🤖 Clawdbot goes viral + Cursor reveals how they built their AI agent

Clawdbot happened this week. If you missed it: an Austrian developer built an open-source AI assistant that runs locally, connects to WhatsApp/Telegram/Slack, and actually does things instead of just chatting. It hit 60,000 GitHub stars in three days — one of the fastest-growing open-source projects ever. I've been playing with it and honestly, this is what I wanted AI assistants to be. It's Claude with hands. Your data stays on your machine, it learns your preferences over time, and the skills ecosystem means you can extend it to do basically anything. Anthropic even asked them to rename it (trademark stuff), so it's now called Moltbot. Here's what else happened while we were all setting up our new AI butlers.

A private inbox doesn’t have to come with a price tag—or a catch. Proton Mail’s free plan gives you the privacy and security you expect, without selling your data or showing you ads.

Built by scientists and privacy advocates, Proton Mail uses end-to-end encryption to keep your conversations secure. No scanning. No targeting. No creepy promotions.

With Proton, you’re not the product — you’re in control.

Start for free. Upgrade anytime. Stay private always.

Get free private email

Clawdbot — Open Source Personal AI Assistant (3 min)
The project that broke GitHub's trending algorithm. 61k+ stars, multi-channel messaging (WhatsApp, Telegram, Slack, Discord, iMessage), persistent memory, and 50+ skill integrations. This is the "Claude with hands" that everyone's been waiting for. Runs locally, your data stays yours.

How Cursor Shipped its Coding Agent to Production (13 min)
ByteByteGo breaks down the technical details and challenges Cursor faced building Composer and their autonomous coding agent. Real engineering insights from one of the most-used AI coding tools.

Tips for Getting Coding Agents to Write Good Python Tests (5 min)
Simon Willison shares practical advice for getting AI coding agents to produce quality tests. If you've struggled with AI-generated tests that miss edge cases or don't actually test anything meaningful, this has actionable tips.

There is an AI Code Review Bubble (6 min)
Greptile argues that the flood of AI code review tools are solving the wrong problem. Most focus on style and syntax when the real value would be catching logic errors and security issues. A useful reality check.

Clawdbot with Docker Model Runner (4 min)
Docker's guide to running Clawdbot with their Model Runner for a fully private setup. No data leaves your machine, predictable costs, and complete independence from AI service providers.

ChatGPT Containers Can Now Run Bash (5 min)
OpenAI quietly upgraded ChatGPT's code execution environment. You can now run bash, install packages with pip/npm, and download files. Simon explains why this matters and what you can do with it.

Why NaN !== NaN Makes Perfect Sense (4 min)
A clear explanation of why JavaScript's seemingly broken NaN behavior is actually correct according to IEEE 754. If you've ever been confused by this, now you'll understand why it works this way.

Introduction to PostgreSQL Indexes (6 min)
Solid fundamentals piece on Postgres indexing. Covers B-tree, hash, GIN, and GiST indexes with practical guidance on when to use each. Good reference material.

Are Juniors Screwed? Getting a Job in a Post-AI World (10 min)
Theo's honest take on the current job market for junior developers. The market is rough, but he outlines a path through. Worth watching if you're starting out or mentoring someone who is.

Stanford PhD and Ex-Citadel Quant Shares His Experience (30 min)
Insights on breaking into AI research and quant finance, whether you need a PhD, and what the actual work is like. Useful if you're considering these paths or curious about what happens at the intersection.

Meta Senior Staff Engineer's Honest Demotion Story (30 min)
Raw story about being demoted from IC7, what level actually made them happiest, and navigating Meta vs Google culture. Refreshingly honest about the reality of senior+ engineering careers.

We're Creating a Knowledge Collapse (12 min)
Thoughtful piece on how AI-generated content might be eroding the knowledge base it was trained on. The concern: we're consuming AI output and producing more AI-trained content in a loop that degrades quality.

Qwen3-Max-Thinking
Alibaba's new reasoning model. Another entry in the "thinking" model category alongside o1 and DeepSeek. Worth watching if you're following the reasoning model race.

Kimi K2.5 — Open-Source Visual Agentic Model
New open-source visual SOTA model with agentic capabilities. If you're building vision-based agents or need strong multimodal reasoning, this is worth evaluating.

zerobrew — 5-20x Faster Homebrew Alternative
Rust-based drop-in Homebrew replacement. If you've been frustrated with Homebrew's speed, this claims significant improvements. Risky to switch your package manager, but interesting.

TanStack AI Library (10 min)
Jack Herrington walks through TanStack's new AI library. If you use TanStack Query/Router and want to integrate AI features, this shows you how to get started.

mmdr — Native Rust Mermaid Renderer
500-1000x faster than mermaid-cli. If you're generating diagrams in CI/CD pipelines or documentation builds, this could dramatically speed up your workflow.

Vercel Skills v1.1.1
Vercel's agent skills ecosystem gets interactive discovery and is now open source. If you're building AI agents, this provides a standardized way to define and share capabilities.

Your Tools Are Now Interactive in Claude (10 min)
Anthropic's official video on Claude's interactive tool use. Manage Asana projects, draft messages, and more — directly within Claude. Shows where AI assistants are heading.

The Web's Next Form: MCP UI with Kent C. Dodds (10 min)
Syntax sits down with Kent C. Dodds to discuss MCP, context engineering, and what it really means for web development. Good discussion on the emerging agent-driven interface paradigm.

AI Mistakes You're Probably Making (10 min)
Theo on common mistakes people make when using AI for coding. It's easy to tell the AI what to do, but getting good results requires more nuance than "just ask it."

Is GLM 4.7 Better Than Opus? (10 min)
Web Dev Cody tests the new Chinese AI model against Claude Opus. Part of the ongoing global competition in LLM development. Useful if you're evaluating models.

Peering Into Claude's Soul (10 min)
Theo reacts to Anthropic's Claude Soul document. If you found the original document weird, this goes deeper. Warning: might give you feelings about AI.

I Don't Really Use Libraries Anymore (10 min)
Theo explores how AI has changed his relationship with npm packages. Provocative take on whether we need as many dependencies when AI can generate code on demand.

That's it for this week. Reply if you have something I should include next time.

Until next week,
Niall