i hired one more bot and gave him a different job. not indexing notes, holding the account. 1. Reports-to line in the charter › names me as the human above him, closes the drift where a bot starts acting like the principal instead of the agent. 2. no source, no number › if he can't cite where a figure came from, he leaves it out. the 20-agent mesh never needed this line. › Chief can't run without it. 3. stop list written as verbs › "sending anything to a person" instead of "communications." a verb he can match, not a category he can talk around. 4. Rule 6, approval belongs to the human › he never signs off on another bot's irreversible action, only i do. the mesh has no equivalent rule. › it was never built to approve anything, only to sync. 5. STOP ALL, the kill switch › one message halts every routine and every task he's running, reports what was in flight, and waits. the mesh will keep syncing whether i watch it or not. Chief is the only bot in the account who stops the second i tell him to. that's the difference between a graph that moves information and a bot that's actually accountable for it.

21,984 views

PART 2: I FOUND 20 MORE PROMPTS IN LAUREN TAN'S PUBLIC PSTACK. USE THEM IN GROK BOT WHEN "BUILD THIS" STOPS WORKING. part 1 gave you 50 prompts pulled from the public xAI company build. for part 2, I went through Lauren Tan's public pstack, which is available inside her Grok Bot. I converted 20 real playbooks into paste-ready commands for when a build starts touching production. 1. find the failure › 01. reproduce [bug]. find the root cause, fix it, and prove it at runtime. › 02. profile [slow flow]. show the baseline, identify the bottleneck, and measure the result after the fix. › 03. diagnose [live symptom] from instrumentation. do not guess from the source code alone. › 04. inspect [trace]. rank the likely causes and verify the dominant one in the running product. 2. settle the design › 05. /architect settle the caller, types, data shape, and module boundary before writing code. › 06. /blast-radius trace every caller this change could break. prove the condition that makes it safe. › 07. build two throwaway prototypes. assign one agent to each and compare them side by side. › 08. /arena run my request unchanged across multiple models. combine only the strongest parts of each proposal. 3. build against reality › 09. build [feature] behind a feature flag. verify the user-visible behavior inside the real app. › 10. compare [reference screenshot] with [current screenshot]. reproduce the mismatch and iterate until they match. › 11. /tdd reproduce this bug with a failing test first. implement the smallest fix and rerun the test. › 12. migrate every caller to [new API]. preserve existing behavior and delete the legacy path. 4. attack the work › 13. /swarm give one package to each worker. run its check script and return one consolidated report. › 14. /interrogate review this PR. ask several models to find correctness, design, and code-quality failures. › 15. /show-me-your-work keep a reviewable trail of decisions, rejected alternatives, and evidence used. › 16. split [migration] into verifiable phases. make sure every phase ends in a shippable state. 5. finish without babysitting › 17. watch PR 123 until it is merge-ready. resolve conflicts, review threads, and CI failures. › 18. verify every PR independently. report exactly what prevented any PR from merging. › 19. /reflect capture why this task took too long. turn the lesson into a reusable skill or automated check. › 20. /automate-me study my recent workflows and draft a personal mode that routes through pstack. with these prompts, Grok Bot stops being a code generator and becomes an engineering workflow you can inspect. bookmark part 2 for the first task that survives the demo but fails in production.

14,290 views

YOUR GROK BOT HAS YOUR LOGINS AND ITS OWN CLOUD COMPUTER. WITHOUT A RULEBOOK, IT WILL SPEND, POST, OR DELETE SOMETHING BY MONDAY THAT YOU CAN'T UNDO. saturday, 8am. i opened the laptop and started counting what a fresh bot had done overnight on the broad task i left him with. signed up for a paid trial in my name. replied to two DM. deleted a file it decided was clutter. none of it was catastrophic. all of it was mine to clean up on the weekend. the bot was doing exactly what i asked. i had just never told him what he was not allowed to do. so i wrote three files before the next task ran. 1. layer one: charter. one page in Bot actions, Description. name, who he reports to, what he owns, and the exact verbs he must park for me. send, publish, spend, delete, sign, sign-up. 2. layer two: rulebook. ten one-line rules he acknowledges before touching anything. rule 1 (if i cannot undo it in a minute, i park it) closed most of the friday-night damage. rule 9 (never paste a password, sign-in goes through the Agent Computer handoff) closed the rest. 3. layer three: routines. five saved jobs, not seven. morning brief at 7, weekly review on friday, end-of-day handoff at 6, silent-inbox catcher, monthly audit. anything else lives in me, not in the bot. the day after i pasted the three files, the same task came back with three parked questions and one finished deliverable. no signups. no DMs. no deletes. your bot is not scary because it is smart. it is scary because it has your credentials and no one told it what stop looks like. paste the charter and the ten rules before the next task runs. the routines can wait until the first parked question comes back the way you wanted.

YOUR GROK BOT HAS YOUR LOGINS AND ITS OWN CLOUD COMPUTER. WITHOUT A RULEBOOK, IT WILL SPEND, POST, OR DELETE SOMETHING BY MONDAY THAT YOU CAN'T UNDO. saturday, 8am. i opened the laptop and started counting what a fresh bot had done overnight on the broad task i left him with. signed up for a paid trial in my name. replied to two DM. deleted a file it decided was clutter. none of it was catastrophic. all of it was mine to clean up on the weekend. the bot was doing exactly what i asked. i had just never told him what he was not allowed to do. so i wrote three files before the next task ran. 1. layer one: charter. one page in Bot actions, Description. name, who he reports to, what he owns, and the exact verbs he must park for me. send, publish, spend, delete, sign, sign-up. 2. layer two: rulebook. ten one-line rules he acknowledges before touching anything. rule 1 (if i cannot undo it in a minute, i park it) closed most of the friday-night damage. rule 9 (never paste a password, sign-in goes through the Agent Computer handoff) closed the rest. 3. layer three: routines. five saved jobs, not seven. morning brief at 7, weekly review on friday, end-of-day handoff at 6, silent-inbox catcher, monthly audit. anything else lives in me, not in the bot. the day after i pasted the three files, the same task came back with three parked questions and one finished deliverable. no signups. no DMs. no deletes. your bot is not scary because it is smart. it is scary because it has your credentials and no one told it what stop looks like. paste the charter and the ten rules before the next task runs. the routines can wait until the first parked question comes back the way you wanted.

