Posts

Claude Code’s Tool Search implementation slashes token consumption by 95%, paving the way for AI agents that can handle thousands of tools without context bloat.

2026 research from UBC and Anthropic’s engineering team reveals a surprising trend: simplifying to a single AI agent with skills can boost performance by 50% over complex multi-agent systems.

Give AI a task before bed, wake up to 6 completed libraries! Learn how the Ralph Wiggum Loop plugin keeps Claude Code iterating and self-correcting 24/7 without rest.

Claude Code now works on web, mobile, and Slack—letting developers fix bugs on the subway, handle requests from coffee shops, and turn Slack threads into PRs.

Claude Code slash commands let you package repetitive prompts into one-click shortcuts. This guide covers everything from basic concepts to advanced features like parameter passing, bash execution, and file references.

Do you take screenshots all day? I discovered with CleanShot X that screenshots can be incredibly smooth: OCR text recognition, scrolling capture, quick editing, and these features upgraded my workflow.

Elon Musk predicts phones and apps will vanish in 5 years. Learn how AI agents and MCP (Model Context Protocol) are reshaping digital commerce, why Amazon is suing Perplexity, and what this means for the experience economy.

Anthropic’s Claude Skills enables AI to dynamically load specialized knowledge without consuming memory. Finally, someone’s solved the context window overload problem!

Master 10 advanced CLAUDE.md techniques, from context management to workflow automation, making your Claude Code smarter and more efficient

Deep dive into three A/B testing trigger methods: User-Level Triggering, Exposure Logging, and Counterfactual Logging. Learn from real industry cases at DoorDash and Spotify to solve experiment dilution problems effectively.

Introducing A/B testing trigger mechanisms that shows how precision targeting of affected users can cut sample size requirements in half. Essential reading for understanding why lots of experiments fail due to diluted effects and how to design more powerful tests.

When every day brings another AI product launch, it’s easy to become numb to innovation. Yet Comet, Cluely, and nao (Cursor for data team) reignite my excitement about technology’s future (even though I’ve never actually used them).

Tired of repeatedly explaining project architecture and coding styles? CLAUDE.md lets your AI coding partner truly remember your needs. Set once, use forever.

Tech giants are simultaneously investing in both GUI and CLI-based AI coding tools, creating a fascinating paradox in the development world. This post explores why companies like Anthropic, Google, and even Cursor are betting big on terminal-based AI coding solutions despite the success of visual interfaces.

VoiceInk’s $25 speech-to-text capabilities doubled my productivity when working with AI tools like ChatGPT and Claude. This comprehensive review covers why it outperforms expensive alternatives for Mac users seeking efficient voice-to-text solutions.

After weeks of testing Claude Code, this terminal AI tool has genuinely improved my development workflow. Here’s my honest assessment of what works, what doesn’t, and who should consider it.

DoubleML combines machine learning with econometrics to extract causal insights from observational data. This tutorial covers its core concepts, real-world applications, and how it overcomes traditional causal inference limitations.

Discover how causal inference unlocks insights when A/B testing isn’t possible. Data scientists can construct ‘parallel worlds’!

P-value < 0.05 ≠ You have 95% chance of being right

Learn the crucial differences between hyphens, en dashes, and em dashes with practical examples and typing shortcuts for macOS users.

The limitations of human intuition and the power of controlled experiments in modern product development.

I have written quite a lot in Chinese: haosquare.com. Now I am trying to write in English.