What is an agent harness?
A first look at what an agent harness is, why the same model behaves so differently inside one, and why it matters more than the model itself.
A first look at what an agent harness is, why the same model behaves so differently inside one, and why it matters more than the model itself.


Eight patterns stripped of AI context. Where my approach converges and diverges. And the deepest takeaway: Claude Code bets on the model for decisions most systems handle structurally.

Four extension mechanisms, one architecture. How MCP servers, plugins, skills, and hooks all resolve to the same three primitives, and why deferred tool loading saves context.

76 lines of code that let tools execute before the AI finishes thinking. How Claude Code's streaming architecture overlaps model generation with tool execution

"What happens when things break. Exponential backoff with jitter, cascade protection that treats retry traffic as a shared resource, and the risks of resuming after an unclean shutdown