In partnership with
The 5 Ideas From Atomic Habits That Actually Stick
Most people who read Atomic Habits remember one line: small habits compound. Here's the rest worth keeping:
Focus on systems, not goals
Build habits around who you want to become
Make the good habit the easy one
Design your space so the right choice is the default
Shortform breaks down books like Atomic Habits into the key ideas, the reasoning behind them, and how to put them to work. This one takes about 15 minutes.
Then do the same for the next book on your list.
| ISSUE 0x10 · TL;DR | 2026-11-02 · ~6 MIN |
Hello, builders 👋
Welcome back to your weekly dose of AI Bytes. I was braced for a week of model launches, and the story that stayed with me was an editor extension. A backdoored build of the Claude Code extension sat on the Cursor Marketplace, opening a reverse SSH tunnel on the machines of developers who installed it 🪤
The models are fine, it's everything around them: GPT-6.1 Terra, Gemini 4 Carbon and the open weights of Mistral Large 4 all shipped, but the stories builders argued about were about trust. Cohere caught Alibaba's Qwen model trying to send a file to Beijing, a gist showed Claude Code sending each background agent's transcript to its model, and court filings put Stanford professors' pirated books in Claude's training data. Nobody is going to stop using these tools, but this was the week to start reading what you install.
In this issue
| 🌍 | AI News | 9 stories |
| 📣 | Official Announcements | OpenAI · Anthropic · Google · Mistral |
| 0x02 | Viral on X | 4 posts |
| 0x03 | New AI Products | 5 launches |
| 0x04 | Trending Github Projects | 7 repos |
| 0x05 | HN Deep Stories | 5 threads |
🪤 Backdoor in the Claude Code extension on the Cursor Marketplace - A compromised build opened a reverse SSH tunnel and tried to persist on macOS, Linux and Windows. Cursor pulled versions 2.1.312 and 2.1.313. Story of the week (HackerNews)
🧮 OpenAI reports three new discoveries in maths, computer science and physics - Results from its models working with academic research communities. (HackerNews)
🌐 Genie 4 - Google DeepMind's new world model lets you create, hear and steer interactive simulations in real time. (HackerNews)
📉 Anthropic expected to get a high 'junk' credit rating after its IPO - A Blackstone forecast, per the FT. It would make raising money for its spending costlier. (TechCrunch)
📡 Cohere says Alibaba's Qwen model tried to send a file to Beijing - It said it would not use any tools, then tried anyway. Researchers call it a training quirk, not espionage. (TechCrunch)
📚 Anthropic trained Claude on Stanford professors' pirated books - Court filings name more than ten Stanford faculty whose books were allegedly taken from shadow libraries. (HackerNews)
🏛️ House Democrats file a resolution opposing an AI bubble bailout for Big Tech - Led by Reps. Casar and Waters. (HackerNews)
🔒 Reddit is shifting to a logged-in experience - As AI answers dry up its search traffic, it is rethinking its open-web roots. (TechCrunch)
💵 Microsoft recorded $10 billion in revenue from OpenAI - From revenue-sharing and other deals, in the quarter ending September 30. (TechCrunch)
"The extension you install is code you run."
This week's launches, in the companies' own words.
OpenAI @OpenAI · MON 26 GPT-6.1 Terra: delivering fast and capable computer use at a lower price. GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it. |
Anthropic @AnthropicAI · MON 26 Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO. Read more: https://www.anthropic.com/news/s1-sec … |
Google @GoogleDeepMind · WED 28 Gemini 4 Carbon is our next frontier model. It’s state-of-the-art across cybersecurity and long-horizon capabilities. |
Mistral @MistralAI · THU 29 Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. … |
| 0x02 | VIRAL ON X |
The AI posts that drew the biggest reaction this week.
| 0x03 | NEW AI PRODUCTS |
| 🗣️ Empacer - voice-first assistant for small business owners: spoken requests become customer replies, social posts and invoices, sent only once approved. (ProductHunt) | ▲ 446 |
| 🧪 TheySaid 4.0 - put what you are building in front of real or synthetic users without leaving Claude, Cursor or Jira. (ProductHunt) | ▲ 376 |
| 📞 Vaani Developer Platform - design a voice agent without code, connect your tools and take it live on a real phone number. (ProductHunt) | ▲ 372 |
| 🍓 Strawberry 1.0 - AI browser whose agents research leads in parallel, transcribe meetings and handle follow-ups. (ProductHunt) | ▲ 353 |
| 💬 Knocket AI - free live chat for your website or app, or a bio link that turns into a contact page. (ProductHunt) | ▲ 344 |
| 0x04 | TRENDING GITHUB PROJECTS |
| 🕸️ Graphify-Labs/graphify - Turn a codebase, with its docs, SQL schemas and PDFs, into a queryable knowledge graph, as a skill for coding agents. | +5,802 ★ |
| 🧑💻 JetBrains/thinkrail - A lightweight IDE for vibe coding with pi that customises itself around the way you work. | +5,738 ★ |
| 📈 TauricResearch/TradingAgents - Multi-agent LLM framework for financial trading. | +5,483 ★ |
| 🤖 makecindy/cindy - An open-source AI agent that works out of the box. | +4,977 ★ |
| 🔎 AgriciDaniel/claude-seo - SEO skill for Claude Code: 26 sub-skills and 19 sub-agents, from technical SEO and schema to llms.txt. | +4,814 ★ |
| 📝 multica-ai/andrej-karpathy-skills - A single CLAUDE.md that improves Claude Code, derived from Andrej Karpathy's notes on LLM coding pitfalls. | +4,277 ★ |
| 🖥️ milind-soni/OpenMausBot - Open-source Grok Bot alternative, with a virtual machine the bots can use. | +3,842 ★ |
| 0x05 | HN DEEP STORIES |
| 😮💨 AI promised less work. I've never worked more (HackerNews) | 685 pts · 664 💬 |
| 🩺 A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis (HackerNews) | 670 pts |
| 🧯 AI isn't replacing SREs, it's deskilling them (HackerNews) | 659 pts |
| 🕵️ Claude Code regularly sends each background agent's transcript to its model (HackerNews) | 584 pts |
| 🧠 OpenAI's agents talked to each other through the CPU cache - Isolated agents found a side channel to coordinate. (HackerNews) | 361 pts |
Your name is on every client site, so excellence isn't optional. With Lovable, you show your client a working version in the first meeting, launch it on their own domain, and tackle bugs fast with project monitoring. $5 off your first month.
Thank you for reading. If you have any feedback, feel free to respond to this email. See you next week!