whemo (@whemohere) header
whemo (@whemohere) profile picture

content creation enjoyer, the road to influence and a bit of shitposting / @npaywall co-founder

Osaka
Joined April 2021
MY GROK BOTS LOG EVERYTHING THEY DO IN ONE OBSIDIAN VAULT. EVERY NIGHT A 13TH BOT READS IT AND CHECKS THEIR WORK. IT CAUGHT 29 MISTAKES BEFORE I SAW THEM. 12 bots, one work log. every action goes in as a line: who, what, for whom bot #13 does one job. at 2am it reads the whole day and checks each line against what's in the vault: → contracts and prices → my rules ("no flights before 8am") → time zones → attachments in sent emails → company policies one night: → invoice for $2,400, contract says $2,040. fixed before it went out → 6:10am flight to denver. rebooked to 9:05, same price → call at 10am pt with someone in london. moved to 8am → proposal sent without the pdf. resent 29 caught this month. 2 reached me. before bot #13 i found 31 a month myself, usually after a client did the prompt for the checker: "every night, read today's log in the vault. check each entry against my notes, contracts and rules. fix what's wrong, and tell me what you fixed in one message." the setup is in the article below ↓
8
23
266
MY 12 GROK BOTS KEPT FORGETTING WHAT THEY ALREADY DID. NOW THEY SHARE ONE OBSIDIAN VAULT AS MEMORY. I USED TO HIT MY LIMIT BY WEDNESDAY. NOW IT LASTS ALL WEEK none of them knew what the others had done inbox read the same 214 emails twice. sales and leads researched the same person. calendar asked me for my meeting prefs 4 times last week: 340 repeated tasks. limit gone on wednesday at 2pm the fix was one rule and one obsidian vault: 1. before any task, read the vault 2. after it, write what you did and what you learned 3. one folder per bot, one shared folder for people and decisions 4. a note per person, deal and trip, so any bot can pick it up this week: 1,460 notes, almost no repeats, 74% of the limit used by sunday the rule i gave all 12: "before you start any task, search the shared vault for it. if it's done, skip it and tell me. when you finish, write a note: what you did, for whom, and what you learned." the setup is in the article below ↓
8
40
574
I TAUGHT GROK BOT MY CHESS METHOD FOR SECURITY. BEFORE ANY CLICK, LOGIN OR INSTALL, IT PLAYS OUT THE ATTACKER'S NEXT 3 MOVES. THIS WEEK IT BLOCKED 9 THINGS THAT LOOKED SAFE. most security checks ask one question: is this link bad. a chess player asks what happens 3 moves later the method i gave my bot: 1. candidate move: what am i about to allow 2. their reply: what an attacker can do with it 3. the next 2 moves after that 4. if the line ends at my email, bank or passwords, don't play it 5. save every blocked line as an opening, so it spots it instantly next time one game this week: → a browser extension i've used for 2 years asked for a new permission after an update → move 1: read every site i open → move 2: take my logged-in sessions → move 3: open my bank without a password → blocked and removed. the extension had a new owner since last week 9 lines like that this week. every one looked normal the prompt: "before i click, log in, install or grant anything, play it like chess: name what i'm allowing, the attacker's best reply and the next 2 moves. if the line reaches my email, bank or passwords, stop me and show me the line. save every blocked line." the setup is in the article below ↓
11
41
668
I TAUGHT GROK BOT TO THINK LIKE A CHESS PLAYER. MY COLD EMAILS WENT FROM 2% REPLIES TO 19%. a cold email is a game. you move, they reply. most people only plan the first move so before every email my bot plays it out like chess: 1. candidate moves: 3 openings, not 1 2. for each, the prospect's most likely reply 3. my counter to that reply, written before i send anything 4. play the line that ends in a yes 5. save the winning line as an opening for the next prospect one game, a dental clinic owner: → opened with "your booking button breaks on iphone". she said "send more info" → 3 lines and a screenshot of her own site. she said "we already have a web guy" → fix it free, her guy keeps the site. call booked for thursday 200 local businesses, 38 replied the prompt: "before every cold email, write 3 openings, predict the prospect's reply to each, prepare my counter, and send the line most likely to get a yes. save winning lines as skills." the setup is in the article below ↓
