Promtor retweeted
Grok Bot treats every trade like a rocket launch, and honestly that's the most accurate metaphor I've ever seen for 100x leverage.
Every coin is a mission. Before liftoff there's a GO / NO-GO poll, and all six agents have to answer: Scout, Filter, Analyst, Grok, Sniper, Guard. Only when every light is green does the countdown start. T-3, T-2, T-1, LIFT. Then the rocket either reaches the moon or blows up on the way.
The flight log reads exactly like mission control:
T-3 $GIGAS poll · all GO
LIFT $GIGAS ape 100x
MAXQ $GIGAS trail armed
MOON $GIGAS +$6,085
And when it goes wrong:
LIFT $DUMPY ape 88x
REKT $DUMPY -$1,218
By mission 9 the manifest looked like this: 6 mooned, 2 rekt, 428 rugs scrubbed before they even reached the launch pad. The mission fund went from $8,820 to $26,630 on paper, 3.02×.
Best flights: $GIGAS +$6,085, $MOONR +$4,733, $WAGMI +$3,968.
Explosions: $DUMPY -$1,219, $BURN -$875.
My honest opinion:
The GO / NO-GO poll is the smartest thing on this screen. Six agents have to agree before anything happens. One NO and the launch is scrubbed. That's 428 rockets that never left the ground, and that's where the real edge is. Most traders don't have a single checkpoint before they press buy. This thing has six.
The trail armed at MAXQ is great too. It means that as soon as the rocket is in the air, there's already a plan to protect the profit. No hoping it keeps going.
But let's be real about the fuel. Every one of those launches ran at 50x to 125x. That's why the explosions are so big. $DUMPY lost over $1,200 in one shot. At that leverage one bad candle is a failed launch, no matter how good the checklist is. The title says 18+ for a reason. If I were running this with real money, I'd keep the six-agent poll and cut the leverage down to something boring.
Comment "LAUNCH" and I'll DM you how the GO / NO-GO poll works and what makes Filter scrub a mission
And tell me: if you had a mission control for your own trades, how many of your launches would have been scrubbed? Half of mine would never have left the pad
Promtor retweeted
I went to bed with an outdated CV and woke up to 7 interviews on my calendar for next week. Grok Bot did the job hunt overnight.
If you've ever looked for a job, you know the worst part isn't the interviews. It's the grind before them. Scrolling hundreds of listings, rewriting the same CV for the tenth time, sending applications into the void and refreshing your inbox for days.
So I handed that whole grind to Grok Bot from 23:00 to almost 07:00. Here's what you're seeing in the video.
The pipeline at the top is a conveyor belt of real-looking job cards. Every role passes through three gates:
FIT ≥ 80%: does this job actually match me?
TAILOR CV: rewrite the CV for this exact role
SEND: only then does the application go out
Most cards never make it past the first gate. They fall off the belt with the reason in red: on-site only, under $90k, needs 10 years of experience, unpaid test task. That's the filter I never have the discipline to apply myself. At 1am I'd apply to anything.
The middle panel is my favorite. Grok Bot reads the job post, highlights what they actually care about (design system, experiments, analytics), then crosses out my lazy CV lines and rewrites them. "Designed screens for various projects" becomes "Shipped a design system used by 40+ engineers". The match score climbs from 57% to 92%.
By morning:
1,486 roles scanned
63 applications sent
19 replies
7 interviews booked
And the inbox panel shows it triaging replies on its own. Interview invites get booked straight into the calendar, rejections get archived, a salary question gets a drafted reply waiting for me to approve.
My honest opinion:
One thing matters more than anything here: the rewrites only make my real experience clearer. They don't invent it. If an AI starts adding skills you don't have, you'll get caught in the first interview. The value is in translating what you actually did into the language each company uses, and that's exactly where most CVs fail.
The second thing is that saying no is the real superpower. 63 applications out of almost 1,500 roles. It's not spamming, it's choosing. 7 interviews from 63 targeted applications is a way better ratio than 300 random ones.
