Induced from Ganesh Muralidharan of Vaisesika Systems (transcribed as “Ganesh Muralidharan”) at the AI in GxP conference (2026-09-15-1340-kenneth-corley-duran-scaling-ai-in-regulated-industries), 2026-09-15.
The pilot paradox in numbers
Every life-sciences leader has experimented with generative AI (McKinsey puts it at 100 percent), yet only 5 percent realize significant financial value and 75 percent lack a strategic vision for it. Of manufacturers who started AI programs (68 percent per BCG and the WEF), 98 percent face scaling challenges and only 16 percent met their targets. The failure is not disagreement about whether to adopt AI; it is the order in which the functions engage.
QA brought in last stalls the pilot
Manufacturing owns workflow fit, IT owns sustainability, and QA owns trust, control, and evidence. When they work in sequence, each optimizing its own slice with QA consulted at the end, the pilot stalls; Duran’s supply-chain procurement case study died exactly there. The fix is concurrent design: all three functions define the system together before it is built. That is the organizational form of Compliance controls belong in the product from inception.
Shared language before design
Vaisesika’s 5C framework (Context, Criticality, Controls, Confidence, Continuity) gives the three functions a shared pre-design vocabulary, and the Assist, Augment, Automate classification ties autonomy level to assurance rigor: more autonomy demands more evidence. Assurance must scale with autonomy is that classification stated as a control rule. The unified delivery model assigns per-function ownership (process context to Manufacturing, quality risk to QA, architecture to IT) with shared ownership of intended use, data lineage, evidence, and human oversight. Shared ownership of intended use is why The intended use statement is the master control lever for GxP AI has to be written before the pilot, not after QA arrives. Merging the three functions into one unit sounds simpler but succeeded in only one of ten Vaisesika projects; the working linchpin in practice is often an IT-business-partners model, SME-embedded IT staff inside quality, regulatory, and safety. Once the three functions stay concurrent, Validating an AI system in a regulated industry means validating a measurement system becomes a shared measurement job rather than a late QA document.