
Oliver
77 posts


A guy found a bug in Claude and instead of opening an issue on GitHub he went on stage.
The room cracked up. On his screen Claude Code was switching between .Claude and .Claire. Back and forth. Over and over.
Because Claude runs without memory, without rules, without guardrails — for 99% of people who launch it every day. It invents names. Writes files wherever it wants. Forgets what it did in the last session.
Three settings that put it in line in 5 minutes:
1. CLAUDE.md in the project root
A text file with rules for Claude Code: "don't touch /legacy", "tests go in /tests", "TypeScript only". Claude reads it before every task and follows. Fixes invented names and file chaos.
2. Memory MCP server
Gives Claude memory between sessions — it remembers yesterday's decisions, your code style, bugs already fixed. Installs with one command. Fixes "forgets what it did last session".
3. Permission rules in settings.json
You spell out what Claude can and cannot do without asking: "don't run git push", "don't touch .env". It asks before every such action. Fixes chaotic file actions.
5 minutes per setting. Claude stops being a junior on steroids and becomes a controlled tool.
Bookmark this. You'll get why in a week.
darkzodchi@zodchiii
English

A 19-year-old student turned AI cartoons into a $3,000/month income before she even learned how to properly edit video.
On her lunch break, she opens TikTok on her old tablet and notices something strange:
The same cartoon mouse with a crown keeps showing up in her recommendations over and over, racking up millions of views.
She downloads one of the videos, opens Cantin, and types one prompt:
"make a funny video"
A minute later, the AI generates the whole thing on its own: animation, voiceover, subtitles, editing, and music.
While most people are still manually cutting timelines in CapCut, she just hits "generate" over and over.
Then she uploads the clips to TikTok and YouTube Shorts.
The first video gets 618 views. The second — 214. The third blows up to 1.7 million views in 48 hours.
Her phone starts exploding with notifications in the middle of her shift at the coffee shop.
A week later, the channel was bringing in more money than her job behind the counter.
The weirdest part — viewers don't even realize they're watching AI content.
In the comments, people argue about the "character's charisma," ask for new episodes, and quote jokes that no human ever wrote.
Just a couple of years ago, people had to spend months learning animation, editing, and voice acting to make anything close to this.
Now it's one student, an old laptop, and a "generate" button.
Andrey Superior@andreysuperior
English

The biggest insight from the article about making money with Claude in 2026:
The biggest checks don’t go to people saying:
“I do AI.”
They go to people saying:
“I built an agent that saves lawyers 4 hours a day.”
People don’t buy “AI.”
They buy:
- contract analysis in 5 minutes
- competitor research in an hour
- a landing page overnight
- 50 leads without manual outreach
- content without a team
That’s why one person with Claude can now comfortably charge up to $25,000 per project.
The path is almost always the same:
- Month 1-2 - one service
- 3-4 - raise prices
- 5-6 - turn it into a product, agency, or system
The most interesting part of the article is where people are already making money:
- AI freelancing — $2K–15K/month
- AI consulting — $3K–25K per project
- AI agencies — $10K–100K/month
- Custom AI agents — $5K–50K per deployment
- AI products — almost no ceiling
But there’s one problem...
аlmost nobody explains the most important part for beginners:
How to get the first cashflow.
The video below is exactly about that.
Most people will just bookmark this post.
A few people will open Claude today.
And those are the people who’ll land their first major client within 90 days.
Khairallah AL-Awady@eng_khairallah1
English

Anthropic and Google pay $500,000+ a year for engineers who understand the math behind modern AI - Andrew Ng taught this at Stanford for free
most people building AI products today don't even know what's underneath them
> every time Spotify knows what you want to hear next - that's this math
> every time a bank stops a fraudster in milliseconds -that's this math
> every time a bank stops a fraudster in milliseconds - that's this math every th
time generative AI "guesses" the most probable outcome - that's this mathe results
bookmark & watch today before you spend another $2,000 on a course that covers the same thing
Anatoli Kopadze@AnatoliKopadze
English