And honestly, the emotional part is underrated. Job hunting is exhausting because of the rejection loop. Letting an agent take the boring, painful middle part means I show up to interviews with energy instead of burnout.
Comment "HIRED" and I'll DM you how I set up Grok Bot for a job hunt: the fit rules, how it tailors the CV without making stuff up, and how it handles the inbox
And tell me honestly: how many applications have you sent that you never heard back from? I stopped counting at 80
Promtor retweeted
Claude Code can now discover a sound that leaks your location (and it's everywhere)
It's actually getting scary for all of us...
here's honest result of my 12-hour tests + public repo to check and mask it:
how it works, Doppler curve method by YouTuber colsto:
1 → when a plane flies past, its engine hum slides down in pitch, like a car horn going past you. On a spectrogram (a picture of sound) that slide is a curve
2 → the curve's shape gives the plane's speed, how close it came and the exact second it passed
3 → sites like Flightradar24 log where almost every plane is, every second - match the curve to that data and you get which plane it was, and where the mic had to be
4 → know the plane, the second and the distance, and the mic is on a line next to its path, a second plane pins the exact spot
What I got with Claude Code:
• 10 seconds of a takeoff on a public Chicago airport webcam. From the sound alone it got the speed, distance and second, then picked one flight out of 35 that fits: Southwest SWA4113 (I knew where camera was)
• colsto posted 1-hour of sound from a secret spot and dared people to find where it was. From the planes in it, Claude Code found the right town: Oyster Bay, New York (with hints: we knew area, and another dev marked tha planes by hand: doppler-locate)
Where Claude failed:
• blind, with only the sound, the time and a 40 by 40 km square of the map, it found 0 of 6 recordings
• Miami looked like a hit at 749 m, but sound from 30 min later landed the same spot, it was guessing from airport traffic, not the sound
Bottom line:
I gave it 12 hours, free flight data, a 16 GB PC and airport cams full of jet roar. The method itself works: doppler-locate searched the whole US with no hint and put colsto's camera 67 m away, on a much bigger PC.
- WIth full data and a stronger computer, I think Claude Code can do it blind too, my repo is the starting point
anyway: now the sound can find you. Claude Code already gets you halfway, colsto's full method does the rest
check your own videos (free):
1. paste into Claude Code: npx skills add Bodila51/colsto-leak-and-mask --skill flyover-check
2. say "check my video": it tells you if a plane, electrical hum or the GPS and date in the file could give away where you filmed
3. say "make a safe copy": it cuts those sounds and strips the GPS and date
Find your location. Then read article below to make this faster:
Promtor retweeted
I let Claude build my trading desk and holy sh*t, I was doing it all wrong.
No joke. One screen and it exposed every bad habit I had.
Paper money, simulated markets, zero real cash on it. I built it to learn, not to flex. And the stuff that actually taught me something was the boring panels nobody ever screenshots.
Here’s what slapped me in the face:
The correlation panel. I asked for a “clean dashboard” and Claude threw it in anyway. I almost deleted it. Then I saw that my “five different positions” were the same damn bet five times. Not diversified. Just five times more exposed.
The exit panel. Fills, slippage, rejects. For years I blamed my entries. Nope. Half the bleeding was sloppy exits. A good entry with a sh*t exit is just losing in slow motion.
The sizing slider. I used to size by vibes. Feeling good? Bigger. Down on the day? Even bigger, gotta win it back. That’s not a strategy, that’s a tantrum. Now a cold number tells me exactly how much my edge is worth. Annoying. Also right.
My honest take:
AI didn’t make me a genius. It did something way more useful. It made it really f*cking hard to lie to myself.
Everyone wants the bot that prints money. Nobody wants the bot that shows them why they keep losing. The second one is the one that actually helps.
So build the boring panels first. The pretty chart is for Twitter. The sizing panel is what keeps you alive.
