Induced from Ganesh Muralidharan’s scaling-AI lecture at the KENX AI in GxP East 2026 conference (2026-09-15-1340-kenneth-corley-duran-scaling-ai-in-regulated-industries), 2026-09-15.
The assist, augment, automate ladder
Classify each use case by how much it does on its own. An assist drafts; an augment decides with a human; an automate executes. Not everything should climb the ladder, and not everyone wants it to. The governance consequence is proportional: more autonomy means more rigor, because the system initiates actions, triggers workflows, and executes tasks with GxP impact.
What the ladder changes on the autonomous rung
The intended-use lecture lands the same claim from the output side: advisory outputs shift the burden to proving a trained human is in the loop, while autonomous outputs demand harder confidence scoring, fallback logic, and failure-mode containment for the excursion that will eventually come. That is The wording of an intended use statement sets the regulatory burden applied to the automate rung, and The intended use statement is the master control lever for GxP AI is the lever that chooses the rung before the system is built. Regulators audit against principles, so size quality rigor to risk is the same proportionality stated from the inspection side. The organizational failure mode is still AI pilots in GxP fail on sequencing, not technology: autonomy gets automated before QA has concurrent ownership of the evidence.