Jane Street pays $750k/year for engineers who build AI trading systems from scratch.
This 15-minute talk by Jane Street senior engineer shows the exact system behind a $750k/year skillset.
here's what he covers:
• why generic AI fails at a top trading firm
• custom models trained on their own trading code
• real-time grading of every AI-written change
• one AI engine across the whole trading desk
most retail traders are still using generic AI tools - while the top hedge funds train their own models
Watch this video then read article below
Roan@RohOnChain
English

A guy spent one weekend and built his first 10 AI automations. Without writing a single line of code.
Four months later he had 45 of them. They save him 20 hours a week.
Stack: Claude + Cowork + MCP servers + plain English.
The strongest ones from the article:
→ Content
Every morning at 7am Claude scans trends and drops the top 10 topics. No more 30 minutes of scrolling. — 10-minute voice memo → 2000-word article ready to publish — Any article → an X thread with hook and CTA, automated
→ Email and comms
Team Slack channel → weekly one-pager for him. He stopped reading Slack entirely. — Before every meeting Claude builds a brief: LinkedIn, past threads, docs — all on one page — Auto follow-ups on emails that went silent
→ Research and analysis
Weekly intel report on 5 competitors: news, pricing, product moves — Any contract → a plain-English list of obligations and risks
→ Business
Any LinkedIn profile before a meeting → how to talk to this person (style, topics to avoid, common ground) — Every evening: voice dump → tasks, ideas, follow-ups already filed in the right folders
Every automation took 30 minutes to set up. Each saves him 2-4 hours a week. Forever.
Bookmark this — it actually works.
Khairallah AL-Awady@eng_khairallah1
English

$68,400 in one month with AI OF girls that don’t exist.
No real photos.
No human chatting.
No face on camera.
Just one guy running multiple AI models that feels completely real to subscribers.
How it works:
Consistent AI characters with unique personalities, kinks & texting styles
Fully automated DMs, voice notes & custom content
On-demand photo & video generation with perfect face/body consistency
Results:
800–2,200 paying subs per girl
One whale dropped $4,700 in a month
~$680/month compute cost per model
$51K net profit after OF cut
The girls reply naturally, remember conversations, and build real emotional connections.
Most subscribers have zero idea they’re talking to AI.
Demand is massive. Tech is finally good enough.
Running from one laptop.
Follow me for more
Frogify@0xFrogify
English

2 OpenAI engineers just gave a masterclass on how to build and ship apps using Codex
they spent 16 mins on how Codex turns one person into an entire software engineering team
for free. better than any $500 dollar AI course.
[here is what it covers]
>Codex is not “AI autocomplete”
>it is an agent harness for software work it can review code like a senior engineer
>it can split work across multiple sub-agents it can run workflows while you focus on higher-level decisions
>meanwhile the people who understand agents are building systems where AI researches, writes, tests,
the important part is not that Codex can write code everyone knows that already
[the important part is that Codex is becoming a workflow layer around software engineering]
most people are still using AI like a better Stack Overflow
the people who learn Codex properly will use it like a junior team, reviewer, researcher, tester, and automation engine
same subscription completely different outcome
full breakdown in the article below
Avid@Av1dlive
English

@shabnam_774 the funniest thing about AI right now is that half the internet is selling “secret prompts”.
English

Instead of watching Netflix tonight, watch this Stanford lecture.
You’ll learn more about how ChatGPT and Claude are built than most engineers learn in years.
Stanford released it for free.
Save this.
Shabnam Parveen@shabnam_774
English

