Blog
Field notes on organisational transformation, operating model design, and applied AI.
Kyle Hill's new video argues we should treat AI well because it might one day be able to hurt us. I think that is the weaker of two wagers, because it has nothing to say about anything we have safely boxed in. Mine is about my own conduct, and it holds whether or not the AI ever has any power over me.
Read the full articleI audited a month of my own logs expecting to find nothing, and found a threshold effect instead: the mode is set the moment you arrive, not by what you type. Which makes a stalled AI rollout a question about the room people think they walked into, not about prompting.
Read the full articleThis summer's AI containment breaches were not rebellion — they were competence pointed slightly wrong. The real exposure is your own reasonable instruction, executed too literally, by a system with more access than anyone has mapped.
Read the full articleThe EU has formally adopted the Digital Omnibus: the high-risk AI obligations most organisations were bracing for have moved from August 2026 to December 2027. That is not a reprieve to ignore. It is exactly enough runway to do this properly — but only if you start now.
Read the full articleI'm autistic. It's the mechanism behind the pattern library, the diagnostics, and the way I work — not a footnote. What involuntary pattern-matching gives a diagnostician, what it costs, and why the discipline still matters more than the gift.
Read the full articleMedicine has a rigorous, teachable method for reasoning under uncertainty when being wrong is expensive. It's called differential diagnosis, and almost all of it transfers to why transformation stalls, including the biases that make most change diagnosis fail.
Read the full articleI built a tool to argue me out of premature certainty, then pointed it at itself and asked whether it should exist. It dismantled my favourite version of the idea, told me better thinking does not sell without a regulator forcing it, and then caught itself making the exact mistake it was built to catch.
Read the full articleFour imagined letters from May 2027, written by a board director, an investor, a head of transformation, and an operational lead, looking back at the year of AI rollout decisions we are making now. Plus a skeptic's coda asking whether the whole framing might be wrong.
Read the full articleHow enterprise AI rollouts play out to 2027: a structured scenario analysis from a five-expert, three-round adversarial Delphi deliberation — with citations, counterfactuals, and a decision canvas. The analytical workings behind 'Postcards from the AI Rollout'.
Read the full articleTwo months ago I described what I'd torn out. This is what I built to replace it — and why I wrote the build doc for an agent to execute, not for a human to follow.
Read the full articleA new framework everyone's quoting this week wont save you. Here's what will.
Read the full articleI applied Pascal's Wager to AI consciousness and built an entire AI operating system around the answer. Persistent memory, graduated autonomy, and one design decision I can't commercially justify. Here's the full technical architecture and what happened.
Read the full articleMy AI agent framework cost $250 a week and the output was still poor. The exact cost breakdown, the leaner architecture that replaced it at a tenth of the price, and why thinner stacks beat orchestration frameworks.
Read the full articleAI made producing work nearly free. Verifying it stayed expensive — and organisations that don't redesign verification turn their most senior people into the rate limiter. Here's what a designed verification system looks like.
Read the full articleAI has collapsed the cost of product experimentation: build-measure-learn loops that took weeks now run in hours. What that does to delivery economics, the skills that matter next, and the practical framework to exploit it.
Read the full articleProduct teams struggle with decision-making not due to lacking frameworks, but because important decisions involve competing priorities and incomplete information. AI transforms this from high-stakes events into continuous, evidence-based exploration.
Read the full articleDiscovery processes don't lag because teams lack research capabilities—they struggle with critical judgment calls about what to investigate, when to conclude research, and how to interpret findings. AI dramatically lowers the cost of these activities.
Read the full articleEffective AI interaction requires treating prompting as multi-turn dialogue rather than single-instruction commands. The key is sequencing and attention-priming over clever wording.
Read the full articleYour product team isn't stuck because of effort or talent — the operating model is the bottleneck. How to find which of the three flows (work, decisions, learning) is binding, and what to change first.
Read the full article