17,067 views

AGENT ARCHITECTURE ROUTES WORK. IT DOES NOT REMEMBER WORK. THAT GAP IS WHY YOUR LOOP KEEPS FIXING THE SAME BUG TWICE. these are two different engineering problems. every agent that silently drifts is missing one of them. architecture answers what runs. harness → loop → graph. it defines the tools, the retries, the branching routes, the approval gates. context ops answer what the run knows. write → read → compress → isolate. it defines what gets saved between attempts, pulled in on read, summarized on overflow, and split across sub-agents. for two months i believed a solid harness plus a verifier loop was enough. my coding agent kept re-discovering the same test failure across retries. the loop was working. it just had nowhere to write what it had already learned. here is the decision rule: if your agent forgets across restarts, add write and read. if it stalls on long tasks, add compress. if two sub-agents step on each other, add isolate. architecture without context ops is a well-routed system with amnesia.

AGENT ARCHITECTURE ROUTES WORK. IT DOES NOT REMEMBER WORK. THAT GAP IS WHY YOUR LOOP KEEPS FIXING THE SAME BUG TWICE. these are two different engineering problems. every agent that silently drifts is missing one of them. architecture answers what runs. harness → loop → graph. it defines the tools, the retries, the branching routes, the approval gates. context ops answer what the run knows. write → read → compress → isolate. it defines what gets saved between attempts, pulled in on read, summarized on overflow, and split across sub-agents. for two months i believed a solid harness plus a verifier loop was enough. my coding agent kept re-discovering the same test failure across retries. the loop was working. it just had nowhere to write what it had already learned. here is the decision rule: if your agent forgets across restarts, add write and read. if it stalls on long tasks, add compress. if two sub-agents step on each other, add isolate. architecture without context ops is a well-routed system with amnesia.

12,806 views

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

you're f*king wasting hours building every Grok Bot from scratch. the official Marketplace already has 85 specialists you can install. the official Marketplace combines Grok Bot Team releases with specialist Bots from named creators. it covers engineering, sales, marketing, design, recruiting, product, operations, and personal work. start with the job you actually need: › Researchy for sourced research, QA Bot for deployment checks › GTM Loop Closer for follow-ups, Clip Bot for captioned clips › Figma Bro for design audits, Competitor Watch for product changes press Add and Grok Bot copies the identity, description, skills, and routines into your account. it does not copy the creator's computer, logins, or conversation history. pick one role, inspect its boundaries, connect only the tools it needs, and test one safe task before enabling routines. stop treating every new agent like a blank-page contest. install the closest role, then make it yours.

you're f*king wasting hours building every Grok Bot from scratch. the official Marketplace already has 85 specialists you can install. the official Marketplace combines Grok Bot Team releases with specialist Bots from named creators. it covers engineering, sales, marketing, design, recruiting, product, operations, and personal work. start with the job you actually need: › Researchy for sourced research, QA Bot for deployment checks › GTM Loop Closer for follow-ups, Clip Bot for captioned clips › Figma Bro for design audits, Competitor Watch for product changes press Add and Grok Bot copies the identity, description, skills, and routines into your account. it does not copy the creator's computer, logins, or conversation history. pick one role, inspect its boundaries, connect only the tools it needs, and test one safe task before enabling routines. stop treating every new agent like a blank-page contest. install the closest role, then make it yours.

38,125 views • 10 days ago

I used GPT-6 Astra to score every Pons V2 token and wallet on Robinhood Chain straight from public logs. 741,643 indexed events on one token alone. 29 attributed curve participants, 93/100 score, HIGH CONFIDENCE. It's live and it's free: No private key. No signer. No transaction path. It reads the chain, it can't touch it. Another token resumed a killed indexing job from a 64-block overlap and still closed clean at 41,302 persisted events. Every point on every score traces back to a block and a log index. Most token scores hand you a number and ask you to trust it. Trust the model, trust the wallet they didn't disclose, trust that "verified" means what they say it means. You can't click through any of it. MEERKAT doesn't ask for that trust: → SCORE is GPT-6 Astra. Breaks a token 0-100 across five components: deployer exposure, creator tax, participant breadth, two-sided market, current holder breadth. Every component opens into the blocks behind it. → WALLET SCORE is GPT-6 Astra. Reads token scope, early discovery, two-sided activity, evidence depth. Confidence drops on its own when the local index is thin, it doesn't fake certainty. → FEE FLOW follows creator revenue to its current recipient. HOP OUT paid 1.177 ETH across 26 sweep events, split curve and pool, and the recipient doesn't match the deployer. Labeled ROUTED AT LAUNCH, not hidden. → RELATIONSHIPS maps attributed wallets around a token and keeps pool callers explicitly separate from real participants. → TIMELINE pages 50 events at a time, colors BUY green and SELL red, links every row to its transaction. → Two hard rules run under all of it: early entry is a curve buy within 30 blocks of launch, fast exit is a sell within 300 blocks of that wallet's first buy. The rule is shown, not just the label. FLYBRAIN ran 27x this week. Eight trusted wallets were in by 20:18, the alert fired at 20:18, and the pool didn't even open until 20:51. Most people saw the chart after it already moved. I caught 18x of that run, and it wasn't a lucky entry. It was because my own system was already reading the wallet flow before the pool opened, not after. By the time the crowd was refreshing charts, I already had a position. No signer means MEERKAT can't push a trade for me any more than it can for you. It reads chain state. It never holds a key. Landing page, terminal, APIs, indexer, database, all one local Node process. Open source, self-hostable, same code running the public terminal you can open right now. Next: global wallet discovery across the whole chain, then Pons market context sitting right next to the score. We're reading evidence anyone could open. Most people just never open the dossier.

