Skip to content
Join 7,000+ leaders following Alastair's work on LinkedIn.

Standard practice

Defensive Programming

Put predictable rules around unpredictable AI: rules first to handle the bulk, rules after to check the output.

Predictable software checks unpredictable AI.

Rules firstthe Deterministic-First Cascade

Cheap, predictable rules handle the bulk. The model only catches what the rules miss, so the rules do most of the work.

Rules afterbefore anything acts

Validate the output against rules before anything happens downstream. Never give the AI a button that acts on its own: a human or a hard rule controls the irreversible step.

When to use  Designing an AI workflow that has to stay cheap at volume, or where a wrong output could do real damage.

Read the full post →
topic: risk & safety type: principle

Want help putting this to work?

A Focus Call is a free, no-pitch conversation about where AI fits in your business.

Book a Focus Call