Lab Notes. Short entries from the building.
Short, precise entries from the work of building and operating dependable intelligent systems.
Where should an agent stop?
A practical test for deciding where preparation ends and accountable authority begins.
- Name the permitted action, evidence shown, uncertainty threshold, escalation route, reversible state and person who remains responsible.
Observability is part of the product.
An intelligent workflow is not dependable if its reasoning and outcomes cannot be inspected.
- Retain the input, sources, recommendation, decision, actor, action and resulting state as one operational trace.
Integration before replacement.
Useful intelligence often belongs between the systems people already use, not inside another destination.
- Connect existing sources of truth, prepare work in motion and replace technology only where the replacement earns its place.
Bring us the problem that keeps repeating.
A free conversation, thirty to forty-five minutes. If it repeats across businesses, it may already be on its way to becoming a pattern.