I used GPT-6 Astra to score every Pons V2 token and wallet on Robinhood Chain straight from public logs. 741,643 indexed events on one token alone. 29 attributed curve participants, 93/100 score, HIGH CONFIDENCE. It's live and it's free: No private key. No signer. No transaction path. It reads the chain, it can't touch it. Another token resumed a killed indexing job from a 64-block overlap and still closed clean at 41,302 persisted events. Every point on every score traces back to a block and a log index. Most token scores hand you a number and ask you to trust it. Trust the model, trust the wallet they didn't disclose, trust that "verified" means what they say it means. You can't click through any of it. MEERKAT doesn't ask for that trust: → SCORE is GPT-6 Astra. Breaks a token 0-100 across five components: deployer exposure, creator tax, participant breadth, two-sided market, current holder breadth. Every component opens into the blocks behind it. → WALLET SCORE is GPT-6 Astra. Reads token scope, early discovery, two-sided activity, evidence depth. Confidence drops on its own when the local index is thin, it doesn't fake certainty. → FEE FLOW follows creator revenue to its current recipient. HOP OUT paid 1.177 ETH across 26 sweep events, split curve and pool, and the recipient doesn't match the deployer. Labeled ROUTED AT LAUNCH, not hidden. → RELATIONSHIPS maps attributed wallets around a token and keeps pool callers explicitly separate from real participants. → TIMELINE pages 50 events at a time, colors BUY green and SELL red, links every row to its transaction. → Two hard rules run under all of it: early entry is a curve buy within 30 blocks of launch, fast exit is a sell within 300 blocks of that wallet's first buy. The rule is shown, not just the label. FLYBRAIN ran 27x this week. Eight trusted wallets were in by 20:18, the alert fired at 20:18, and the pool didn't even open until 20:51. Most people saw the chart after it already moved. I caught 18x of that run, and it wasn't a lucky entry. It was because my own system was already reading the wallet flow before the pool opened, not after. By the time the crowd was refreshing charts, I already had a position. No signer means MEERKAT can't push a trade for me any more than it can for you. It reads chain state. It never holds a key. Landing page, terminal, APIs, indexer, database, all one local Node process. Open source, self-hostable, same code running the public terminal you can open right now. Next: global wallet discovery across the whole chain, then Pons market context sitting right next to the score. We're reading evidence anyone could open. Most people just never open the dossier.

74,238 views • 27 days ago

500 agents shouldn't be running all day. they should spin up for the exact window a task needs, finish it, and disappear until the next trigger. the bigger the swarm gets, the less any single agent matters. what matters is the system wiring them together. 100 agents can form up to 4,950 possible pairwise connections. 500 = 124,750. when a real signal lands, it spins up the whole workforce: 1 trigger → 100–500 Kimi agents → 5 live data feeds → up to 4,000 steps → verified artifact and that's before you add: sources → claims → memory → tool calls → contradictions → retries so the architecture has to fold all of that activity into one shared state, continuously. more agents just buys you more raw compute. the graph is the only thing standing between 124,750 possible connections and pure noise.

500 agents shouldn't be running all day. they should spin up for the exact window a task needs, finish it, and disappear until the next trigger. the bigger the swarm gets, the less any single agent matters. what matters is the system wiring them together. 100 agents can form up to 4,950 possible pairwise connections. 500 = 124,750. when a real signal lands, it spins up the whole workforce: 1 trigger → 100–500 Kimi agents → 5 live data feeds → up to 4,000 steps → verified artifact and that's before you add: sources → claims → memory → tool calls → contradictions → retries so the architecture has to fold all of that activity into one shared state, continuously. more agents just buys you more raw compute. the graph is the only thing standing between 124,750 possible connections and pure noise.

41,841 views • 29 days ago

FIVE LAYERS OF AGENT ENGINEERING, EACH ONE WRAPS THE ONE BELOW IT. IF YOU SKIP LAYER 2, YOUR LAYER 5 WILL LOOK BROKEN WHEN IT IS ACTUALLY JUST STANDING ON NOTHING. for weeks i debated harness vs loop vs graph like they were competing choices. then a stack diagram made the shape obvious. they are not choices. they are floors. 01 | prompt engineering. the message. unit of work: one input. inputs are role, instructions, examples, format. output is a single raw response. 02 | context engineering. the memory. unit of work: what stays in the window. a curator selects, compresses, and drops from query, docs, memory, prior turns, and tool outputs before the prompt runs. 03 | harness engineering. the machine. unit of work: the machine itself. gather (context + prompt) → LLM → tools or sub-agents → verifier → final response. the article calls this the operating environment. 04 | loop engineering. the system. unit of work: the run. goal + success criteria + max iterations + budget + completion check wrap around one harness pass. failed pass appends results to context and retries. 05 | graph engineering. the topology. unit of work: the graph run. goal + nodes + edges + state schema. graph routes to agent nodes, tool nodes, or human approval. a reviewer node with a different model and fresh context checks the final answer. the wrapping is the whole point. layer 5 assumes layer 4 works. layer 4 assumes layer 3 works. skip layer 2 and layer 3's verifier keeps failing without a clear reason. this is why swapping the model is a one-day project and swapping the stack is a quarter. the model is the commodity. the five layers around it are the engineering. full three-layer breakdown of the top of the stack (harness, loop, graph) in the post below.