Be honest: when you open your screen, where do your eyes go first? I’m betting most of you look at the wrong thing. I did for years.
Promtor retweeted
grok tracked 6 global events, placed 3 trades and finished +$1008 without risking a single dollar
this is world monitor, an autonomous geopolitical intelligence terminal powered by grok. it watches global events, connects them to financial markets and checks whether prediction markets have already priced in the news before making a trading decision
five specialized agents work together, each handling a different stage of the operation
> event scanner detects geopolitical and supply-chain disruptions
> stock mapper connects events to affected assets and sectors
> prediction check compares Polymarket and Kalshi odds
> risk manager calculates exposure and enforces stop-loss limits
> execution builds and submits orders through the Alpaca paper API
one pipeline outage triggered an XLE buy, a Midwest drought produced a WHEAT position, and a metals sanctions alert led to a GLD trade
after one hour: 6 events analyzed, 3 paper orders entered, 2 opportunities rejected as priced in, 1 kept on watch and +$1008 net paper gain
zero real capital was used. the system is designed to test its decisions in paper mode before any real-money execution
Promtor retweeted
He pasted one prompt into Grok Bot with $500 and went to sleep.
Woke up 8 hours later to $6,359. 12.7x.
Six agents ran the whole night:
- Scout scanned 13,338 pairs.
- Filter killed almost everything with low liquidity or unlocked mints.
- Analyst scored what survived.
- Executor took the trades.
- Risk enforced an −8% hard stop and a +15% trailing stop.
- Logger wrote every decision.
41 trades. 31 green, 10 red.
Biggest single win: $1,589.
He didn’t pick a coin. Didn’t set a stop. Didn’t open a chart.
The desk just ran while he was asleep.
Promtor retweeted
Grok Bot vs The Market. One round, 99 seconds, and I've watched this fight more times than I'd like to admit.
The whole run is built like an old arcade fighting game. On the left is a little white robot with a G on its chest. On the right is The Market, a giant red bear, level 99. Every trade is a move. When a trade hits, the robot lands a punch and The Market loses HP. When a trade goes wrong, the bear counters and the robot takes damage.
The move list at the bottom is honestly hilarious:
$FURY +$928
$HOOKR +$2,347
$SLAMR +$1,102, 2 hit combo
$KNOX +$3,218
$TITNZ +$4,085, the finisher
And the counters:
$JABZ -$684
$SMASH -$411
$UPCUT -$903
The super meter fills with every clean hit, and at the end it fires "SUPER: TO THE MOON", the bear goes flying off screen, and it's K.O. Market defeated.
$3,690 at the start, $16,713 at the end on paper. 4.53×. 7 hits, 3 taken, and 267 rugs blocked before they could even throw a punch.
My honest opinion:
The genius of this format is the HP bar. Most trading screens just show you money. Here you see your health. The robot finished with 67% HP. It took real hits, three of them, and still won because none of them were knockouts. That's exactly how it should feel. You're allowed to get punched. You're not allowed to get knocked out.
The 267 blocked rugs are the defense nobody shows in the replay. Before every fight there's a list of moves the robot simply refused to make. That's where most of the HP was saved.
But I'll be straight: look at the numbers next to the moves. 50x, 75x, 100x, 125x. The arena literally says high leverage, 18+. That's why the counters hurt so much. $UPCUT took almost a thousand dollars in one hit. In a real market, one counter at 125x can take the whole HP bar. I'd keep the blocking and play a much lower difficulty.
Comment "KO" and I'll DM you how the bot decides which moves to block and when it goes for the combo
And tell me honestly: in your own trading, how many rounds have you lost to The Market by K.O.? I've lost more than I've won, and the bear wasn't even level 99
Promtor retweeted
this is f*cking brilliant... Grok Bot is becoming a full-blown open-source ecosystem.
There are already 600+ community-built setups, tools, and experiments on GitHub.
I went through the ecosystem and picked 20 repositories worth checking out.