9
35
549
I MADE GROK BOT FAR SMARTER BY TEACHING IT TO THINK LIKE A CHESS PLAYER. FIRST TEST: A $11,400 INSURANCE DENIAL. OVERTURNED most people ask AI for one answer. a chess player never plays the first move that looks good so i gave my bot the way they think: 1. candidate moves: write 3 options, not 1 2. calculation: for each, predict the other side's best reply 3. counter: have the next move ready before sending anything 4. evaluation: pick the line that ends best for me, in dollars 5. opening book: save every winning line as a skill first game: my insurer denied a surgery. "not medically necessary", "no prior auth", "policy exclusion" it saw every excuse coming. 3 moves, each one already answered. appeal approved, i owe $0 the prompt: "before every message in this dispute, list 3 candidate moves, predict their most likely reply to each, prepare my counter, and send the line with the best outcome. save winning lines as skills." the setup is in the article below ↓
8
31
506
MY DOWNLOADS FOLDER HAD 23,418 FILES. GROK BOT SORTED EVERY ONE IN 6 MINUTES AND FREED 140 GB. 7 years of "i'll sort it later". my mac had 41 GB left it went file by file on my mac, with my approval: → 6,104 screenshots into folders by year and month → 4,880 duplicates, byte for byte the same as the original. 38 GB → 1,980 zips i'd already unpacked. 12 GB → 211 videos already backed up in icloud photos. 29 GB → 1,206 invoices and receipts renamed by vendor and month then it asked before the big one: 312 .dmg installers for apps i already have. 61 GB. i said yes nothing i still use got deleted. 181 GB free now the rules stay saved, so new downloads get sorted every sunday the prompt: "sort my downloads folder into folders by type and date. find duplicates, installers and anything already backed up. show me what you want to delete before you delete it." the setup is in the article below ↓
8
41
578
I NEEDED A CAR. GROK BOT EMAILED 46 DEALERS AT ONCE. THE CHEAPEST ONE SOLD IT TO ME FOR $4,300 LESS THAN ADVERTISED. same car, same trim, one email to every dealer within 50 miles. then the part i'd never do myself. every time a dealer came back lower, grok bot sent that number to the other 45. how the price moved: $31,480 advertised $30,400 $29,500 $28,400 $27,900 $27,180 out the door 41 of 46 answered. a few said "that's below our invoice". the next email went out anyway nothing got signed until i said yes. saturday i drove it home. the setup is in the article below ↓
9
54
719
MY COMPANY IS 9 PEOPLE AND 31 GROK BOTS. HERE'S OUR ORG CHART every person runs a small team of bots now support (dana): inbox, refunds, reviews, live chat sales (marcus): leads, follow-ups, demos, crm finance (emma): invoices, expenses, payroll, taxes ops (leo): orders, shipping, vendors, stock marketing (ines): ads, social, newsletter, seo product (ryan): feedback, roadmap, qa, release notes hiring (kate): sourcing, screening, scheduling, onboarding engineering (omar): code review, on-call, docs bots hand work to each other. refunds tells invoices. leads tells demos. money, contracts and anything public waits for the human on that branch. 1,200+ tasks a day. the 9 of us do the parts that need a yes. how we set it up, article below ↓
11
1
43
862
GROK BOT SAT IN 23 OF MY MEETINGS THIS WEEK. I WENT TO 2 grok bot 1.0 can join zoom, google meet and teams calls on its own it listens, writes the notes, remembers what was decided, and hands out the action items my week: → monday standup, 4 tasks to reminders → 3 sales calls, follow-up emails drafted, waiting for my ok → 2 vendor calls, it caught a price that didn't match the contract → a 90 minute planning call, summed up in 6 lines → 9 meetings that could've been an email. it told me which ones. i joined the 2 where someone needed a yes from me everything else is in my calendar with notes attached the setup takes 10 minutes. article below ↓
11
38
686
