Latest

Same Skill, Different Model

A skill that works beautifully on Opus can quietly fall apart on Haiku, and the reverse happens too. What two 2026 skill studies and the earlier prompt-sensitivity papers say about why, with numbers, and how to find out where your own skill holds up.

ai research llm agents agentic-ai
15 min read · October 10, 2026

Act Slow Fast

September 3, 2026 · 14 min read

Give Your Agent a Sandbox

August 21, 2026 · 24 min read

Agents Are Jobs, Not Services

A resident agent idling at 2GB of RAM is nothing, expensively. Most of the agent work people...

ai llm agents
July 26, 2026 · 4 min read

Lessons from Building an Agent Harness for Local Models

The loop, the context juggling, the per-model prompt formatting, the tool calls. Here is what...

ai llm agents
June 24, 2026 · 18 min read

World-Building with WASM

A walkable browser world needs something to generate it, something to draw it, and something to make...

webassembly wasm rapier
June 8, 2026 · 7 min read

Controlled Chaos

I've a love hate relationship with rules. They give me a map, but they also draw the walls. This is...

life theory chaos
April 26, 2026 · 6 min read

Writing Your Own Claude Plugin and Shipping It to the World

Claude Code has a plugin system now. Build a plugin, push it to GitHub, list it on AgentHub, and...

ai claude-code plugins
April 20, 2026 · 19 min read

N Reviewers Walk Into a PR...

How many people should review this PR? I built a GitHub Action that answers that question...

github open-source devtools
April 2, 2026 · 6 min read

Showing 9 of 62 articles

ai wasm thought research cyberspace build