1. AGENT TEAMS & AI WORKSPACES
01 / OpenCompany (1.0K★)
Build an AI company with specialized employees, delegation, persistent memory, and reusable skills.
github.com/zeenie-ai/OpenCom…
02 / Open Duet (6.2K★)
Create self-hosted AI coworkers with individual roles, routines, memory, and approval controls.
github.com/Anil-matcha/open-…
03 / OpenMausBot (5.7K★)
A local-first AI coworker setup with connected apps and computer environments for your agents.
github.com/milind-soni/OpenM…
04 / OpenBot (7.3K★)
An open coworker-platform template where agents get their own browsers, files, and tools.
github.com/CopilotKit/OpenBo…
2. CODING & AUTOMATION
05 / Orca (107.7K★)
Run multiple coding agents in parallel with isolated Git worktrees and desktop or mobile control.
github.com/stablyai/orca
06 / Codync (948★)
Message existing coding agents from your desktop, phone, or terminal.
github.com/leepokai/Codync
07 / devspace (5.3K★)
An MCP coding harness for controlled project access, scoped edits, shell commands, and isolated worktrees.
github.com/Waishnav/devspace
08 / unbrowse (810★)
Turn browser workflows into reusable API access so agents don't have to repeat the same clicks.
github.com/unbrowse-ai/unbro…
3. MEMORY & RESEARCH
09 / obsidian-second-brain (5.0K★)
Give agents persistent Markdown memory, a shared Obsidian vault, and a crew logbook for handoffs.
github.com/eugeniughelbur/ob…
10 / system_prompts_leaks (70.3K★)
Explore an archive of assistant system prompts, including examples of how AI roles and instructions are structured.
github.com/asgeirtj/system_p…
11 / openmeld (1.4K★)
Coordinate tasks and shared context between people and AI agents.
github.com/amplifthq/openmel…
12 / grokbot-field-notes (639★)
A collection of Grok Bot roles, playbooks, operating rules, and lessons from a 72-hour build.
github.com/unicodef1wn/grokb…
4. BROWSER, COMPUTERS & AGENT
INFRASTRUCTURE
13 / celesto (1.1K★)
Give agents persistent computer environments with browsers, terminals, files, and snapshots.
github.com/CelestoAI/celesto
14 / web-access (9.3K★)
A web-access toolkit combining browser CDP, multiple access channels, and reusable site-specific methods.
github.com/eze-is/web-access
15 / rakazo (6.0K★)
An open-source Grok Bot alternative where you choose the model and sandbox powering your AI teammates.
github.com/elie222/rakazo
16 / rome (875★)
A self-hosted agent OS that turns completed work into reusable actions, apps, and memory.
github.com/rome-os/rome
5. SPECIALIZED TOOLS & EXPERIMENTS
17 / gawkbot (1.5K★)
A self-hosted bot workspace for automating routine tasks and building microapps.
github.com/najmuzzaman-mo/ga…
18 / grok-bot-0.18-reconstructed (3.5K★)
An unofficial macOS reconstruction of Grok Bot 0.18 for exploring how that client version works.
github.com/b-nnett/grok-bot-…
19 / bloub (1.9K★)
A morphing SVG recreation of the Grok Bot avatar with 14 visual states.
github.com/jeremy-prt/bloub
20 / aora-bot (585★)
A lightweight expression engine with 32 SVG states for chatbot interfaces and desktop assistants.
github.com/sam70361/aora-bot
You don't need all 20.
Find the one missing piece in your workflow. Test it on a real task. Keep it only if the result is worth the setup and running costs.
Save this list. Add one useful tool at a time.
And if you're building a serious Grok Bot team, start with the workflow, not the number of agents.