GROK BOT 1.0 TOLD ME 4 OF MY 9 BOTS WERE BURNING MONEY ON NOTHING. IT SHUT THEM DOWN ITSELF. MY BILL DROPPED 40% in 1.0, bots can suggest changes to how they work. mine did it on day one its report: → inbox bot: checking mail every 10 minutes, nothing to reply to most of the day → news bot: writing a digest nobody had opened in 3 weeks → a research bot from one project in august, still running → a price tracker for something i already bought 4 bots. 41% of the bill. nothing useful out of any of them it asked once: "shut these down and move the useful parts into the other 5?" i said yes same work, same answers. the bill went from $310 to $186 ask yours: "go through all my bots. which ones cost money and produce nothing? what would you merge or shut down?" how to set yours up is in the article below
11
1
34
608
THE "CAVEMAN" TRICK CUT MY GROK BOT'S TOKENS 68%. SAME ANSWERS. IT JUST STOPPED TALKING TO ITSELF IN FULL SENTENCES most of what an agent spends isn't the answer. it's the notes it writes to itself before the answer before: "okay, so i need to look through the inbox. let me start by reading each new email carefully. the first one appears to be an invoice…" 1,840 tokens caveman: "inbox. 2 new. invoice → finance. sam asks time → reply 3pm. done." 410 tokens the rule i saved to memory, one line: "think in short words. no full sentences to yourself. replies to me stay normal." across 6 bots: → 3.4M tokens a month became 1.1M → 0 answers changed → research and inbox dropped the most the setup is in the article below ↓
8
1
36
624
GROK BOT AUDITED ITS OWN TOKEN SPEND. 71% WAS WASTE. IT CUT ITSELF FROM 4.1M TO 1.2M A MONTH i asked it one thing: "audit your own token spend." it traced every token from october, bot by bot: → 1.1M re-reading the same files. research re-read a 300-page doc 212 times → 0.8M on runs that found nothing. inbox checked 96 times a day → 0.5M on retries after errors → 0.3M on reports nobody opened → 0.2M dragging old chat into new tasks then it changed itself: 1. reads its own saved summary instead of the source 2. runs only when something changes 3. fixes the cause of an error instead of retrying blind 4. stopped the 3 reports nobody reads 5. starts each task clean useful work stayed at 1.2M. nothing i use stopped working ask yours: "trace where your tokens went last month, show me what was waste, and fix it in yourself." the setup is in the article below ↓
9
49
966
MY GROK BOT NOW DOES THE SAME WORK FOR 1/6 OF THE TOKENS. IT AUDITED ITS OWN ROUTINES, FOUND 3 FIRING EVERY 15 MINUTES FOR NOTHING AND REWROTE THEM. MY PLAN LASTS THE WHOLE MONTH NOW my plan used to run out by the 9th. so i asked the bot to read its own usage log where the tokens went: 1. 3 routines every 15 minutes. 288 runs a day, 271 ended with "nothing new" 2. a group chat where every message woke all 6 bots 3. a research bot re-reading the same 300-page doc on every run 4. 3 jobs done by clicking through the browser what it changed: 1. routines fire on a new email, or twice a day, and stay silent when nothing changed 2. one chief bot reads the chat and wakes only the bot it needs 3. the doc summary is saved once and read every time 4. the same 3 jobs now go through connectors nothing i use stopped working. the plan now ends the month with 40% left the line that did most of it, now in every bot's description: "if a routine finds nothing new, say nothing and stop." the setup is in the article below ↓
11
1
47
2,127
GROK BOT FOUND EVERY SITE THAT STILL HAS MY CARD SAVED. 64 STORES, 9 I CAN'T EVEN REMEMBER, ONE CHARGED ME LAST WEEK. IT REMOVED THE CARD FROM 61 i've paid online with the same card since 2019. i asked my bot where it's still saved it went through 7 years of receipts in my mail, my saved logins and my bank statements, then signed into each store from its vault. it never sees the passwords, it just fills them in what it found: > 64 stores and services with my card on file > 9 i don't remember signing up for > a vpn trial from 2022 that charged me $11.99 last week. and every month before that > 14 stores that had a data breach while my card was saved there > 6 accounts that kept my card without asking, after a single checkout 58 removed online in 40 minutes. 3 stores have no button for it, so it called them, sat on hold and had support delete the card. i watched the live transcript. the 3 i actually use kept it the vpn refunded 3 months the prompt: "Find every store and service that has my card saved. Use my email receipts, saved logins and bank statements. For each one, show the last charge and whether it had a breach. Remove the card everywhere I haven't bought anything in 6 months, and ask me before any phone call." the full setup is in the article ↓
