Claude Safety Is Out of Hand, and That Probably Isn't a Bad Thing
A refusal wall in the middle of ordinary product work, why the brakes should exist anyway, and the practical boundary that keeps security-adjacent tooling out of everyday context.
Newsletter
Practical AI insights for Australian businesses, delivered weekly. No fluff. Just actionable advice from the field.
Follow Hayden on LinkedInPractical AI insights you can implement today. No fluff, just actionable advice.
A refusal wall in the middle of ordinary product work, why the brakes should exist anyway, and the practical boundary that keeps security-adjacent tooling out of everyday context.
One day your coding agents are brilliant, the next they are useless. It is not random: they lose signal in noise. Treat AI behaviour like a project and version control it.
Chezmoi solved two core problems I’ve been struggling with for a long time: I couldn’t see my AI, and I didn’t know what changed. Here’s how tracking config files in git gave me observability for my AI’s behavior.
The Ralph Loop is a forever-running agent plugin from Anthropic. But without backpressure, more iterations can actually widen the scope of regressions. Here is why TDD is the answer.
AI slop is everywhere. Backpressure, a flow control mechanism borrowed from fluid dynamics and software systems, is how you stop it reaching you.