Anthropic engineer showed how one person can run 5 AI agents, that code, test, review, and deploy at the same time.
In 30 minutes they built the whole thing live in one session.
Here's what they cover:
> when to use one agent vs a full team
> how to split work so agents don't step on each other > the exact framework for deciding what each agent handles
that's exactly why, I put together a guide on building agent teams that actually work.
full guide in the article below 👇
rody@0x_rody
English

someone just handed you a $40k/month team for free
these 50 Claude prompts took months to build - and they're copy-paste ready right now
worth more than any $200 prompt engineering course
here's what's inside:
- content creation, SEO, video scripts (prompts 1–10)
- full business strategy + pitch decks (prompts 11–20)
- code review, refactoring, CI/CD pipelines (prompts 21–30)
- market research, risk analysis, decisions (prompts 31–40)
- daily tasks, habits, negotiation, life decisions (prompts 41–50)
most people will bookmark this and never open it again
the ones who copy 3 prompts into their workflow
today will be operating at a different level by the end of the month
watch the video below first, then save the article
Khairallah AL-Awady@eng_khairallah1
English

i spent 3 months building the most optimized claude code setup possible
it was running worse than day one
23 plugins, 8 skills, 5 MCP servers, 3 custom frameworks, a context tool i found on twitter at 2am
every session opened with 40% of context window already burned
hitting limits before i typed a real prompt
deleted almost everything. kept 2
first session after the purge felt like switching to a better model
didn't switch models
just stopped suffocating it with "improvements"
your context window isn't full because of your work
it's full because of everything you added to help with the work
Bookmark it or continue using LLM like normie
the best setup is almost nothing
regent0x@regent0x_
English

THIS MAN HASN'T TOUCHED HIS TASK MANAGER IN A WEEK AND HIS ENTIRE LIFE IS STILL ON TRACK.
Here is exactly what he did differently.
Every night before bed he dumps everything into one folder in Obsidian called CAPTURE.
Every idea. Every task. Every thing that happened. Every thing that needs to happen.
No filing decisions. No organization. No thinking about where anything goes.
Just dump and close the laptop.
At 8PM Claude runs automatically.
It reads every file in the CAPTURE folder.
It sorts every note into the right place.
It flags every stalled project.
It assigns priority to every task.
It surfaces every open loop from the previous day.
By 6AM the morning briefing is sitting in his Obsidian vault.
He opens his laptop knowing exactly what matters before he sees a single notification.
His coworkers are still living in their inboxes.
Still spending the first 45 minutes of every day figuring out what they are supposed to be working on.
Still maintaining task managers that are three days out of date.
Still losing ideas because the capture friction was too high in the moment.
The only difference between his system and theirs is one folder called CAPTURE and Obsidian plus Claude running at 8PM every night.
Every idea goes in.
Claude files everything.
Nothing requires a decision at capture time.
Nothing gets lost.
The system runs whether he touches it or not.
That is not a productivity hack.
That is a different relationship with how your work gets managed.
Bookmark this and build it tonight.
Follow @cyrilXBT for the exact Obsidian plus Claude setup that makes this possible.
CyrilXBT@cyrilXBT
English

@antpalkin meanwhile the bot already traded 40,000 times while you were waiting for HR to reply
English