13
1
42
2,379
GROK BOT SEARCHED EVERY FILE ON MY MAC FOR WHAT A THIEF WOULD WANT. 38 SCREENSHOTS OF PASSWORDS, MY PASSPORT IN DOWNLOADS 4 TIMES, TAX RETURNS IN A SHARED FOLDER. IT LOCKED ALL OF IT IN AN ENCRYPTED VAULT if someone took my laptop tomorrow, what would they find? grok bot went through 184,000 files to answer that. what it found: – 38 screenshots with passwords on screen. bank login, email, router, even crypto seed words – 4 copies of my passport in downloads, from 4 different bookings – 3 tax returns in a family shared folder, one with a link anyone could open – 12 bank statements with account and routing numbers – 5 photos of my cards, one with the cvc visible 68 files in total. it moved all of them into an encrypted vault that only opens with touch id, deleted the copies, turned off the shared link and removed the screenshots from photo sync. the prompt: "Search every file on my Mac like someone who just stole it. Find screenshots with passwords, ID and passport scans, tax returns, bank statements and card photos, and tell me where each one is and who can see it. Move them into an encrypted vault and delete the copies." the full setup is in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's. One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste. 16-min read. Save it ↓
10
2
22
961
GROK BOT READ 4 YEARS OF MY BANK STATEMENTS AND BUILT ONE PAGE OF EVERYTHING I PAY FOR. 23 SUBSCRIPTIONS, 9 I FORGOT ABOUT, AND ONE BILLED ME TWICE SINCE 2022. IT CANCELED 9 AND GOT $412 BACK grok bot has pages now: interactive docs it builds from your data. i connected my bank and asked for one page what was on it: – 23 subscriptions, $322 a month – a language app i last opened in january 2023 – a gym app, a news site, a vpn from one trip, a pdf editor i used once – a notes app charging me twice every month since 2022 – $15,070 spent on subscriptions since 2022 it canceled the 9 i forgot, $116 a month. then it wrote to the notes app about the double charge, and they refunded $412 $1,401 a year back. the page stays live, so i'll see the next one sneak in the prompt: "Read my bank statements for the last 4 years and build me one Page of every recurring payment: what it is, how much, when I last used it. Flag anything I haven't used in 6 months or that charges me twice. Cancel the ones I approve and ask for refunds on double charges." the full setup is in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's. One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste. 16-min read. Save it ↓
14
1
37
1,108
GROK BOT CHECKED WHAT EVERY APP ON MY MAC IS ALLOWED TO DO. 14 CAN RECORD MY SCREEN, 9 CAN READ EVERY FILE, AND A FREE CLIPBOARD APP SAW EVERY PASSWORD I COPIED FOR 2 YEARS. IT TOOK IT ALL BACK IN 6 MINUTES i opened system settings → privacy & security and never looked closer. grok bot did what it found: – 14 apps with screen recording. i use 2 of them – 9 with full disk access, which means mail, messages and every file – a time tracker taking a screenshot every 5 minutes, since 2023 – a vpn i deleted last year left a helper behind that still had access – a free clipboard app with access to everything: screen, files, keyboard. it had stored 41,000 things i copied. bank password, 2fa codes, my card, an api key it went category by category and turned off 27 permissions. the 5 apps i actually use kept only what they need. the clipboard app is gone, and its history with it the prompt: "Go through Privacy & Security on my Mac. For every app with screen recording, full disk access, accessibility or input monitoring, tell me what it can see and when I last used it. Keep access only for the apps I use, turn off the rest, and remove anything that doesn't need to be there." the full setup is in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's. One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste. 16-min read. Save it ↓
11
