Notes from client work on applied AI, AI coding tools and search. Each piece starts from something I built or measured: the readiness assessment I run with new clients, the coding tools I use every day, the code a local 27B model shipped and the sources answer engines cite. Where a tool fell short, I say so. Newest first; the longer, ordered Claude Code guides sit in Guides.
How I run Claude Code on client projects, in reading order. Guide 00 asks when AI is the wrong tool; 01–03 cover CLAUDE.md, pricing and token costs, and choosing a model; 04–07 memory, sessions, skills and hooks, and MCP servers; 08–12 planning, parallel agents, quality gates, work beyond code and evals; 13–14 my open-source hook-rule plugin and transcription skill. Each one names what fails as well as what works.
Three ways I work with founders and small-business owners. As a fractional CTO, I own the technical decisions one to two days a week. For organic growth, I build search and AI-answer visibility and trace it to leads and revenue. In applied AI, I build agents, automation and software for a defined job, with human review. Each page sets out when the service fits and what you get first.
Three engagements, each with the evidence behind its result. An Australian education provider's revenue grew 129% on the same quarter of 2025, its best on record; the decisive fix was consolidating the category and landing pages that outranked its course pages. In July–September 2026 a UK training provider's online course orders reached 2.4x the same quarter of 2025. A technology recruiter's monthly organic clicks grew 24x within four months. Each study sets out the diagnosis, the work in order and what came next.