Blog

Notes from work in progress

Essays and field notes — things I have noticed while building businesses, delivering complex projects, directing an AI-native ERP and trying to make AI-assisted work dependable. Written when there is something worth saying, not on a schedule.

TopicsAll Building and operating businesses Markets and commercial systems Physical and digital delivery Enterprise technology AI-assisted work and governance

LatestPlaybook · 3 September 2026 · 8 min read

Why Are You Teaching AI the Same Job Every Week?

Stop writing better prompts. Start building reusable AI skills. If you've explained the same task to AI three times, you probably don't need a better prompt — you need to save the way you want the work done.

Read the article

More writing

29 August 2026Essay · 8 min read

Agentic AI Doesn't Have an Intelligence Problem. It Has a Continuity Problem.

Agentic AI will change how we build, but intelligence alone won't be enough. A context window is not project memory — and the more autonomy we hand to agents, the more that distinction matters.

AI-assisted work and governance

26 August 2026Essay · 6 min read

The Next Breakthrough in AI Coding Won't Come From a Better Model

AI coding tools keep getting smarter, but our projects still have poor memory. The next real improvement may not be a better model at all — it may be giving projects a memory of their own.

AI-assisted work and governance

12 July 2026Essay · 3 min read

The Decision That Disappeared

AI tools helped our team move faster. They also made it easier to lose the reason behind a decision. What we changed, and what it taught me about working with AI.

AI-assisted work and governance

8 July 2026Retrospective · 6 min read

The Only Metric That Matters: Are We Re-Explaining Less?

Most metrics for a knowledge tool measure activity — searches run, entries created. The one that actually predicts whether it's working is quieter: are we explaining the same thing to the same tool less often than we used to?

AI-assisted work and governance

6 July 2026Essay · 6 min read

Protecting the Core Without Freezing the Project

Constitutions, templates and core rules keep a project coherent — until they calcify into rules nobody can change even when the project has clearly outgrown them. Finding the line between governance and paralysis.

AI-assisted work and governance

3 July 2026Essay · 6 min read

Why Every AI Agent Should Read the Same Project Memory

Claude, Codex, Gemini, Cursor — each one builds its own private understanding of a project unless something makes them read from the same source. What that costs, and what a shared, tool-neutral memory actually requires.

AI-assisted work and governance

30 June 2026Field Note · 6 min read

The Handover Is Where AI-Assisted Work Breaks

Human to AI, AI to human, session to session — every handover is a place where context can silently drop. What a minimum viable handover actually needs to carry, learned from watching plenty of bad ones.

AI-assisted work and governance

18 June 2026Essay · 6 min read

From Requirement to Release: Keeping the Chain Intact

A requirement, a decision, a work package, an implementation, a release. Six months later, can you still walk the whole chain backward? Most teams can't — and don't notice until it matters.

AI-assisted work and governance

4 June 2026Essay · 6 min read

Project Memory Is Not Chat History

A searchable transcript of every conversation isn't the same as a project that remembers what it decided and why. The difference between chat history and project memory, and why conflating them cost us more than once.

AI-assisted work and governance

21 May 2026Essay · 6 min read

Why Some Operations Should Never Run Themselves

Disaster recovery is the one place where finishing the job on schedule matters less than refusing to run an operation you can't yet prove is safe. A principle, not a war story — and why I'd rather admit that than dress it up as one.

AI-assisted work and governance

30 April 2026Essay · 6 min read

An AI System Should Earn the Right to Act

Observation, explanation and recommendation are cheap. Action is expensive. A case for making AI-assisted systems climb a ladder of trust before they're allowed to touch anything that matters.

AI-assisted work and governance