Induced from Shelly LaPointe’s data-foundation lecture at the KENX AI in GxP East 2026 conference (2026-09-15-1201-shelly-lapointe-data-foundation-for-ai-in-gxp), 2026-09-15.
Silos make every AI pilot rebuild its plumbing
LIS, MES, and QMS each carry their own format, access model, and governance, and GxP environments are tuned to be inspection-ready rather than insight-ready (GxP systems are optimized to be inspection-ready, not insight-ready). The result is that each AI pilot builds its own non-reusable pipeline, and agentic AI multiplies the problem because every additional agent touching the data multiplies the traceability obligations.
Build the estate once, validate it once, reuse it everywhere
Her reference architecture: sources feed an ingestion framework that normalizes, tags, and captures lineage; storage splits into raw, curated, and consumption zones beside a metadata catalog; role-based access control, encryption, and a GxP validation envelope wrap pipelines, storage, access, and every model and agent. One governed dataset then serves autonomous agents, training pipelines, compliance dashboards, and humans. Centralize evaluation when every AI pairing carries its own KPIs is the evaluation-estate version of the same reuse.
ALCOA+ still governs; the enforcement surface changes
AI does not change the data-integrity requirements, only where they bite: audit trails, e-signatures, and human-understandable agent naming matter more, and “enduring” is the hardest attribute to keep as models keep changing. That naming requirement is why Autonomous agents need their own auditable identities in regulated systems. GAMP 5-style risk-based validation applied well cut validation cycle time about 35 percent.