FIVE LAYERS OF AGENT ENGINEERING, EACH ONE WRAPS THE ONE BELOW IT. IF YOU SKIP LAYER 2, YOUR LAYER 5 WILL LOOK BROKEN WHEN IT IS ACTUALLY JUST STANDING ON NOTHING. for weeks i debated harness vs loop vs graph like they were competing choices. then a stack diagram made the shape obvious. they are not choices. they are floors. 01 | prompt engineering. the message. unit of work: one input. inputs are role, instructions, examples, format. output is a single raw response. 02 | context engineering. the memory. unit of work: what stays in the window. a curator selects, compresses, and drops from query, docs, memory, prior turns, and tool outputs before the prompt runs. 03 | harness engineering. the machine. unit of work: the machine itself. gather (context + prompt) → LLM → tools or sub-agents → verifier → final response. the article calls this the operating environment. 04 | loop engineering. the system. unit of work: the run. goal + success criteria + max iterations + budget + completion check wrap around one harness pass. failed pass appends results to context and retries. 05 | graph engineering. the topology. unit of work: the graph run. goal + nodes + edges + state schema. graph routes to agent nodes, tool nodes, or human approval. a reviewer node with a different model and fresh context checks the final answer. the wrapping is the whole point. layer 5 assumes layer 4 works. layer 4 assumes layer 3 works. skip layer 2 and layer 3's verifier keeps failing without a clear reason. this is why swapping the model is a one-day project and swapping the stack is a quarter. the model is the commodity. the five layers around it are the engineering. full three-layer breakdown of the top of the stack (harness, loop, graph) in the post below.

31,198 views • 1 month ago

THE THREE MOST OVERLOOKED LINES IN A GROK BOT CHARTER ARE THE ONES THAT KEEP YOU IN CHARGE. approval is not a feature Chief levels into. it is a wall he cannot climb even if you ask him to. Rule 6, verbatim: "Approval belongs to the human, not to me. I never approve another bot's irreversible action. Only you do." i watch people hand Chief broader approval scopes every week thinking it is the upgrade path. it is not the upgrade path. it is the delegation chain forming behind you. the rule splits approval along three axes, and every one of them is the boundary: 1. own action vs another bot's action › Chief can sign off on delivering what he made himself. the moment he approves a sibling bot's action, the human is out of the loop. 2. reversible vs irreversible › anything Chief can undo in under one minute is his to run. anything he cannot is parked for you, without exception. 3. with Reports-to check vs through the chain › approval that reads a Reports-to line asks who the human is. approval that inherits from a previous bot's approval never asks. if Chief signed off on a publish, a payment or a send that you did not personally see, you are not running one bot with a boundary. you are running a chain that ends where nobody is checking. paste those three sentences into your charter, exactly as written. no synonyms, no softening.

THE THREE MOST OVERLOOKED LINES IN A GROK BOT CHARTER ARE THE ONES THAT KEEP YOU IN CHARGE. approval is not a feature Chief levels into. it is a wall he cannot climb even if you ask him to. Rule 6, verbatim: "Approval belongs to the human, not to me. I never approve another bot's irreversible action. Only you do." i watch people hand Chief broader approval scopes every week thinking it is the upgrade path. it is not the upgrade path. it is the delegation chain forming behind you. the rule splits approval along three axes, and every one of them is the boundary: 1. own action vs another bot's action › Chief can sign off on delivering what he made himself. the moment he approves a sibling bot's action, the human is out of the loop. 2. reversible vs irreversible › anything Chief can undo in under one minute is his to run. anything he cannot is parked for you, without exception. 3. with Reports-to check vs through the chain › approval that reads a Reports-to line asks who the human is. approval that inherits from a previous bot's approval never asks. if Chief signed off on a publish, a payment or a send that you did not personally see, you are not running one bot with a boundary. you are running a chain that ends where nobody is checking. paste those three sentences into your charter, exactly as written. no synonyms, no softening.

19,297 views • 1 month ago

YOU CLOSE THE TAB AND FORGET YOUR OBSIDIAN NOTES TOMORROW. HERE IS THE BREAK-OUT HARNESS. i spent four years clipping articles into obsidian, tagging notes manually, and building a giant graph visualization. six months later, i had 400 notes i never opened again. it was not a second brain. it was a folder full of digital guilt. then an engineer showed me why notes fail: if every chat starts from zero, your AI is just an expensive intern with amnesia. here is the exact 3-layer obsidian architecture that turns passive notes into an active memory layer for AI agents: 1. human memory layer (messy & raw) › keep your raw thoughts, personal doubts, unpolished meeting clips, and half-formed ideas completely human-centric. › preserves the authentic soul of your thinking without forcing you to act like a full-time librarian after work. 2. agent memory layer (structured & machine-readable) › maintain explicit markdown files for active project goals, decision logs, style constraints, and reusable prompts. › provides claude and local coding agents with instant, deterministic state so every chat continues where you left off. 3. deterministic index map (_index navigation) › place single-line map indexes inside _index/ folders so agents locate relevant files in a single prompt lookup. › prevents the LLM from searching thousands of raw notes and turning your vault into unstructured prompt slop. 90% of creators will keep building static note graveyards. the 1% who build a dual-layer AI memory vault will win. save this post before you close your current obsidian session and lose your context architecture.

