StarHaze

259 pull requests. 30 days. 100% written by Claude. The man who shipped them built Claude Code. He hasn't opened an IDE since November 2025. He runs the whole thing from his phone. Most people use Claude like a smarter search engine. The person who built the tool uses it like infrastructure. The full system is in the article.

He didn't replace one tool. He replaced the entire workflow. One YouTube channel. Claude Code handles research, scripting, scheduling, and analytics. What used to take a team of people with five separate tool subscriptions now runs on one model that gets better every week. Most people are still asking "which tool should I try first?" The people getting results are asking a different question: what does the whole stack look like if I build around one thing that actually compounds? That question has a different answer. And a very different monthly bill.

A bot made $438,000 on Polymarket in 30 days. Not a hedge fund. Not a trading desk with Bloomberg terminals. Someone built it with Claude Code. The strategy is public. The code was generated by the same model you can open right now. The mechanics are simple: the bot reads market data faster than any human, identifies mispriced probabilities, and places positions before the market corrects. It doesn't sleep. It doesn't second-guess itself. It doesn't have opinions about what should happen -only about what the data says is underpriced. This is what Polymarket looks like from the other side of the trade. The 84% of users finishing in the red are losing to something. This is one of the things they're losing to.

@0xCodez It’s hidden GEM

A senior Anthropic engineer just dropped 11-page PDF on "Loop Engineering" for agentic systems. The shift: you stop prompting the agent. You build the system that prompts it instead. Schedule → Discover → Build → Verify → Repeat Every loop runs one turn, five moves: • Discovery: it finds its own work - failing CI, open issues, recent commits - instead of being handed a list. • Handoff: each task gets an isolated git worktree so parallel agents don't collide. • Verification: a second agent, told to assume the code is broken, reviews the first. The "thing that can say no." • Persistence: results get written to disk, never left in a context window that gets flushed. • Scheduling: an automation wakes it on a timer. That's what makes it a loop. The key insight: an agent grading its own work always praises it. This 11-page PDF changed how I'm building agentic systems today. Read it now, then explore the article below.

Single-purpose tools charge you the same in year three as year one. They never get better. You just get used to them. Claude gets more useful the more context you give it. The workflow compounds. The subscription doesn't. Full breakdown — what I replaced and what I kept:

$141 a month. Five tools. Five subscriptions. Five tabs permanently open "just in case." I cancelled all of them. Claude replaced every single one for $20. The one that didn't work — and exactly why — is in the article:

$141 a month. Five tools. Five logins. Five single-purpose subscriptions that each did exactly one thing. I cancelled all of them. Gave myself 30 days with Claude only. The research tool went first. Then the writing assistant. Then the meeting notes app. The math was obvious from day one — but cost wasn't even the main finding. The main finding was that single-purpose tools don't compound. You pay them the same amount in year three as you did in year one, and they're exactly as useful as the day you signed up. Claude builds on itself. Every context file you save, every workflow you wire up, every prompt you refine — the system gets better. The subscription doesn't. $1,452 back per year. And a workflow I actually own. Full breakdown — what worked, what didn't, and the one tool I kept:

@RohOnChain Thanks to you, I created my first Loop. It's incredible.

this is f*cking dangerous build hedge fund using "loop engineering" that prints alpha 24/7 (Full Guide) if I had this a year ago, I would've built my hedge fund in a week instead of a year bookmark before someone takes it down

"Stop Prompting Claude. Start Loop Engineering." That's the title of a video that dropped 6 days ago. Not a research paper. Not a corporate announcement. A YouTube video. That phrase is now the headline of the conversation. A year ago the advice was learn to prompt better. Six months ago it was "build agents." The language has a new name now, and it moved faster than most people noticed. The difference between prompting and loop engineering is the same as the difference between driving somewhere and building a road. One gets you there once. The other gets everyone there, every time, without you in the car. Full system — all 5 components:

Someone made $10,025 in 24 hours using Claude. Not a trading bot. Not a faceless YouTube channel. Not passive income. He sat down, built a system, ran it, and closed the day with five figures. The part nobody talks about is how unremarkable the setup was. A few prompts. A loop that handled the repetitive work. The same Claude that 100 million people are using to write emails. The difference between people who make money with AI and people who watch videos about making money with AI is almost always the same thing — one of them stopped prompting and started building a workflow they could run again tomorrow. $10,025 in 24 hours. The math gets uncomfortable if you think about what your last 24 hours produced.

