My Books

The book I wrote for the moment design stopped being something you do only in a canvas. It covers the agentic design paradigm end to end: agent platforms, design-for-agents harnesses, design-as-code workflows, and orchestrating more than one agent without the work turning to mush.

It is written for two people at once — the designer who is curious about agents and suspicious of the hype, and the engineer building design tooling who needs to understand what designers actually protect. Both get the same material: how the operating model changes, what moves to code, what stays judgement.

Fourteen chapters plus three appendices of battle-tested prompts and copy-paste examples, and four real case studies written with honest assessments — including the parts that fell apart.

A practical roadmap for visual designers and product managers moving into design engineering. It starts where most guides refuse to: at the terminal, assuming no prior comfort with it, and gets you to a working prototype before asking you to understand anything clever.

From there it covers the parts that actually decide whether AI-assisted design work survives contact with a real project — prompting for design rather than for prose, teaching an agent your design language, tokens and design systems at scale, and the handoff from design to production.

Ten main chapters and three reference appendices, with prompt templates and step-by-step terminal guides. The goal throughout is to work directly with coding agents without turning the job into a programming manual, and without losing your designer sensibility.

Open Design is an open-source design harness for AI coding agents, and this is its field guide. It runs locally — skills, DESIGN.md brand systems, plugins, an MCP server, and a sandboxed preview and export loop — so an agent can produce real design artifacts instead of describing them.

The argument underneath is about independence. Local-first and agent-native means your brand system, your components and your output pipeline stay yours: no vendor holds the file format, and no cloud service sits between an idea and a rendered artifact.

Sixteen chapters and four appendices, including 150 brand systems, the plugin ecosystem, multi-agent design teams, case studies, and an honest chapter on evaluating whether this stack is right for your organisation.

The evidence-contract field guide: how to build automated sport analytics that you can actually defend. It is organised around five verbs — watching, tagging, modelling, interpreting, and acting — and it refuses to let a number travel without the provenance that earned it.

The full chain is here and measured, not asserted: capture and lens choice, homography and per-frame calibration under broadcast dynamics, detection and the AGPL licensing decision that comes with it, tracking and re-identification, segmentation with foundation models, 2D keypoints to 3D biomechanics, and spatio-temporal ball-trajectory recovery.

Then the harder half — simulation and counterfactuals done honestly, where vision-language models help and where they lie, the agent harness itself, a live coaching cockpit with real latency numbers, sensors and fusion, and a full chapter on deployment, licensing, rights and ethics. Everything is measured end to end on pickleball and rugby league, with a laboratory chapter for reproducing the book's claims yourself.