#ai
4 posts tagged ai. all posts
- Securing and monitoring an MCP server in production MCP turns a chat UI into a system that can take actions on your behalf. That changes the security and observability story. Here is the stack I run for netrecon.
- Writing a diagnostic LLM that doesn't lie How netrecon keeps an LLM-narrated diff grounded: schema-bound outputs, no snapshots in the prompt, and citation whitelisting that silently drops hallucinated field names.
- Applied AI for infra teams: patterns that actually work LLMs are good at narrow, read-only, well-evaluated tool calls. They are bad at long autonomous loops in production. Here is the shape of the systems that actually hold up.
- MCP for SRE: giving agents real tools, safely Model Context Protocol is the right primitive for on-call agents. It is also a footgun if you expose write tools without thinking. Here is a safer default.