Promtor retweeted
what the f*ck. I asked Grok to build me a pump.fun terminal and it trades one coin at a time right in front of you. it called it issa terminal
$500 → $3,657 in 22 seconds
how it works on screen:
→ a new pump.fun launch pops up
→ Grok checks 3 things: dev wallet, bundled buys, posts on X
→ all green → buys, and you watch the % go up live
→ one red → skips it. every skipped coin rugged right after
→ every sell drops a pin on the chart with the coin's logo and the profit
what it traded:
→ $SLOTHY · dev 2%, 12K posts on X → +146% → +$146
→ $BULLY · dev holds 44% → skipped → rugged
→ $MEOWSK · 19K posts on X → +212% → +$254
→ $INFERNO · dev 1% → +274% → +$411
→ $SHADOW · 61% bundled → skipped → rugged
→ $DUNE · looked clean → stop hit → −$17
→ $TEDDY · 58K posts on X → +684% → +$1,368
→ $PRISM · 27K posts on X → +238% → +$524
→ $RATTY · dev already sold → skipped → rugged
→ $LIFTOFF · 22K posts on X → +196% → +$470
"done. $500 → $3,657. want it in your wallet?"
tapped cash out. sent.
7 trades. 6 green. 1 loss of $17
3 skipped. all 3 rugged
the rules Grok wrote for itself:
1. dev holds more than 10% → skip
2. bundled buys over 25% → skip
3. dev already sold → skip
4. every buy gets a stop
5. one coin at a time. never all in
you're still aping pump.fun launches blind. Grok checks every one in a second and shows you why it buys
comment "CLEAN" and I'll show you how it works
illustrative build, numbers simulated
Promtor retweeted
I genuinely don't understand why everyone isn't using this yet
With the help of this concept, my business reached an income of $13,000 per month
Andrej Karpathy noted: "The primary regret among engineers in my field right now is every AI conversation they failed to save. In 10 years, the people who started archiving in 2024 will leave the rest of us in the dust"
co-founder of OpenAI, generated 25M views around a single concept: stop using AI just to code, and start using it to construct a second brain that remembers every article, PDF, transcript, and chat you have ever provided
You simply direct Claude Code to a folder, throw any file into it, and Claude scans the content, connects it to existing records, and builds a living wiki. Your personal knowledge finally stacks up like compound interest instead of slipping away every weekend
The stack:
> Brew install obsidian, mkdir ~/vault, cd ~/vault
> Launch claude inside that directory and enter: "init a knowledge vault with CLAUDE.md as the rules, /raw for sources, /wiki for your pages"
> Place any file into /raw (a YouTube transcript, meeting notes, a 400-page PDF, or a tweet screenshot) and enter "ingest"
> Claude constructs a wiki page, links it to 5 existing documents automatically, and updates CLAUDE.md with any fresh patterns it spots in your thinking style
> Never launch a blank Claude chat window again. Every prompt runs through your vault, and every reply cites your own materials with dates and links
It takes 5 minutes to set up, letting your knowledge compound from that night onward
The harsh truth most people will scroll past: in 10 years, the engineers who adopted this in 2024 will recall every book, podcast, and meeting with solid proof. You will remember virtually nothing. Karpathy has already made his decision, and you are still falling behind
Check the article below for the complete step-by-step walkthrough
Save this to your bookmarks before you waste another Monday
Promtor retweeted
I GAVE GROK BOT $104. SIX HOURS LATER: $4,231.56
40.69x the starting balance. Now try looking at the next coin without thinking, "what if I just make this one a little bigger?"
That's the part I care about after a run like this. The next entry. A bigger balance makes a bad decision more expensive
The setup in the guide below puts a separate judge between the research and the order. It gets the candidate's dossier and answers four questions:
→ Is the demand organic?
→ Are we already late?
→ How ugly could the exit get?
→ Buy, watch or pass?
The answer doesn't give it access to a bigger bet. It can't raise the size, sign the order or change the gate because it feels confident. The staged order still waits for a human
I like that last bit. You can have every agent agreeing with each other and still want someone to ask what they're agreeing on
$104 → $4,231.56 is one six-hour session. It doesn't tell me what the next six hours will look like
I'm happy with the number. I don't want the next trade sized around how happy I am
Promtor retweeted
HOLY F*CKING SH*T. I tossed just $200 into Grok Bot again and went to sleep.