YOU CLOSE THE TAB AND FORGET YOUR OBSIDIAN NOTES TOMORROW. HERE IS THE BREAK-OUT HARNESS. i spent four years clipping articles into obsidian, tagging notes manually, and building a giant graph visualization. six months later, i had 400 notes i never opened again. it was not a second brain. it was a folder full of digital guilt. then an engineer showed me why notes fail: if every chat starts from zero, your AI is just an expensive intern with amnesia. here is the exact 3-layer obsidian architecture that turns passive notes into an active memory layer for AI agents: 1. human memory layer (messy & raw) › keep your raw thoughts, personal doubts, unpolished meeting clips, and half-formed ideas completely human-centric. › preserves the authentic soul of your thinking without forcing you to act like a full-time librarian after work. 2. agent memory layer (structured & machine-readable) › maintain explicit markdown files for active project goals, decision logs, style constraints, and reusable prompts. › provides claude and local coding agents with instant, deterministic state so every chat continues where you left off. 3. deterministic index map (_index navigation) › place single-line map indexes inside _index/ folders so agents locate relevant files in a single prompt lookup. › prevents the LLM from searching thousands of raw notes and turning your vault into unstructured prompt slop. 90% of creators will keep building static note graveyards. the 1% who build a dual-layer AI memory vault will win. save this post before you close your current obsidian session and lose your context architecture.

24,898 views • 2 months ago

THIS GUY BUILT AN AUTONOMOUS AI AGENT OUT OF CLAUDE CODE + OBSIDIAN and this is way more interesting than another “use AI to take notes” demo the trick is simple: Obsidian is not the writing app here. it becomes the agent’s memory, task board, and context folder. Claude Code is not just answering prompts. it reads the vault, edits files, follows instructions, and keeps moving through the work like a junior operator with a filesystem. the reusable setup looks like this: 1. create an Obsidian vault for one project 2. keep goals, rules, tasks, decisions, and references as markdown files 3. point Claude Code at the folder 4. give it a clear operating loop: read context → choose next task → execute → write back what changed 5. use the notes as persistent memory instead of re-explaining the project every chat that’s the part people miss. the “agent” is not magic. it’s the boring combination of: - local files - explicit rules - task state - write access - a model that can run through the repo/vault Obsidian makes the memory human-readable. Claude Code makes the memory executable. that combo is why the video worked: it turns a notes app into an operating surface for actual work. best use cases: - content systems - research vaults - coding projects - client ops docs - personal knowledge bases that need actions, not just storage the caveat: if your vault is messy, your agent becomes messy too. folders, naming, “done” criteria, and forbidden actions matter more than the prompt. but once the structure is clean, this is one of the easiest ways to build an agent that remembers what happened yesterday without paying for a full custom app.

THIS GUY BUILT AN AUTONOMOUS AI AGENT OUT OF CLAUDE CODE + OBSIDIAN and this is way more interesting than another “use AI to take notes” demo the trick is simple: Obsidian is not the writing app here. it becomes the agent’s memory, task board, and context folder. Claude Code is not just answering prompts. it reads the vault, edits files, follows instructions, and keeps moving through the work like a junior operator with a filesystem. the reusable setup looks like this: 1. create an Obsidian vault for one project 2. keep goals, rules, tasks, decisions, and references as markdown files 3. point Claude Code at the folder 4. give it a clear operating loop: read context → choose next task → execute → write back what changed 5. use the notes as persistent memory instead of re-explaining the project every chat that’s the part people miss. the “agent” is not magic. it’s the boring combination of: - local files - explicit rules - task state - write access - a model that can run through the repo/vault Obsidian makes the memory human-readable. Claude Code makes the memory executable. that combo is why the video worked: it turns a notes app into an operating surface for actual work. best use cases: - content systems - research vaults - coding projects - client ops docs - personal knowledge bases that need actions, not just storage the caveat: if your vault is messy, your agent becomes messy too. folders, naming, “done” criteria, and forbidden actions matter more than the prompt. but once the structure is clean, this is one of the easiest ways to build an agent that remembers what happened yesterday without paying for a full custom app.

30,403 views • 3 months ago

STOP TREATING YOUR OBSIDIAN VAULT AS AI MEMORY. WITHOUT ONE ROUTING FILE, CLAUDE IS BURNING TOKENS ON NOISE YOU FORGOT YOU SAVED. a folder full of markdown does not make Claude smart. what makes it smart is what your agent can find without rereading the whole vault. storage: write → tag → graph → agent scans everything the model touches the entire vault, hallucinates half of it, forgets by the next session. memory: write → index → route → agent opens two files the model reads the index first, follows one path, answers with a link back to the note. i have rebuilt my Obsidian vault three times this year. the first two collapsed inside a month because Claude kept guessing which folder to open. the third one holds because the first thing i ship now is the routing file. Markdown, links, and the graph view are not what changed. what changed was a one-line description per folder, written for the agent, kept current, loaded before any retrieval call. if your agent still starts every session by scanning your whole vault, you are running storage. write the routing file today.

