Topic
AI
Why My Best Productivity Tools Are the Ones Nobody Else Can Use
Custom software used to be a luxury reserved for people with a budget. With AI, a tool shaped exactly like my workflow costs less than adapting to a generic one — which quietly changed what I build, what I still buy, and where the line between the two sits.
Obsidian' Doesn't Exist — It's Four Layers
What we call 'Obsidian' is really separate, replaceable layers — storage, format, UI, agent. Counting how many you can swap without touching the others is the real test of whether your notes app owns you.
LM Studio Bionic: open models finally get their own agent
LM Studio just turned from a model runner into a full AI agent for open models. Here's what Bionic is, how it works, and what it says about where local AI is heading.
The Simplest Way to Save AI Tokens: Make Your Agent Talk Less
Every AI session has a token economy: prompt, context, agent work, and the answer. I break down where the tokens really go, what Caveman actually saves, and why the trick works far beyond coding agents.
GPT-5.6 isn't the real news — the new ChatGPT app is
OpenAI shipped a new model family, but the real shift is the app: ChatGPT now bundles Chat, Work, and Codex — and picking the right space matters more than picking the right model.
How I built an Obsidian vault no single AI tool can hold hostage
CLAUDE.md, AGENTS.md, a one-page identity file, a skills folder, and a set of maps to my key notes — the whole AI layer on top of my Obsidian vault is plain markdown I own, not configuration locked inside any single tool.
I finally ran an LLM on my own machine — and it changed how I think about owning my AI
I own my notes. I own my blog. So why does my data leave the building every time I prompt an AI? Running a model locally is the next chapter of the same old conviction.
How I made Claude the librarian of my Obsidian PKM — the complete process
The exact pipeline I run to turn web clippings and highlights into atomic, linked permanent notes — Claude does the structuring, I stay the curator.
How I use AI in my Obsidian vault without losing the plot
I kept AI out of my Obsidian vault for years. MCP changed the equation — not by making the vault smarter, but by giving AI a specific job with a clear boundary and a review step.
Coding agents don't need more power. They need rules.
A simple CLAUDE.md file based on Andrej Karpathy's observations about coding agents became wildly popular because it solves a real problem: AI agents are powerful, but not disciplined by default.
Stop using AI coding agents like autocomplete
AI coding agents become more useful when I stop treating them as smarter autocomplete and start treating them as delegated workers with clear inputs, boundaries, and verification.
Claude Code vs Codex: two tools, two modes of thinking
I stopped asking whether Claude Code or Codex is better. The useful question is simpler: am I exploring a problem, or am I delegating a bounded task?
How I Built an iOS App with Vibe Coding — From Blank Page to App Store
The story of using AI tools like Cursor, ChatGPT, and V0 to build and publish an iOS subscription tracker app from idea to App Store.
How to supercharge your writing process with AI and chatGPT
How to use AI and ChatGPT to speed up your writing process without losing clarity or structure.