Insights
What we learn building AI inside regulated organisations, written up while it is still inconvenient. No vendor press releases.
Everything Worth Knowing About Working in ClaudeClaude is not one thing. It is six decisions stacked on top of each other: where you open it, which model you pick, how hard it thinks, how you work, what you get back, and what it is plugged into. This field guide lays out all six on a single page.PDF · 55 KB
The Claude Working ChecklistMost people use Claude. Far fewer get real work out of it. This one-page checklist covers sixty-four habits across setup, projects, skills, prompting, connectors, Cowork, Claude Code, writing and cost control. Print it, tick it, keep it by the desk.PDF · 84 KB
Nine Claude Skills, in the Order You Build ThemFrom a messy idea to a prompt that holds up. Nine skill files, each invoked with a slash command, laid out as a single pipeline you can follow or hand to a colleague.PDF · 87 KB
Claude Cheat SheetWhich Claude Code extension to reach for, how CLAUDE.md, skills, subagents and hooks differ, and what each one costs you in context. One page, verified against the Claude Code documentation.PDF · 117 KB
MCP in Five QuestionsThe Model Context Protocol, drawn rather than described. What it is for, who the pieces are, what a server hands over, what happens on the wire, and what will trip you up. Six pages for beginners.PDF · 367 KB
The AI DictionarySixty terms you actually need, each defined in one sentence, with the common misreading named rather than skirted. The vocabulary you will meet in vendor decks, architecture reviews and board papers this quarter.PDF · 142 KBOne email a month. Nothing else.
New insights, new resources, and one thing we got wrong recently and how we fixed it.
No sharing, no selling.