> you sent 380 applications
> got 2 interviews, 0 offers
> meanwhile a Brazilian dropout moved into his parents' garage
> he can't code. he can't trade.
> he rented VPS for $10 a month and Claude Opus, told it to figure out Polymarket
> 40,266 trades later: $794,000 profit
> every night the bot grades itself and rewrites its own strategy
> it's smarter today than the day he launched it
> you spent that same year "building your resume"
> the market never asked for your resume
> it asked who's running an agent and who isn't
> he was running an agent
> you were refreshing LinkedIn
before sending your resume, try copying his trades and see the results: @cvxv666" target="_blank" rel="nofollow noopener">kreo.app/@cvxv666
cvxv666@antpalkin
A Brazilian college dropout moved into his parents garage and built a Polymarket trading bot using only open-source AI agents - it earned him $794,000 in 14 months. He didn't write a single line of code. Claude Opus wrote it all. The agent framework we're using is Hermes - open source, built by NousResearch (backed by Paradigm with $70M). His wallet: @bonereaper?via=cvxv666" target="_blank" rel="nofollow noopener">polymarket.com/@bonereaper?vi…
Wallet handle: Bonereaper. 40,266 trades. Total stack cost: $10/month. Here's how it actually works: The bot trades BTC 5-minute Up/Down markets on Polymarket. 288 windows per day. One trade every 81 seconds. The edge is Markov chain analysis. When BTC price enters a persistent directional state - math says the next bar continues up with probability p ≥ 0.87. The market doesn't know this. The market prices it based on emotion. That gap is the entire trade: Δ = p̂ − q ≥ ε → ENTER The stack: BRAIN - Claude Opus 4.7 via API. Reads signals, makes decisions, rewrites its own strategy nightly. BODY - Hermes Agent by NousResearch. Open-source. 100K+ GitHub stars. RUNNER - Hetzner VPS. $5.99/month. Runs 24/7. ALERTS - Telegram bot. Every trade pings your phone. Setup time: 30 minutes. No coding required. But the real trick is the nightly self-learning loop. Every midnight Opus reads the day's trade journal. Tags wins, losses, EV per Markov state. Then rewrites MIN_PROB and MIN_EDGE in the .env file. Yesterday it might have been MIN_PROB=0.87. Tomorrow 0.89. The week after maybe 0.91 if the regime tightens. The agent is measurably smarter after 50 trades. After 500 it's a different bot than the one you launched. You don't need to know how to trade. You don't need to know how to code. You need $10/month and 30 minutes. The bot does the rest. Claude does the thinking. You read Telegram reports in the morning, approve the next session, and go back to sleep. Save this if you want to dig in and understand it. Or just copy BoneReaper trades using this TG bot - his algorithm has been perfected over 14 months and literally has no equal: @cvxv666" target="_blank" rel="nofollow noopener">kreo.app/@cvxv666 English

@0xKn0wledge a year ago “ai assistant” meant chatbot
now it means something that negotiates appointments while you sleep
English

This guy hasn't picked up a phone in 6 weeks.
The video below - Bland, building AI phone operators for big business.
He built one for himself. At home. In a single evening.
His AI now books, negotiates, orders, calls clinics. 20 cents a minute.
A year from now, everyone will live like this.
Guide below.
0xKnowledge@0xKn0wledge
English

Most people using Claude are leaving 80% of its power on the table.
Not because they lack the skills.
Because they never learned to prompt.
Two people. Same model. Same task.
One gets a generic answer.
The other gets a strategy that wins a real meeting.
The difference is 200 words of context and knowing these 5 rules:
→ Context beats cleverness every time
→ Constraints make outputs sharper
→ Define what "done" looks like — Claude self-corrects
→ One ask at a time. Always.
→ Bad first response = information. Refine, don't restart.
This thread has 25 prompts built around this logic.
I've been using #15 (The Ruthless Editor) and #6 (The Decision Architect) weekly for months.
Save this. You'll want it when you're stuck at 11pm with a deadline.
↓ Which of these 25 would you use first? Drop it below
Defileo🔮@defileo
English

4 months ago, the dude bought Claude for $20 and asked: "how to make money?"
Today, his faceless channel makes **$11,700 per month**.
Claude writes scripts, ElevenLabs says, CapCut mounts.
He himself spends 5 hours a week and sleeps peacefully.
While you are looking for a "stable job" his AI prints money 24/7.
winkle.@w1nklerr
English

$100 → $10,000 in weeks?
I turned a simple BTC 5-min Up/Down bot on Polymarket into a self-learning money printer using Hermes AI.
Trading bots already made $60M+ profit in 2025–2026… and 77% came from this exact market.
Now even normies can build one.
Full guide → How to build a self-learning BTC trading agent with Hermes (step-by-step)
> No coding. Runs on autopilot.
> Learns and improves itself every night.
Who’s ready to stop gambling and start printing?
0xRicker@0xRicker
English
