Insights

Field notes from the practice.

Essays on AI safety, governance and organisation design — written from inside live engagements, not from the sidelines.

GovernanceAug 2026

The board pack is not a control

Quarterly reporting tells directors what happened. It does not give them a hand on the wheel. What real oversight instrumentation looks like.

Organisation designJun 2026

Name the human: decision rights for autonomous systems

Accountability survives only when it has an address. A pattern for assigning ownership of model decisions that outlives the reorg.

Responsible AIApr 2026

Refusal is a feature: designing systems that can say no

Well-designed AI declines gracefully. On refusal thresholds, calibration, and why a confident 'not this one' builds more trust than a fluent guess.

RegulationFeb 2026

EU AI Act, six months in: what actually changed

Beyond the headlines — the operational duties that bite first, and the three artefacts regulators keep asking to see.

Field notesDec 2025

What we learned instrumenting 40 model fleets

Patterns from a year of deployments: where drift hides, which alerts teams ignore, and the one metric boards actually use.

Prefer the conversation to the essay?

Talk to us