Midyear 2026 Check-In: Agentic AI Cuts Deviation Investigations by 50%

Megha Sinha, founder and CEO of Kolter AI and Kamet Consulting Group, spoke with PharmTech at the beginning of 2026 about her predictions for the industry this year. Now that the year is halfway through, In part 1 of a 3-part interview, she discusses how she believes that the infrastructure around digital twins is "Half built, and the built half is the easier half."

Companies such as Merck, Novartis, and BMS now run real-time what-if simulations on select S&OP processes, and the market for plant-level digital twins has surpassed a billion dollars. Inside the plant, agentic systems are already detecting out-of-spec results, pulling batch records, drafting root cause analyses, and cutting investigation time by 50% or more in validated GxP environments. "That is a shift from AI as a productivity layer to AI as an operational participant," Sinha notes.

But Sinha argues these systems stop short of where the real cost lies: translating a network, process, or manufacturing change into its regulatory consequences market by market. A minor change on the plant floor, she explains, can trigger a major variation requiring prior approval in Brazil, a six-month grace period in Germany, or a bundling opportunity across other markets, none of which existing digital twins or agentic tools are built to compute. Tariff planning, she adds, is still largely a quarterly review exercise rather than a live decision engine.

That gap between physical execution and regulatory execution is what Sinha says Kolter AI was built to close. "To me, that is the next frontier, and companies should be moving on it now," she says, describing an engine that takes a network or process decision.