Pinned Tweet
🚨 شوف الطالب ده عمل حركة غريبة جداً على مكتبه وصدمت كل اللي حواليه..
شوف عمل إيه بالذكاء الاصطناعي وخلى الأجهزة تشتغل مكانه وهو نايم! 😱
القصة بدأت لما الطالب ده اشترى iPad و Mac Mini بـ 2,200 دولار وربطهم بجهازه الـ MacBook.. وبقى عنده 3 شاشات على المكتب لدرجة إن زمايله في السكن افتكروه بيعدّن عملات رقمية (Crypto).
احفظ البوست ده فوراً عشان ما يضيعش منك!
بس الصدمة الحقيقية ظهرت تاني يوم الصبح بعد ما ظبط الأوتوميشن ونام:
- اللي حصل وهو نايم: كل اللي عمله إنه ربط الـ 3 أجهزة ببعض وشغّل سيستم أتمتة (Automation) بالذكاء الاصطناعي، وسابه شغال ودخل ينام بهدوء.
- المفاجأة الصبح: صحي لقى السيستم لوحده استخرج وفلتر مئات العملاء المحتملين (Leads)، وكتب إيميلات مخصصة بالكامل لكل عميل فيهم، وملى بيهم نظام الـ CRM من غير ما يلمس الكيبورد لمسة واحدة!
- الأرقام المرعبة: المجهود ده كان فيه فريق عمل كامل بيعمله قبل كده وبيكلف الشركات 7,000 دولار شهرياً.. الطالب ده عمله بـ 2,200 دولار دفعهم مرة واحدة تمن الأجهزة وخلاص!
الموضوع أكبر من مجرد حالة فردية؛ فيه حوالي 360 مليون شركة في العالم، و310 مليون شركة منهم لسه بتدفع رواتب ومصاريف ضخمة لبشر عشان يعملوا وظائف الماكينة والـ AI بيعملها حالياً أحسن وأسرع بكتير.
والصدمة الأكبر؟ إن فيه حوالي 100,000 شخص بس على كوكب الأرض كله هما اللي فاهمين اللعبة دي وعارفين إزاي يطوعوا الـ AI ويبنوا السيستم ده!
لايك وفولو واحفظ البوست عشان يوصلك كل جديد!
Web3Arabs retweeted
Claude Code tip: your agent writes code faster than you can review it. so 𝘆𝗼𝘂 𝘀𝘁𝗼𝗽𝗽𝗲𝗱 𝗿𝗲𝘃𝗶𝗲𝘄𝗶𝗻𝗴 𝗶𝘁.
Claude's official security plugin flips it. Every bug one agent finds goes to 𝟯 𝗼𝘁𝗵𝗲𝗿 𝗮𝗴𝗲𝗻𝘁𝘀 𝘁𝗿𝘆𝗶𝗻𝗴 𝘁𝗼 𝗽𝗿𝗼𝘃𝗲 𝗶𝘁 𝘄𝗿𝗼𝗻𝗴. If 2 of them can't, it's real and it makes the report.
fix that with three lines in Claude Code (on Opus 5.5):
/plugin install claude-security@claude-plugins-official
/reload-plugins
/claude-security scan the whole repo at high
It sends agents through every file, and 𝗲𝘃𝗲𝗿𝘆 𝗯𝘂𝗴 𝘁𝗵𝗲𝘆 𝗳𝗶𝗻𝗱 𝗵𝗮𝘀 𝘁𝗼 𝘀𝘂𝗿𝘃𝗶𝘃𝗲 𝟯 𝗼𝘁𝗵𝗲𝗿 𝗮𝗴𝗲𝗻𝘁𝘀 𝘁𝗿𝘆𝗶𝗻𝗴 𝘁𝗼 𝗸𝗶𝗹𝗹 𝗶𝘁. Whatever's left is real.
Then run it again, pick Suggest patches, and it hands you the fixes.
Web3Arabs retweeted
While my GrokBot was reverse engineering my smart TV, Claude Code found f*cking scary privacy problem...
My TV takes screenshots of everything on my screen and sends them out, even when I watch from my laptop over HDMI
here's what else it found and 5 steps to fix yours:
what Claude Code found:
• my TV talked to 14 different servers in one evening while I just watched YouTube
• "Viewing Information" was on since first setup and I never turned it on
• every app I opened was reported with the exact time I opened it
• my TV kept sending data in standby while screen was black
• and much more...
