Writing

Essays on AI, agentic systems, enterprise transformation and systems thinking. Arguments, not announcements.

  • 6 min read

    The Step Everyone Skips

    Geoff Woods' CRIT framework asks the AI to interview you before it does anything. Most people skip that step, and the reason they skip it is exactly why it matters.

  • 6 min read

    Encoding Taste So a Machine Can Follow It

    Why 'make it look good' fails as an instruction to an AI coding agent, how TasteSkill turns design judgement into explicit constraints, and where rules stop and taste begins.

  • 6 min read

    AI at the Edge of a Three Dollar Chip

    What AI development honestly looks like near an ESP32: what fits in 520 KB of RAM, what belongs on a server, and the latency and power trade-offs that decide the split.

  • 5 min read

    When the Problem Is Actually a Graph

    How to tell when nodes and edges are the honest shape of a problem, what traversal really costs, and where graph systems start to break as they grow.

  • 7 min read

    Give Juniors the AI. Build Them a Harness.

    The consensus says AI belongs with the senior engineers who can catch its mistakes. I think juniors gain the most from it, provided someone builds the harness that makes a wrong answer safe.

  • 6 min read

    Mentoring Young Minds When Answers Are Cheap

    How to develop a junior engineer's judgement: asking before telling, productive struggle, feedback that lands, and what changes now that plausible answers are instant.

  • 6 min read

    Build the Assistant That Cannot Betray You

    An AI personal assistant is only as safe as its worst possible action. On capability boundaries, read-only defaults, prompt injection and what should always stay human.

  • 5 min read

    Still a Programmer

    An honest accounting of what AI changes about programming and what it does not: where the work moves, where judgment still lives, and what is worth practising now.

  • 5 min read

    Prompting Is Specification, Not Incantation

    Why prompt tricks age badly, what actually generalises across models, and how specificity, context, constraint and evaluation turn prompting into engineering.

  • 6 min read

    An MCP Server Is Infrastructure, Not a Plugin

    What changes when you move a Model Context Protocol server from a local stdio process to a hosted service on Railway, and why secrets and blast radius are the real work.

  • 5 min read

    The Vibe Coders Are Shipping

    People without engineering backgrounds are building real software with AI. What they get right, where the danger actually sits, and how engineers can be useful to them.

  • 6 min read

    The Voice Is the Easy Part

    Speech interfaces live or die on latency, turn-taking and honesty about what the system is. On the engineering behind giving an AI a voice with OpenAI and ElevenLabs.

  • 1 min read

    Hello, world

    A first post, existing so the publishing pipeline has something real to carry end to end.

Stay in touch