Thoughts on code, tools, and the things I'm building.
A tiny tool that embeds a terminal into any app. Smaller than it sounds, more useful than you'd expect.
A tool that scrapes a site and generates a working command-line interface for it. Useful in ways that aren't immediately obvious.
The ecosystem around Claude Code is growing fast. Here are five repos that are actually worth your time.
Organizing AI agent systems using the filesystem as the architecture. Simpler than it sounds, and that's the point.
A heads-up display that surfaces what Claude Code is actually doing while it works. Useful for debugging, useful for trust.
Someone built a distributed SQL database in a month, using AI as a pair programmer for the hard parts. The interesting thing isn't the database.
Despite every chat UI, the terminal remains the most powerful interface for AI workflows — why CLI-first AI tools win for developers.
The creator of Terraform and Vagrant writes honestly about how his relationship with AI tools has changed. Worth reading for the nuance.
Anthropic's engineering team tried building a C compiler using Claude as the primary implementor. The post is more interesting than the result.
Fine-tuning language models has historically required serious infrastructure and ML expertise. Unsloth Studio is trying to change that.
What actually goes wrong when you orchestrate multiple AI agents together — the failures, fixes, and patterns that scale.
Python Speed's guide to building and optimizing Rust extensions. The setup is less scary than it looks, and the results are worth it.
AI code review is only useful if you set it up right — what to automate, what to keep manual, and how to avoid false confidence.
Using local embeddings, vector search, and RAG to build a searchable personal knowledge base from your notes, bookmarks, and articles.
Starting the blog. What I'll be writing about and how this site works.
What companies are actually hiring for in AI development — skills that matter, skills that don't, and how to position yourself.
Most free courses waste your time. Here are the ones that don't — focused on AI, systems, and developer fundamentals.
AI-generated UIs are getting better. What's production-ready, what still needs a human touch, and how to use AI for UI scaffolding.
Skales has a lizard mascot. Why personality in AI products matters and how to build it intentionally.
After a year of daily AI tool use, here's what actually stayed in my workflow and what got dropped.
A stack of local AI tools that run entirely on your machine — Ollama, LM Studio, local embeddings, local vector DBs, and why local matters.
Zero-shot prompting in pipelines — when it works, when it fails, and how to structure prompts for production use.
A step-by-step approach to building a Claude Code plugin from scratch — skills, commands, and hooks.
Eight websites that actually improve how developers work — tools, references, and communities worth bookmarking.
The case for hands-on, project-first learning over tutorials and courses — and how to structure it.
A Claude Code plugin that handles tasks previously requiring multiple developers — and what it says about AI augmenting the solo developer.
The emerging market for custom AI agents — how developers can build, package, and sell them, and what buyers actually want.
DeepSeek, Qwen, and the wave of Chinese AI models — what's real, what's hype, and what it means for developers choosing models.
Seven GitHub repos that are genuinely useful for developers building AI-powered products — not just stars, but tools you'll actually use.
The ecosystem of 380+ Claude Code skills — how to find, install, and build them, and how the plugin model works.
Getting the most out of Claude Code by treating it as a full development environment, not just a coding assistant.
A specific tool from Google addresses a real limitation in Claude Code workflows — here's the problem and the fix.
CLI-Anything wraps any software in a CLI interface for agents — what agent-native software means and why the CLI is the natural interface for AI.
Impeccable is a tool that gives AI models proper design capabilities — here's the problem it's solving and why AI design has been bad.
How to run Claude Code locally for free and privately — steps, tools, and honest trade-offs.
Autoresearch runs AI agents to conduct deep research autonomously — what this means for developers building knowledge-heavy products.
Siftly uses local AI to categorize your Twitter bookmarks — and it's a good example of why local-first AI tools matter.
Paperclip is an open-source orchestration platform for AI pipelines — here's how it compares to LangChain and when to reach for it.
What the Anthropic Agent SDK gives you that the raw API doesn't — walking through a real agent loop.
Three mental models for understanding Retrieval-Augmented Generation — and when to reach for each one.
Fabric is a CLI tool that wraps AI prompts — and for developers, it's a better fit than the ChatGPT web UI.