and it's not just me:
- smart TVs capture what's on screen as often as every 10 milliseconds, even from your laptop over HDMI (UC Davis and UCL)
- researchers found trackers in 89% of Fire TV channels and 69% of Roku channels (Princeton University)
how to fix yours:
1. install repo below in Claude Code (works with GrokBot too) and send it a photo of your TV's Settings → About screen
2. turn off ACR, it hides under names like "Viewing Information Services" or "Live Plus"
3. turn off personalised ads and reset your advertising ID
4. block the tracking domains on your router, Claude Code gives you the full list
5. keep TV offline and plug a streaming stick in, much easier to control
Save it. Then read article below on how to find these problems:
Web3Arabs retweeted
Claude Code found stalkerware on my sister's Android phone. a hidden app with no icon, called "System Service", that had been sending her location, messages and call log to a server every 15 minutes for 5 months.
she told me her ex always seemed to know where she was. we thought she was being paranoid.
i connected her phone to my MacBook with USB debugging on and ran Claude Code. it found a package with no launcher icon, installed by hand from an APK, not from the Play Store.
package → com.sys.***.service
server → 185.***.***.62
location → Netherlands
last upload → 11 minutes ago
installed → 5 months ago
someone had her unlocked phone in their hands for about 2 minutes. that is all this kind of app needs. after that it hides itself and runs in the background.
what it had access to:
→ accessibility service (it can read everything on the screen)
→ notification access (every message preview, every app)
→ device admin (so she couldn't uninstall it the normal way)
→ location, microphone, call log, SMS
Google Play Protect was switched off. she never turned it off. that is the first thing these apps do.
this is not rare. Kaspersky counted over 31,000 people hit by stalkerware in 2023 alone. and this kind of app needs someone holding the unlocked phone to install it.
she didn't cry. she just went very quiet and asked how long. 5 months.
what we did that night:
→ took screenshots of everything first, in case she needs it as evidence
→ removed device admin, then uninstalled the app
→ turned Play Protect back on and ran a full scan
→ changed her Google password and every other password from my phone, not hers
→ set a new screen lock pin that nobody else knows
the full step by step is below. takes 3 minutes. run it on every phone in your house tonight.
full setup below. takes 2 minutes. save it now so you don't lose it.
Web3Arabs retweeted
A Grok Bot prompt that every user should use.
drop this into Grok Bot and watch it build your first team of AI agents with prompts, tools, handoffs, and workflows for you
Steal this before everyone starts using it:
Web3Arabs retweeted
Grok Bot could be the first routine you write instead of repeat
It can wake up on schedule, do the job on its own computer and only ping you when a decision is yours
In this article, I show you how
Web3Arabs retweeted
Claude Code usage limit hitting earlier every week?
Paste this into Claude Code. It digs through your last 7 days of sessions, finds 𝘁𝗵𝗲 𝟯 𝗹𝗲𝗮𝗸𝘀 𝗱𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝘆𝗼𝘂𝗿 𝘂𝘀𝗮𝗴𝗲, shows how much each one costs, and writes rules to 𝘀𝗲𝗮𝗹 𝘁𝗵𝗲𝗺 𝗳𝗼𝗿 𝗴𝗼𝗼𝗱:
Don't edit anything until I approve. Build a quick script that pulls the token counts from my past week of sessions in ~/.claude/projects, and keep the log files out of your context. Rank the 3 patterns that wasted the most, like opening entire files, pasting huge test or build output, launching subagents, or letting one session run for hours. Show the rough percentage each one ate, plus one rule that would have prevented it. Once I approve, save the rules to the instruction file this project uses: CLAUDE.md first, AGENTS.md if there's no CLAUDE.md. Mask any secrets.
Using Grok Bot? Send it the prompt as-is and it'll run it on your machine for you.
Using Codex? Point the same prompt at ~/.codex/sessions and keep the rules in AGENTS.md
Web3Arabs retweeted
Elon did it again. Grok Bot is now the best AI agent you can use.
Most of these Bots would have needed a dev team and a $10K budget a year ago.
🧵: 10 insane Grok Bot agents (just tag bot to install these).👇
Web3Arabs retweeted
This Grok Bot prompt is worth saving.
turns Grok Bot into your personal money-making strategist
it helps you build your first money-making business with products, sales, automations, and everything for you
Steal this before everyone starts using it:
Web3Arabs retweeted
Claude Code tip: 𝗜'𝗺 𝗯𝗲𝗴𝗴𝗶𝗻𝗴 𝘆𝗼𝘂, add these two lines to your AGENTS.md:
"Never trust the frontend for permissions, pricing, or limits. Every check that matters must happen on the server, scoped to the logged-in user."
"Before running any migration or query that changes existing rows, show me the exact SQL and what could be lost, and wait for my approval."
Web3Arabs retweeted
Elon nailed it. Grok Bot is now the most powerful AI agent anyone can use.
Each of these Bots would have cost you a full-time hire a year ago.