STOP TREATING YOUR OBSIDIAN VAULT AS AI MEMORY. WITHOUT ONE ROUTING FILE, CLAUDE IS BURNING TOKENS ON NOISE YOU FORGOT YOU SAVED. a folder full of markdown does not make Claude smart. what makes it smart is what your agent can find without rereading the whole vault. storage: write → tag → graph → agent scans everything the model touches the entire vault, hallucinates half of it, forgets by the next session. memory: write → index → route → agent opens two files the model reads the index first, follows one path, answers with a link back to the note. i have rebuilt my Obsidian vault three times this year. the first two collapsed inside a month because Claude kept guessing which folder to open. the third one holds because the first thing i ship now is the routing file. Markdown, links, and the graph view are not what changed. what changed was a one-line description per folder, written for the agent, kept current, loaded before any retrieval call. if your agent still starts every session by scanning your whole vault, you are running storage. write the routing file today.

17,240 views • 1 month ago

THIS GUY TURNED 5 PROMPTING TIPS INTO A FREE AI CEO CHALLENGE The useful part is treating every prompt like you are briefing a very fast employee who has zero context. Most people open ChatGPT and type a wish. Pros give it a job. Try this instead: 1. Give it a role Not “help me with marketing.” Say: “Act as a B2B SaaS growth operator reviewing a landing page.” 2. Give it the real context Who is the customer? What are they buying? What have you already tried? What does success look like? 3. Give it constraints Length, tone, format, audience, banned words, examples to copy, examples to avoid. A vague prompt gets a vague answer. A constrained prompt gets something you can edit. 4. Ask for options before answers “Give me 5 angles, rank them, then explain the tradeoff.” This turns AI from an autocomplete box into a thinking partner. 5. Force it to show assumptions Before it writes, ask: “What are you assuming, what info is missing, and what would change your answer?” That one line saves a lot of fake confidence. Dan Martell’s video works because the promise is simple: 5 prompting habits that make AI feel less random. The reusable move is even simpler: Stop prompting for outputs. Start prompting for decisions. Bad: “Write me a post.” Better: “Here is the source, here is the reader, here is the angle, give me 3 hooks, choose the strongest, then draft in this style.” That is the difference between getting content-shaped noise and getting work you can actually ship. Caveat: prompts do not fix weak taste, bad data, or unclear strategy. But they do expose those problems faster. If your AI answers are generic, your prompt probably has no job, no context, no constraints, and no standard for what “good” means.

THIS GUY TURNED 5 PROMPTING TIPS INTO A FREE AI CEO CHALLENGE The useful part is treating every prompt like you are briefing a very fast employee who has zero context. Most people open ChatGPT and type a wish. Pros give it a job. Try this instead: 1. Give it a role Not “help me with marketing.” Say: “Act as a B2B SaaS growth operator reviewing a landing page.” 2. Give it the real context Who is the customer? What are they buying? What have you already tried? What does success look like? 3. Give it constraints Length, tone, format, audience, banned words, examples to copy, examples to avoid. A vague prompt gets a vague answer. A constrained prompt gets something you can edit. 4. Ask for options before answers “Give me 5 angles, rank them, then explain the tradeoff.” This turns AI from an autocomplete box into a thinking partner. 5. Force it to show assumptions Before it writes, ask: “What are you assuming, what info is missing, and what would change your answer?” That one line saves a lot of fake confidence. Dan Martell’s video works because the promise is simple: 5 prompting habits that make AI feel less random. The reusable move is even simpler: Stop prompting for outputs. Start prompting for decisions. Bad: “Write me a post.” Better: “Here is the source, here is the reader, here is the angle, give me 3 hooks, choose the strongest, then draft in this style.” That is the difference between getting content-shaped noise and getting work you can actually ship. Caveat: prompts do not fix weak taste, bad data, or unclear strategy. But they do expose those problems faster. If your AI answers are generic, your prompt probably has no job, no context, no constraints, and no standard for what “good” means.

25,573 views • 3 months ago

AgentRouter is handing out $125 in FREE API credits for OPUS 4.8 > No card. > GitHub sign-in. > One API key for Opus 4.8 / Opus 4.7 / Sonnet 4.6 / GLM 5.1 The useful part is not “another model gateway.” It is that you can point coding tools and agents at one OpenAI-compatible base URL instead of juggling separate keys for Claude and GLM, and others. What you get: • $125 signup balance • one API key • OpenAI-compatible endpoints • chat, responses, messages, embeddings, images, audio, rerank endpoints listed • claimed support for Cursor, Claude Code, Hermes, and other agent tools The flow is simple: 1. go to AgentRouter (link in comment) 2. sign in or create an account 3. generate an API key in the console 4. copy the base URL 5. paste it into Cursor / Claude Code / your agent runtime 6. run a small test before moving anything serious Caveat: AgentRouter’s own site currently has a notice saying Claude-series service was recently hit by stability issues and large-scale Claude access was suspended. So don’t treat this as “guaranteed free Claude forever.” Treat it as $125 of free routing credits to test which models are live, how the latency feels, and whether the gateway is stable enough for your workflow. Still a very solid save if you build with agents. Most people burn paid API credits just testing configs. This gives you room to test first, then decide if it belongs in your stack.

