Articles
Notes from the field
Writing on keeping systems reliable, on observability (knowing what a system is doing), and on working with AI agents: the AI tools that carry out tasks on their own.
Claude Isn't Ignoring Your CLAUDE.md. Your Old Rule Is Still in There.
When your agent follows a rule you already changed, the old version is often still sitting in your `CLAUDE.md` or the agent's memory. A smarter model mostly works around it, and you pay for that in time and tokens. Here's how old rules pile up, the one line that fixed it in my own testing, and where to look in your setup this week.
A Status Line for Claude Code That Shows What Each Prompt Costs You
Most status lines show where your session is. Mine shows what each prompt just did to it: context added, cost added, and whether the cache held. Here's what I watch and the repo if you want a starting point.
An AI Agent Exposed 31 Secrets. Here's What I Changed.
A debugging command in a coding agent session printed 31 credentials. What I rotated, what I rebuilt, and the least-privilege rules I now work by.
Observability vs. Monitoring: A Practical Comparison
Monitoring is built around known failure modes. This works well for predictable problems. The limitation is that monitoring is only as good as the assumptions baked into it.
The Reliability Mindset: How Modern Engineering Really Works
Reliability is less about tools or processes and more about a mindset. It’s the discipline of designing for failure, observing reality, and constantly closing the gap between how we think things work and how they actually behave in production.