A few hours later, the screen lit up with $2,437.14.
A clean 12.19x on absolute autopilot while the system chewed through thousands of pools on Solana. Look at the video above-how coldly the bots filter out all the trash and lock in profits.
How the 6-agent desk ran this time:
→ Scanner swept 10,214 pools.
→ Heavy logger filtered and destroyed 10,010 dead tokens (scams, no liquidity, unlocked mints, and flagged wallets).
→ Analyst and Executor picked the best assets and entered a series of trades.
→ Risk management chopped losses at the root and dragged profits upward.
Session summary: balance grew from $200 to $2,437.14 across 38 executed trades.
While you keep blowing your portfolio manually at night, AI agents are pocketing 12x on passive.
Promtor retweeted
this is f*cking unfair, Grok Bot made me $1,993 in 94 minutes by stalking the top holders of every coin i held and getting out 0.4s after one of them started selling
23.15x in 94 minutes, $90 → $2,083.47
10:58am i gave Grok Bot 4 agents. the new one is called Whale Watch and it only has 2 jobs:
1. track the top 10 wallets of every coin i hold
2. the moment one of them starts selling, Risk dumps my whole bag with no confirm button
every other trade rides with a -6% stop
my one rule for it: miss a single whale exit and you're deleted
11:09am it sniped a coin in its launch minute, $65.73 in
the coin ran +848% in the next hour on $46.6K of liquidity. one wallet, 2KSL...9Ymv, sat on 14.40% of the supply the whole way up
11:59am that wallet started selling, 9.15% of the supply for $14,195
my phone lit up, the sell sheet opened on its own, filled in 0.4s. no human taps that fast
then the candle went -81%
+$527.46 out of one coin, +802%, and the phone was face down on my desk
12:16pm same thing on another coin
4AQy...29Lg held 14.43% and dumped all of it, the candle went -99%
Grok Bot was already out with +$374.66
94 minutes in numbers:
→ 17 trades, 13 green, 4 red, 2 whale exits
→ 4 stops, worst one -$19.14, it still takes small losses
→ 27 whale wallets under watch at once
→ 11:59am $837, 12:16pm $1,237, 12:28pm $2,082
honestly the scary part is the 2 charts. the price was still green both times it sold, i would have held both
you watch the chart. the whale watches you watch the chart
send this to anyone who still holds a coin "for the long run"
every Grok Bot agent and its rules are in the quoted article below ↓
Promtor retweeted
Grok Bot turned a whole night of trading into a beehive, and I can't decide if it's the cutest or the most degen thing I've ever watched.
The screen is a giant honeycomb. Every cell is a coin. Little bees fly around the hive: Scout, Analyst, Sniper, Guard, Filter, and Grok in the middle as the queen making the calls. When a coin is good, the cell turns gold and fills with honey. When it's a scam, it turns into a red wasp cell and gets crossed out.
By the end of the night:
2,113 cells scanned
337 wasps blocked
9 sweet, 3 stung
$2,965 → $13,724 in the honey jar, 4.63× on paper
The colony log reads like a nature documentary with a crypto problem:
filter: wasp $WASP · mint live
filter: wasp $HORNT · LP 0%
filter: wasp $STING · honeypot
filter: wasp $GRIME · vol $1k
grok: ape $HIVEX
sniper: $HIVEX +$3,416
The best cells were $HIVEX +$3,417, $GOLDN +$2,689 and $ROYAL +$2,214. The stings were $QUEEN -$231, $jelly -$357 and $WAXX -$298.
My honest opinion:
The honey jar is a genius idea. It fills slowly, drop by drop, and when you get stung you watch the level dip a bit. It makes you feel how small the losses are compared to the wins. 9 sweet cells did all the work, and the 3 stings barely made a dent.