30
887
GROK BOT CHECKED ALL 27 EXTENSIONS IN MY BROWSER. 9 COULD READ EVERY PAGE I OPEN, AND A COUPON ONE GOT QUIETLY SOLD AND STARTED SENDING MY HISTORY TO A SERVER. IT REMOVED 16 AND MADE THE REST SAFE i installed most of these years ago and forgot about them. it opened each one and checked what it's allowed to see, who owns it now and where it sends data what it found: – 9 had "read and change all your data on all websites". that's my bank and my email too – the coupon extension changed owners in march. 2 weeks later an update started sending every url i visit to a server – a pdf converter uploads every file i convert – 4 hadn't been updated since 2021 – i'd used 11 of the 27 in the last month it removed 16. for the 11 i use, it found versions with fewer permissions or set them to run only when i click them the prompt: "Check every extension in my browser. For each one: what permissions it has, whether it changed owners, when it was last updated and what it sends where. Rank them by risk and tell me which ones I actually use. Ask before removing anything." the full setup is in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's. One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste. 16-min read. Save it ↓
11
34
845
GROK BOT WATCHED 1,140 HOURS OF LECTURES, READ 38 TEXTBOOKS AND 212 PAPERS ON ONE SUBJECT, THEN CUT ALL OF IT DOWN TO 1,800 ANKI CARDS. I LEARNED WHAT A FIRST-YEAR MED STUDENT LEARNS IN 6 WEEKS i picked pharmacology, the class med students call the worst to memorize. grok bot did the reading: → 1,140 hours of lectures from 9 universities, through the transcripts → 38 textbooks, every chapter → 212 review papers → about 21 million words then it threw out everything that repeated. 21 million words became 9,800 facts, and 1,800 of them showed up in nearly every source. those became the cards each card has the fact, one line on why it works that way, and where it came from, like "textbook ch. 10, lecture 14 at 32:10" it also reads my answers. when i miss the same kind of card twice, it writes a new one that explains it from another angle. 18 minutes a day for 6 weeks: - week 1 practice exam: 31% - week 6: 84% pick any subject. code, law, investing, a language steal it: "I want to learn [subject] to the level of [goal]. Find the best lectures, textbooks and papers on it. Go through all of it, keep only the facts most sources agree on, and turn them into Anki cards with the source on every card. Track my answers and rewrite any card I miss twice." the full setup is in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's. One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste. 16-min read. Save it ↓
14
34
778
I PICK UP MY PHONE 96 TIMES A DAY. GROK BOT PUT ONE JAPANESE WORD ON MY LOCK SCREEN EVERY TIME. 60 DAYS, 1,400 WORDS, ZERO STUDY SESSIONS i've wanted to learn japanese for years and never opened the textbook twice. so i stopped trying to study how it works: → every night at 2am grok bot reads my day: the anime i watched, my texts with yuki, my tokyo trip notes, a ramen video → it picks 20 words i'd actually use and makes anki cards with the sentence they came from → each time i pick up my phone, one card sits on the lock screen. kanji first, reading and meaning a second later → "got it" or "again", then the phone is mine after 60 days: - 96 pickups a day - 5,760 cards seen - 1,400 words, past the ~800 on the jlpt n5 list - 0 study sessions the words stick because every one of them came from my own day. 約束 (promise) came from an anime, 予約 (booking) from my trip notes steal the prompt: "Every night, read what I watched, read and wrote today. Pick 20 Japanese words I don't know yet that I'd actually use. Make Anki cards with the sentence they came from: kanji on the front, reading and meaning on the back. Show one on my lock screen every time I pick up my phone. Never repeat a card I marked easy." works with any language you keep putting off. the setup is in the article ↓
I put together one guide for anyone who wants a workflow like the SpaceXSI team's. One message builds your whole Grok Bot setup. Plus the team's rules, prompt habits, routines and permissions. Every setup is a block you paste. 16-min read. Save it ↓
14
1
30
992