Q&A: Pharma AI in 2026: Discovery Solved, Execution Still Manual

At the begining of 2026, PharmTech connected with Megha Sinha, founder and CEO, Kolter AI and Founder, Kamet Consulting Group, to learn how she predicted the pharma industry would develop during the year. Now at the end of Q2 of 2026, she responds to her initial predictions and comments further on artificial intelligence (AI) in pharma manufacturing and supply chain.

Watch the 3-part video interview with Sinha:

Part 1: Midyear 2026 Check-In: Agentic AI Cuts Deviation Investigations by 50%

Part 2: Midyear 2026 Check-In: Megha Sinha on Pharma's AI Skills Gap

Part 3: Midyear 2026 Check-In: AI Investment Skipped Pharma's Costliest Operational Problem

Pharmtech: You Initially Predicted Pharma Companies Would Move Toward Real-Time Tariff Decision Engines. How Has That Played Out?

Sinha: Half of that prediction has come true, and the built half was the easier half. Companies like Merck, Novartis, and Bristol Myers Squibb now, based on public data, run real-time what-if simulations as a core part of supply chain planning, and the digital twin market has grown past a billion dollars. But those systems model shipments, inventory, and capacity, they aren't living tariff decision engines, and I haven't found a single pharma company running one. Tariff planning still mostly happens in a quarterly review slide after the disruption has hit. None of the deployed systems reach the regulatory consequences: the label updates, grace periods, and country-by-country submission waves a single tariff change trigger. That half has never been automated, which is exactly why we built an engine that computes a network decision's cost and execution impact market by market.

What About Inside the Plant? How Has AI's Role in Manufacturing Changed?

That side has delivered. Manufacturers are reporting agentic systems that detect an out-of-spec result, pull batch records, cross-reference historical deviations, draft a root cause analysis, and route it to the right reviewer. These systems are cutting investigation time by 50% or more in a validated GxP environment, a real shift from productivity tool to operational participant.

But even the best deployments reason locally. They translate a deviation into its financial and supply impact, but none recognize that a change proposed as minor on the floor might be a major variation requiring prior approval in Brazil, a six-month grace period in Germany, or a bundling opportunity elsewhere. The proposal stops at the fence line. Closing that gap is next.

What's Surprised You Most In 2026?

The bottleneck moved. For 3 years, the assumed constraint was model capability. This year, the technology has arrived, and the binding constraint is organizational readiness, employee sentiment, and data. I'm surprised by how fast the question shifted from if AI can do it to doing ROI math after 18 months of hype; everyone's running pilots, but the real question is converting one into an enterprise-wide solution.

The deeper surprise is where the money didn't go. Investment poured into AI drug discovery and planning pilots, technologies that photograph well. Meanwhile, the hardest, most valuable problem, computing the execution of a geo expansion or lifecycle change across functions and markets, drew almost no capital. The industry proved AI can design a molecule and draft a deviation and left compliant execution across 60 markets exactly where it's always been: manual, tribal, reinvented on every program.

What's the Biggest Challenge for the Industry, And Where Will it Land By 2030?

The biggest challenge is readiness to try new things operationally, the same courage the industry found on the science side. We rethought drug discovery from the ground up with AI; we haven't shown that willingness on operations across a global network.

By 2030, the operational landscape can't stay fragmented, clinical AI here, regulatory AI there, each brilliant in its own lane and blind to the others. My bet is on a connective layer, a nervous system carrying a single change from one function to the next and out across every market it touches, because fragmentation can't survive the pace AI-native discovery is about to create. New tariffs and reshoring incentives are already forcing network redesign faster than teams can absorb, and we've seen the most M&A activity this year. Strategy moves at the speed of the boardroom; execution still moves at the speed of a spreadsheet. The companies that wire operations into one auditable backbone now will be the ones ready when 2030 arrives.