Journal

Browse by subject.

Four threads run through everything we publish. Pick the one you came for.

Engineering

How these systems are actually built: architecture, retrieval, orchestration and the trade-offs behind them.

Measuring model behaviour honestly — harnesses, regressions and the metrics that survive contact with users.

Interfaces for uncertain systems, and the craft of making correction cheap.

How we run engagements: scoping, handover, and the decisions that happen before any code.