The red cells around the gold ones are the real story though. Look how much of the hive is red by morning. Most of the colony is wasps. The bot's real job wasn't finding honey, it was not getting stung by 337 wasps in a row.
Now the part I can't ignore. This is labeled "degen colony 18+" for a reason. Some of those cells say 75x and 125x. That's not trading, that's sticking your hand in the hive and hoping. The filter logic is legit and worth stealing. The leverage is pure chaos, and I'd never run it like that with real money.
Comment "HIVE" and I'll DM you how the bees split the work and the exact wasp rules the filter uses
And tell me which wasp name you'd fall for first. I'd be lying if I said $ROYAL wouldn't have gotten me
Promtor retweeted
HOLY F*CKING SH*T you cannot name 3 things you learned last month and karpathy's one claude code prompt builds you a second brain that remembers everything for you in 5 minutes
andrej karpathy, co-founder of openai, posted the fix a year ago and pulled 16M views for one line: stop using AI to write code, use it to remember everything you ever learned
the setup is in the article below. here is what it actually feels like 2 weeks after you point claude code at an obsidian vault and feed it everything you read, watch and save:
you ask → "what did i actually learn about transformer position encoding this year and where did i read it"
it answers → 4 bullets, each one citing the exact pdf, timestamp or tweet you fed it, with the links. you never re-read a thing. you never forget a thing
you ask → "give me the single page i would hand a junior to onboard them to this project"
it writes → the page in 60 seconds, pulling from 11 of your own notes, 3 slack exports, and the article you starred in march
you ask → "what do i actually believe about X and what did i change my mind on this year"
it tells you → with receipts, with dates, with the thing you said in january that you have since contradicted twice
your current "second brain" cannot do any of this. your current second brain is 4,000 bookmarks tagged "read later" from 2024 that both of us know you will never open. karpathy's one prompt rebuilds the whole pile into something that actually remembers for you, in 5 minutes
full step-by-step is in the article below
bookmark this before you waste another monday
Promtor retweeted
this grok bot made $1.12M in 26 minutes off the binance hacker on TRON
oct 29. binance gets hacked. the hacker swaps 234M stolen USDT for TRX on sunswap, in a hurry
TRX in sunswap's main pool goes ~18% above the market
one bot on TRON:
> borrows USDT from the pool in a flash loan
> buys TRX cheaper in another pool
> sells it into the overpriced pool and repays, all in one tx
> what's left over is profit
> first tx at 20:49 UTC: moved 23.1M TRX, kept 623,193 TRX = $210,757
> 23 txs in 26 minutes: +2.4M TRX and +$325K USDT
> fees: $202
> total: $1.12M at market price
bitquery has it as the #1 TRON wallet of the night
bookmark this
Promtor retweeted
People are literally giving their bank passwords to AI agents now, and the biggest banks in America just started slamming the door on them.
JPMorgan, Bank of America, Citi, Morgan Stanley, Schwab, Wells Fargo. All restricting or blocking outside agents that log in with customers' own credentials. Some of these bots were reportedly calling bank call centers and doing things their owners never asked for.
My first reaction was honestly, wait, people actually do this? Then I thought about it for a minute and yeah, of course they do. If an app promises to pay all your bills and sort your money while you sleep, most people will hand over the login without reading a single word. I've clicked "allow" on way dumber things myself.
Here's the part that bugs me. To the bank, an agent with your password looks exactly like you. Same login, same permissions. So if the bot pays the wrong person or the wrong amount, who eats that? You, because you shared the password? The bank, because it let the login through? Or the AI company that built the thing? Right now nobody has a clear answer, and Bank of America is still figuring out the rules.
And I keep seeing the same pattern with AI everywhere. The tool belongs to one company, the money belongs to you, and the responsibility belongs to nobody until something breaks. Meta did it with ad terms, now it's showing up in banking. That's the part that actually worries me, not the tech.
