Writing

Field notes

Building AI agents in public. What I wire up, what breaks, what saves me time.

Explainer

An agent is six layers. Most people build one.

The model is one layer out of six. Connectors, hands, memory, permissions and loops are the other five, and the last one is where most agents quietly fall apart. With a worked example from my own pipeline.

Read →
Explainer

Open source AI is not free. Here's what it actually costs.

The download is free. The GPUs are not. What self-hosting a 671B model really costs per month, why cheaper per token is not cheaper per job, and the four-level map for picking your stack as you grow.

Read →
Experiment

OpenAI made agent control physical. Here's how to copy it.

Codex Micro sold out in under a day at $230. A Stream Deck and an open-source repo get you most of the way there. The architecture, the India prices, and the one button to be careful with.

Read →
Experiment

Meta built an AI that predicts your brain

Not mind-reading. It predicts the brain activity a scanner would record while you watch a video. I pointed it at real ads, and found the three things its own config gets wrong.

Read →
Free skill

Stop saving reels. Teach them to your agent.

My agent turns tutorial reels into reusable skills: it watches the video, extracts the method, writes the file. The exact skill is inside, free to copy.

Read →
Explainer

Learn Claude Code or Codex, not the tool of the week

Chasing every new AI tool is a waste of time and money. Go deep on one CLI agent instead. Why Claude Code and Codex are so powerful, and how they work.

Read →
Explainer

What is MCP, and how does your data flow through it?

The Model Context Protocol in plain terms, with a step-by-step look at how one question becomes a real answer.

Read →
Setup

How I run my personal AI agent

The skills I actually use, the connectors I rely on, and the two dashboards everyone on Claude should build first.

Read →