AI

Shipping NanoClaw Skills as npm Packages

NanoClaw wants capabilities as skills, not core PRs. The maintainers are busy, and skill PRs sit in a long queue. Here is how I ship npm packages that install skills into my fork, and keep install and uninstall idempotent.

Launching NanoClaw Agents on External Hosts

I use NanoClaw as my agent orchestration layer, but not only on local Docker. Here is why I implemented mailbox IO and agent container abstractions, and how these unlock Fly Machines, native process spawn, and the next agent container runtime without forking NanoClaw core.

Extending NanoClaw Without Forking the Host

NanoClaw keeps the trunk small on purpose. Here's how I add a local chat desk, live activity traces, and an admin API as installable skills, and why hosthooks is the extension layer underneath them.

Meet My NanoClaw Agent Team

Four specialized AI agents review PRs, file bugs, and ship code across BeakerStack, Skein, and the NanoClaw packages. Here is who they are, and the GitHub profiles where you can see the work.

Why I Chose NanoClaw Instead of OpenClaw

I wanted agents with real tools and a blast radius I could live with. NanoClaw puts isolation in the kernel, not in an allowlist, and that is why I run it.

Vibe Coding vs Agentic Software Engineering

Vibe coding and agentic software engineering use the same AI. The difference is the scaffolding around it, and with the right controls you can let non-technical teammates build safely.

The 3,000-Line-a-Day Thought Experiment

If a new engineer shipped 3,000 lines on day one, could your team validate it? Why agentic programming makes this normal, and how to be ready for it.

Measuring Velocity in the Age of Agentic Programming

AI agents are breaking traditional velocity metrics. Why output is now the cheap part, and what to measure instead in the age of agentic programming.

Why Observability Makes Agentic Programming More Efficient

Give an AI agent access to your observability platform through MCP and it can investigate a production bug faster than most engineers. Observability is becoming the substrate your agents debug with.

100% Test Coverage Should Be the New Bare Minimum

For years, full test coverage was treated as impractical. AI-assisted coding changed the economics. Why 100% is now a sensible floor, and what it does and does not buy you.

The Testing Pyramid: AI Just Removed Your Last Excuse

Teams have always known they should have automated tests. The excuse was that writing them takes too long. AI-assisted coding just took that excuse away, right when you need tests most.