Configuring MCP servers correctly — the parts that bite
The two-part declaration people get half right, why an API key is not an MCP credential, and where secrets should live so a prompt injection cannot read them.
Topic
AI & DevOps
Using AI tooling in a delivery pipeline without pretending it is magic — Claude, MCP and evaluation.
5 posts
Using Claude in DevOps without pretending it's magic
Where an LLM helps a delivery pipeline, where it is the wrong tool, and the four architectural choices that decide which of those two you end up with.
Claude Code for platform teams: a setup that survives contact
Project context, permission boundaries, hooks and subagents — the configuration that decides whether agentic coding helps a platform team or generates work for it.
Evaluating an AI feature without relying on vibes
Building an eval set that catches regressions, where LLM-as-judge is trustworthy and where it is not, and how to run it in CI without constant flakiness.
Cutting LLM cost in a pipeline without cutting quality
Prompt caching, model tiering, batching and effort control: four levers that cut spend on an LLM feature, in the order that returns most for least risk.