10 insane Grok Bot agents (just tag @bot to install).👇
Web3Arabs retweeted
I am begging you to paste this into Grok Bot before bed...
drop this into Grok Bot and watch it build your first online income system with offers, customers, and sales workflows for you
save and steal this now:
Web3Arabs retweeted
this is f**king dangerous
someone built a cinematic scene from one JSON prompt with Claude Opus 5.5
I rebuilt the exact same thing with Kimi K3.1 in two hours, at 3x lower cost.
both nailed it first try. Kimi K3.1 Pro just finished faster.
nobody expected that. Opus 5.5 was supposed to dominate long reasoning without competition.
you have less than 24 hours to set this up as this takes away 20% of your usage limit
save and bookmark it no matter what
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro
Web3Arabs retweeted
Vibecoders, 𝗱𝗼 𝘆𝗼𝘂𝗿𝘀𝗲𝗹𝗳 𝗮 𝗳𝗮𝘃𝗼𝗿, add these two lines to your AGENTS.md:
"Cut settings pages, filter bars and mode switches. If a feature needs ten knobs to be useful, pick the right default and hide the rest."
"Never solve a design problem by adding more UI, or a code problem by adding another layer. Remove parts until the right action is the only obvious one."
Web3Arabs retweeted
Grok Bot is becoming the 24/7 AI team everyone needs.
One Chief of Staff bot. Four specialist bots on a shared cloud computer. Running routines every 5 minutes. Even making money for you.
10 wild examples and workflows.
Bookmark this.
Web3Arabs retweeted
stop using grok bot like a chatbot.
drop this into Grok Bot and let it build an entire team of AI agents that research, prompt, review, and work together for you
steal this before everyone starts using it:
Web3Arabs retweeted
holy sh*t. Karpathy's second brain now lives inside Grok Bot: stop re-explaining your work to AI every chat, keep a wiki that updates itself
i gave it a team of three, and each one remembers its own job:
the collector keeps every email, doc and meeting note exactly as it arrived
the organizer updates every project it affects, usually 10 to 15 at a time
the chaser runs every night looking for follow-ups you missed, deadlines that moved and tasks with no owner
→ you ask "where are we on this": the answer becomes a page, so the next question takes two seconds
→ an email quietly moves a deadline: you're told, it's never changed silently
→ a task sits 7 days with nobody on it: it turns red
the most organized-looking desk is usually where the important thing gets lost
Catch it. File it. Chase it.
no project stays on the list without:
> the very next step
> the person who owns it
> when it was last touched
the prompt i use:
"Set up three bots for my work. One keeps my emails and notes untouched, one maintains a page for every project, one checks every night for missed follow-ups and moved deadlines. Walk me through the plan before you build any of it"
pick the project that's stressing you most this week, not your whole job
the whole setup in the article below
save it
Web3Arabs retweeted
the whole idea of a "growth stack" just got rewritten.
Nectar is a set of agents that spot trends, find creators, and reply to every comment and DM in your brand's voice. one platform doing the job of 5+ legacy tools
built by the team that ran News Feed and Creator Monetization at Meta and scaled Groups past a billion users
$150M attributed to brands already. backed by Anthropic's Anthology Fund
growth runs on AI agents now, not tabs.
🤝 Paid partnership
Engineers stopped writing every line of code and started directing agents.
Growth and marketing teams are in that moment now.
Meet the next chapter of Nectar: the agentic growth stack. One platform replacing 5+ legacy tools. Agents that work every message, creator outreach and creative research about your brand, and tie it all back to revenue 🍊
Web3Arabs retweeted
grok bot is f**king insane.
drop this into Grok Bot and it will design your first team of AI agents with prompts, tools, workflows, and everything for you
save and steal this now:
Web3Arabs retweeted
Grok Bot could be the first team you manage instead of join
Six bots can run 100+ agents, split the work, hand it off to each other and send you one summary in the morning
In this article, I show you how
Web3Arabs retweeted
holy sh*t. Elon says Grok Bot is the fastest-growing thing he has ever seen, so the people who win are the ones who make it do work they'd pay for
three bots in one group chat, each with its own memory:
the hunter takes every job you hate doing each week and writes it down exactly as you do it
the builder turns each one into a skill with steps, rules and a stopping point, usually 1 to 3 a week
the checker wakes up on a nightly routine and hunts skills that failed, steps that changed and work nobody approved
→ a skill works 3 times in a row: it becomes a routine, so the team grows from use, not only from setup
→ a website changes and a skill breaks: the checker tags you instead of guessing a fix
→ a skill goes 30 days without running: it gets marked dead
an AI can look busy while doing nothing you'd pay for
Hunter notices. Builder teaches. Checker doubts.
every skill has to show:
> the job it replaces
> the step where it stops for you
> the hours it saved this week
paste this into Grok Bot:
"Build me an automation team with three bots in one group chat. List the weekly jobs I describe in jobs.md, turn each one into a skill with a clear stopping point, run a nightly check for failed skills and never act outside the skill without my yes. Show me the structure first. Do not create anything until I confirm"
start with one job you hate, not every job you have
full setup below

















