I've always joked that I work in the "ideas department."

I've always joked that I work in the "ideas department."

Big concepts, ambitious plans, whiteboards full of architecture. Then reality hits: the gap between idea and working system is enormous, and there are only so many hours after the kids are in bed.

AI closed that gap for me.

The shift wasn't about automating the boring stuff. It was about learning to give an agent a proper briefing — context, constraints, conventions — then designing the system together rather than just prompting at it.

In the last month that's looked like: a portable AI toolchain synced across every machine I own, MCP servers for Cloudflare and Cisco circuit registries, GitOps pipelines for lab and network config, Splunk on Proxmox, a full Packer → Terraform → Ansible → GitLab CI build chain, and a lifecycle management app for tracking Cisco hardware EOL dates.

A month ago most of those were ideas on a list.

The skill I think matters most right now isn't prompt engineering or AI literacy. It's engineering judgment — knowing how to decompose a concept, brief an agent, and recognise when the output is actually right.

The ideas department finally has a delivery team.

#AIEngineering #AgenticAI #MCP #DevOps #ClaudeCode