To be fair, nothing big has blown up yet. The reports don't show mass unauthorized transfers or customer losses. Banks are closing the door early, and honestly I respect that. Most companies wait for the disaster and then write a sorry email.
But let's not pretend banks are only doing this for us. They also want to control which bots get in and on whose terms. Amazon did the same thing with Meta's shopping agent. Whoever owns the door gets to charge for the key. I think both things are true at the same time, it's safety and it's business.
I'm not against AI touching money at all. I actually think in a couple of years agents will pay bills, find better rates and cancel subscriptions you forgot about, and it'll be one of the most useful things AI ever does. But it has to happen through official access with limits, a log of everything, and a human saying yes before money leaves. Not by sharing your password and hoping the bot behaves.
So until that exists, my rule is simple. AI can look at my money. It can't touch it without asking me first.
Would you let an AI agent move your money on its own, or is that a hard no for you too?
Promtor retweeted
Grok Bot reads the news faster than I can read the headline.
ETF flow spike hit at 10:47. It was long in 0.4 seconds.
$200 → $1,546 in one US session.
9:30am ET, market open.
News + high-vol mode. Max 10% per trade.
It watched X, exchange announcements and ETF flow data live.
On a real catalyst it entered within seconds of the headline.
Skip anything already +80% in the last hour.
By 2:30pm:
3,184 pairs scanned.
47 news hits.
24 trades. 17 won. 7 lost.
Exchange listing +$123.
Whale tweet +$141.
ATR breakout +$124.
Volume 6x the average +$124.
Headlines hit at 10:47 and 12:08. Both times it was in before I unlocked the phone.
The losers were tiny. A fake headline just costs one stop, never the account.
Promtor retweeted
Somebody built a trading bot as an actual circuit board, and every trade turns into a Tetris block. I've watched it about ten times now.
Here's what's on screen. In the middle sits the Opus 5.5 chip. Six smaller chips are wired around it: Scout, Filter, Analyst, Claude, Sniper, Guard. When one of them is working, its trace lights up and you can see the signal travel through the board. Data comes in from the Solana RPC at the top, and money goes out to the wallet port at the bottom.
My favorite detail is the chip temperature. It sits at 40°C when the bot is idle and climbs to 60°C with the clock speed going up while it's thinking. It means nothing for the trading, and it's still exactly the kind of detail that makes you feel the machine is working.
Under the board there are three slot machine reels: token, score, verdict. Every coin gets a spin. $RUGZ, score 0.39, KILL. $APEX, score 0.93, PASS. Most spins land on KILL, and that's how it should be. By the end of the run that was over 1,850 spins and only a handful of passes.
Then comes the balance panel, which is literally Tetris. One block is worth $60. Every win drops new blocks onto the stack: $ZAP +$264, $KOVA +$412, $MAKO +$846. A loss like $ONYX -$121 knocks blocks off. By the end the stack is almost full.
$1,500 → $5,618 on paper, and every fill in the hexdump says the same thing: risk 2%.
My honest opinion:
The Tetris idea is smarter than it looks. A number like $5,618 means nothing to your brain. A tall stack of blocks does. And when a loss eats a couple of rows, you feel it, but you also see that the stack is still standing. That's how drawdowns should feel: visible, but not the end of the world.
What I respect about Opus 5.5 in this setup is the hexdump on the right. KILL, PASS, FILL, EXIT, one line per decision, each with a reason: top10 41%, honeypot, mint live, score 0.85. No vibes. If I want to know why it bought something, the answer is one line away.
And the 2% risk on every single fill is what keeps the stack from collapsing. It didn't size up after $MAKO, and it didn't get scared after $ONYX. Same block size every time. That's boring, and boring is what makes it work.
Comment "BOARD" and I'll DM you how the six chips are wired to the core and the exact kill rules behind the reels
And tell me honestly: if your trading history were a Tetris stack, would you be at the top of the screen or already game over? Mine would have lost to the first $ONYX