AgentRouter is handing out $125 in FREE API credits for OPUS 4.8 > No card. > GitHub sign-in. > One API key for Opus 4.8 / Opus 4.7 / Sonnet 4.6 / GLM 5.1 The useful part is not “another model gateway.” It is that you can point coding tools and agents at one OpenAI-compatible base URL instead of juggling separate keys for Claude and GLM, and others. What you get: • $125 signup balance • one API key • OpenAI-compatible endpoints • chat, responses, messages, embeddings, images, audio, rerank endpoints listed • claimed support for Cursor, Claude Code, Hermes, and other agent tools The flow is simple: 1. go to AgentRouter (link in comment) 2. sign in or create an account 3. generate an API key in the console 4. copy the base URL 5. paste it into Cursor / Claude Code / your agent runtime 6. run a small test before moving anything serious Caveat: AgentRouter’s own site currently has a notice saying Claude-series service was recently hit by stability issues and large-scale Claude access was suspended. So don’t treat this as “guaranteed free Claude forever.” Treat it as $125 of free routing credits to test which models are live, how the latency feels, and whether the gateway is stable enough for your workflow. Still a very solid save if you build with agents. Most people burn paid API credits just testing configs. This gives you room to test first, then decide if it belongs in your stack.

25,316 views • 3 months ago

THIS GUY BUILT A BUSINESS SECOND BRAIN WITH CLAUDE CODE + OBSIDIAN IN 3 STEPS Most teams do not need another Notion workspace. They need a place where the company can remember how it works. The video shows a simple setup: 1. Create one empty folder called second brain. 2. Split it into 3 buckets: raw new knowledge wiki 3. Let Claude Code turn messy company material into connected notes. The useful part is the separation. Raw is where your existing stuff goes: SOPs, sales docs, process notes, client delivery checklists, old Loom summaries, onboarding docs. New knowledge is where fresh outside material lands: articles, clips, tactics, examples, market notes. Wiki is the cleaned version: concepts, roles, processes, SOPs, gaps, reusable decisions. That is where Claude Code becomes more useful than a normal chat window. Instead of asking it to remember random context forever, you give it a folder it can read, edit, and reorganize. Then Obsidian becomes the human interface. The Obsidian Web Clipper captures useful pages into the vault. Claude Code ingests them. The wiki gets updated. Then you can ask questions like: “Does my current workflow actually hold up?” That is the real point. Not “AI notes.” A business memory system that can compare what you do today against new information tomorrow. The caveat: this is not magic company intelligence. If your raw docs are vague, outdated, or full of tribal knowledge, Claude will organize weak inputs into cleaner weak outputs. You still need naming rules, review habits, and someone responsible for deleting junk. But the setup is refreshingly practical. Folder first. Clipper second. Claude Code as the maintainer. No giant knowledge base migration. No complex setup. Just a local vault that can slowly turn scattered business memory into something searchable, editable, and actually reusable.

THIS GUY BUILT A BUSINESS SECOND BRAIN WITH CLAUDE CODE + OBSIDIAN IN 3 STEPS Most teams do not need another Notion workspace. They need a place where the company can remember how it works. The video shows a simple setup: 1. Create one empty folder called second brain. 2. Split it into 3 buckets: raw new knowledge wiki 3. Let Claude Code turn messy company material into connected notes. The useful part is the separation. Raw is where your existing stuff goes: SOPs, sales docs, process notes, client delivery checklists, old Loom summaries, onboarding docs. New knowledge is where fresh outside material lands: articles, clips, tactics, examples, market notes. Wiki is the cleaned version: concepts, roles, processes, SOPs, gaps, reusable decisions. That is where Claude Code becomes more useful than a normal chat window. Instead of asking it to remember random context forever, you give it a folder it can read, edit, and reorganize. Then Obsidian becomes the human interface. The Obsidian Web Clipper captures useful pages into the vault. Claude Code ingests them. The wiki gets updated. Then you can ask questions like: “Does my current workflow actually hold up?” That is the real point. Not “AI notes.” A business memory system that can compare what you do today against new information tomorrow. The caveat: this is not magic company intelligence. If your raw docs are vague, outdated, or full of tribal knowledge, Claude will organize weak inputs into cleaner weak outputs. You still need naming rules, review habits, and someone responsible for deleting junk. But the setup is refreshingly practical. Folder first. Clipper second. Claude Code as the maintainer. No giant knowledge base migration. No complex setup. Just a local vault that can slowly turn scattered business memory into something searchable, editable, and actually reusable.

16,698 views • 3 months ago

THIS GUY BUILT A CLAUDE CODE X OBSIDIAN MAP OF HIS ENTIRE CONTENT SYSTEM This is the useful version of “AI second brain.” Not dumping more notes. Not asking Claude for a prettier folder system. Not making a canvas because it looks smart. In the video, he points Claude Code at his Obsidian setup and shows a visual map of the actual content pipeline: Analysis Ideation Prep Scripting Prep Performance The interesting part is the shape. Each stage is connected to the next one. Some boxes show sub-processes. One section shows a router detecting content type and routing a short into the next step. There are references attached to the flow. That is a real payoff: you stop treating your vault like storage and start treating it like an inspectable machine. The move is simple: 1. Put the real workflow in markdown 2. Let Claude Code inspect the vault 3. Ask it to find stages, dependencies, and missing links 4. Turn the output into an Obsidian map 5. Use the map to see what is manual, duplicated, or broken This works because Claude Code is not just summarizing a note. It can read across prompts, docs, scripts, references, and messy process files, then expose the structure you stopped seeing. That is why the demo hits. The video is not really about “better note-taking.” It is about making your private operating system visible enough to debug. Caveat: a beautiful graph does not mean you have a working system. If the notes are vague, the map will be vague. If the process is fake, Claude will draw a fake process very cleanly. If nothing feeds back into performance, the canvas is just decoration. But if the vault already contains real work, Claude Code x Obsidian becomes a powerful audit tool. Your notes stop being a pile. They become a map of what you actually do.

