Merck is using AI agents to cut drug discovery cycles by a third and ship compliant marketing materials up to 80% faster — but VP of Digital Platforms Sean Finnerty says the only reason it’s working is because they built the infrastructure first. The pharmaceutical company’s approach highlights a growing trend: agentic AI requires foundational systems before it can deliver results.
Merck’s early results show AI generating marketing drafts that are “99% right” when it comes to compliance. Review cycles that once took months now take days, and delivery speeds have accelerated by 70% to 80%. In medical research, one AI-assisted discovery cycle was reduced by 33% — a year shaved off the timeline for a drug to reach patients.
“If we do one-offs, we’re gonna end up with thousands of things that are ultimately just gonna be debt,” Finnerty said at a recent AI Impact Series event. The company’s strategy mirrors lessons from the 2010s cloud transition, where Merck built infrastructure from the ground up. Today, that system supports 2,500 AWS accounts, Microsoft Azure subscriptions, and Google Cloud integrations.
Related: Alibaba Unveils Long Lasting AI Model Qwen3
AI agents, Finnerty explained, will soon number in the thousands. The challenge lies in registering, securing, and connecting them to the right tools. Merck’s edge locations, databases, and petabytes of structured and unstructured data — stored in Oracle, SQL, Excel, and phone transcripts — require careful context delivery.
“There’s no one solution to solve every single problem,” Finnerty said. His team uses Databricks, Amazon Redshift, and other platforms to organize data. The goal is to make workflows frictionless, secure, and compatible with protocols like A2A and MCP, ensuring agents can operate across GCP, AWS, or other environments.
AI’s impact extends beyond discovery. Marketing materials, once mired in months of human review, now get first drafts in days. A 99% accurate initial version cuts time to approval, allowing materials to ship faster. For app modernization, AI discovers architecture, documents APIs, and writes code — tasks that previously took weeks and hundreds of thousands of dollars.
But challenges remain. Finnerty’s team has seen AI generate “wackiness,” like fabricating non-existent test functions. To combat this, they use guardrails: AI supervising AI, with confidence scores applied across multiple checks. “If you ask something once, have AI check it, then ask it a third time, the confidence increases,” Finnerty said.
Related: Why Hiring the Best SEO Company in Los Angeles Is Key to Digital Success
At Mastercard, Chief Data Officer Andrew Reiskind is experimenting with agentic AI in transaction and dispute workflows. Disputes aren’t single events; they trigger back-end processes involving merchants, networks, and consumers. Reiskind described a system balancing deterministic and probabilistic decisions — structured data like lost cards versus unstructured consumer complaints.
AI could speed these processes, but risks loom. If an agent misclassifies a dispute, a consumer might be wrongly accused. Reiskind stressed the need for cost-benefit analysis: “Is it an acceptable risk if one percent of the time it makes the mistake?”
Enterprises must weigh risks early, he said. Whether the error is serving a wrong sandwich or risking a celiac patient’s health, the stakes vary. “It’s not a simple process to get to the cost,” Reiskind admitted. “But it is doable.”
Leave a Reply