@shmidtqq I always find myself wondering whether I’ve strayed from the path

Head of Engineering for Claude Code at Anthropic. Fiona Fung: "After a while, it could start being a lonely experience - we were all just working with our agents so much." Claude now writes 80% of Anthropic's code. The team ships 8x more. An engineer, in Anthropic's own report: "On the days everything works, I can't help but think nothing I do matters." She names the fix in the episode. The dream was less typing - is this what you wanted?

8.9 million people just learned what a loop is. Almost none of them will build one. There's a word for that gap. It's not a knowledge problem.

The people who built AI stopped prompting it months ago. They run loops now. 5 components. Systems that read, check, fix, and ship — while they sleep. Most people are missing at least 3 of them. Here's the full setup:

@cyrilXBT Really appreciate you for this info

I genuinely don't understand why everyone isn't using this yet. Andrej Karpathy, OpenAI co-founder, posted a simple idea that went massively viral: Stop using AI to write code. Use it to build a second brain. You point Claude Code at a folder. Drop in any source: an article, a transcript, a PDF. Claude reads it, links it, files it into a living wiki of everything you know. It compounds like interest. The more you feed it, the smarter it gets. Here's the whole thing: 1) Install Obsidian 2) Create a vault 3) Open it in Claude Code 4) Paste Karpathy's wiki idea and tell Claude to build it 5) Claude makes three folders: - raw (for sources) - wiki (for its pages) - CLAUDE. md (that runs it) 6) Drop any source into raw and say: "ingest this" 7) Ask questions across everything, forever Five minutes to set up and you never start from a blank chat again. Full step by step guide below.

Boris Cherny — the man who built Claude Code — shipped 259 pull requests in 30 days. Every single line written by Claude. He hasn't opened an IDE since November 2025. That's not optimization. That's a completely different job description. Full system breakdown:

@ridark_eth I think it’s future of analytic

How a 24-year-old programmer from Portugal made $18,200 in a month on football betting He created an AI analyst that finds flaws in bookmakers' live lines in real time and delivers predictions with an 84% win rate. Costs: $0 (Used free APIs and Windsurf IDE) He launched a Python script that maps out match videos in real time: Top layer: A Computer Vision algorithm recognizes the positions of players from both teams (blue and pink dots) and the ball, instantly transferring them onto a 2D pitch layout. This allows the AI to track team formations and open spaces in high detail, things regular bettors completely miss. Bottom layer: Python code (written alongside the Windsurf AI assistant), where the SoccerPitchConfiguration class defines the field, penalty box, and center circle dimensions down to the centimeter for perfect player-distance calculations. The AI constantly correlates the real-time movement of players on the pitch with live bookmaker odds. The moment the algorithm detects that a team has pinned their opponent into a specific zone or exposed their flanks, while the bookmaker hasn't adjusted the odds yet, the script automatically fires a betting signal. First week: >Live bets placed: 142 > Won bets: 119 > Net profit: +$4,350 using a flat $50 stake The AI completely automated the entire cycle: Windsurf and Claude wrote the tracking code, and the algorithm autonomously parses live odds, calculates the mathematical expectation of value bets, generates player heatmaps, and spots hidden tactical anomalies. It runs 100% autonomously. Bookmark it and check out the article below 👇

CNBC sits down with the head of Claude Code at Anthropic's developer conference and asks him about the future of work. He doesn't give a keynote answer. He describes what he did that morning. Checked the loop's output, approved two PRs, pushed a feature — all before the interview. No IDE. No terminal. The agent ran overnight, the work was waiting. The part that lands is the framing. This isn't a prediction about what's coming. It's a description of what's already happening, inside the team that builds the tool. Most people watching this are still treating Claude as a chat window. The people who built it stopped doing that months ago.

The creator of Claude Code sits down with Sequoia at AI Ascent 2026 and says something that should make every developer uncomfortable. He hasn't written a single line of code in 2026. He ships dozens of pull requests a day — from his phone. His IDE has been closed since November. 259 PRs landed in one month, all written by Claude, none by him. The uncomfortable part isn't the number. It's what he says his job is now: writing loops. Not prompting Claude. Not reviewing code line by line. Designing the system that does the prompting for him, checks its own work, and keeps running after he puts the phone down. That's the whole video. 24 minutes of what "the job" looks like when the leverage point moves.