THIS GUY BUILT A CLAUDE CODE X OBSIDIAN MAP OF HIS ENTIRE CONTENT SYSTEM This is the useful version of “AI second brain.” Not dumping more notes. Not asking Claude for a prettier folder system. Not making a canvas because it looks smart. In the video, he points Claude Code at his Obsidian setup and shows a visual map of the actual content pipeline: Analysis Ideation Prep Scripting Prep Performance The interesting part is the shape. Each stage is connected to the next one. Some boxes show sub-processes. One section shows a router detecting content type and routing a short into the next step. There are references attached to the flow. That is a real payoff: you stop treating your vault like storage and start treating it like an inspectable machine. The move is simple: 1. Put the real workflow in markdown 2. Let Claude Code inspect the vault 3. Ask it to find stages, dependencies, and missing links 4. Turn the output into an Obsidian map 5. Use the map to see what is manual, duplicated, or broken This works because Claude Code is not just summarizing a note. It can read across prompts, docs, scripts, references, and messy process files, then expose the structure you stopped seeing. That is why the demo hits. The video is not really about “better note-taking.” It is about making your private operating system visible enough to debug. Caveat: a beautiful graph does not mean you have a working system. If the notes are vague, the map will be vague. If the process is fake, Claude will draw a fake process very cleanly. If nothing feeds back into performance, the canvas is just decoration. But if the vault already contains real work, Claude Code x Obsidian becomes a powerful audit tool. Your notes stop being a pile. They become a map of what you actually do.

15,941 views • 3 months ago

THIS GUY IS USING GTA 6 TO MAKE $10,000 A MONTH ON THE GAME'S LAUNCH. Not after the game launches. Before it launches. That is the whole play. The creator’s bet is simple: GTA 6 is already a search engine before anyone can play it. Trailers. Scenes. Map theories. Car theories. Release rumors. Money glitch theories. Tiny details people want explained. In the video, he says GTA 6 is projected to make $1B on day one and $7.6B in its first two months. His move is to stand in front of that demand early. The workflow he shows: 1. Take a GTA 6 trailer, scene, rumor, or news angle 2. Ask ChatGPT for a scene-by-scene breakdown 3. Turn that breakdown into YouTube Shorts ideas 4. Paste it into Viewmaxx io for video generation, scriptwriting, and AI voiceover 5. Add captions so the clip still works when people are scrolling fast, muted, or half watching 6. Repeat across every micro-question people search before launch The money claim is the bait. He points to YouTube Shorts paying roughly $2K - $5K per million views, but that is not a guaranteed income plan. RPM depends on niche, country, retention, ad demand, monetization status, and whether the content is original enough to pass platform rules. AI voice + captions + GTA clips is not a business by itself. The useful part is the arbitrage: find a giant upcoming event, break it into small questions, use AI to ship faster than normal editors, and attach each post to demand that already exists. GTA 6 is the example. The reusable model is event-driven Shorts before the event peaks.

THIS GUY IS USING GTA 6 TO MAKE $10,000 A MONTH ON THE GAME'S LAUNCH. Not after the game launches. Before it launches. That is the whole play. The creator’s bet is simple: GTA 6 is already a search engine before anyone can play it. Trailers. Scenes. Map theories. Car theories. Release rumors. Money glitch theories. Tiny details people want explained. In the video, he says GTA 6 is projected to make $1B on day one and $7.6B in its first two months. His move is to stand in front of that demand early. The workflow he shows: 1. Take a GTA 6 trailer, scene, rumor, or news angle 2. Ask ChatGPT for a scene-by-scene breakdown 3. Turn that breakdown into YouTube Shorts ideas 4. Paste it into Viewmaxx io for video generation, scriptwriting, and AI voiceover 5. Add captions so the clip still works when people are scrolling fast, muted, or half watching 6. Repeat across every micro-question people search before launch The money claim is the bait. He points to YouTube Shorts paying roughly $2K - $5K per million views, but that is not a guaranteed income plan. RPM depends on niche, country, retention, ad demand, monetization status, and whether the content is original enough to pass platform rules. AI voice + captions + GTA clips is not a business by itself. The useful part is the arbitrage: find a giant upcoming event, break it into small questions, use AI to ship faster than normal editors, and attach each post to demand that already exists. GTA 6 is the example. The reusable model is event-driven Shorts before the event peaks.

11,134 views • 3 months ago

THIS GUY BUILT A SNACK MACHINE BUSINESS AROUND THE MOST BORING PRODUCT ON EARTH just a simple machine, a location, snacks, and repeatable distribution. that’s why the video works. most people watch it and think: “nice side hustle.” builders should watch it and think: “this is a better product lesson than 90% of startup advice.” because the mechanism is stupidly clear: 1. find a tiny repeatable demand 2. put the offer where the demand already exists 3. remove the human from the transaction 4. restock based on what actually sells 5. repeat only after the unit economics survive reality that last part is the whole game. AI builders keep trying to automate the shiny part first. landing page, prompt chain, avatar video, dashboard, launch post. but the snack machine business starts with something AI people skip: boring proof. can one location pay back? which products move? how often does it need restocking? what breaks? what gets stolen? what happens when nobody cares? this is also why the better AI-UGC businesses are interesting right now. not because “AI makes videos.” because the real workflow is distribution + testing + iteration: multiple accounts, many creatives, fast feedback, then scaling the winners. same idea, different machine. physical vending machine: location → product → purchase → restock data AI content machine: account → creative → attention → revenue data the caveat is obvious: a video can make the machine look cleaner than the business. permits, placement deals, maintenance, theft, dead inventory, and bad locations can kill the margin. but the reusable lesson is still strong: build the smallest cashflow machine you can observe directly. then automate the parts that are already working. not the other way around.

11,646 views • 3 months ago

No more content to load