Merck and Mastercard are seeing real agentic AI results. Both say the plumbing came first.
Merck says agentic AI is cutting drug-discovery cycles by a third and speeding compliant marketing drafts up to 80%, but only after building the right infrastructure.
Intelligence analysis by GPT-5.4 Mini
The article argues that agentic AI only works at scale when the underlying systems come first. Merck says its cloud, data, security, and integration setup lets agents help with research, compliance, and app modernization.
Merck is using AI helpers to do parts of big jobs. Some help scientists look for new medicines faster. Other helpers make sure ads and documents follow the rules.
The company says these helpers work well only because the company built strong roads for them first. Those roads are things like secure computers, organized data, and clear rules.
It is like giving a toy car a smooth track instead of a muddy path. The car can go faster, but only if the track is built well first.
Analysis
Plumbing first
Merck VP of Digital Platforms Sean Finnerty says the company’s agentic AI progress depends on infrastructure built before the current wave of AI excitement. The core idea is simple: if companies create many one-off agent experiments, they risk accumulating technical debt that slows later innovation.
Merck compares the current AI moment to the early cloud era, when getting the foundations right required patience and discipline. The company now has thousands of AWS accounts, Microsoft Azure subscriptions, and growing Google Cloud Platform integrations. Finnerty says the same pattern will apply to AI agents, which means companies need ways to register them, secure them, connect them to the right tools, and provide the right data and context.
What the agents are doing
The company is using agents in regulated enterprise operations, scientific research, and app modernization. In drug discovery, Merck says one AI-assisted research cycle was reduced by 33%, which Finnerty framed as a year saved in the discovery timeline. In marketing, AI drafts are reportedly “99% right” on compliance, cutting review cycles from months to days and speeding delivery by 70% to 80%.
The article also says agents can help inspect architecture, document data flows, check authentication and authorization, write Terraform, and refactor code between languages. Finnerty’s team is building scaffolding for context delivery because Merck’s data is spread across many databases and repositories, and there is no single system that solves every problem.
Guardrails still matter
The company has also run into AI mistakes, including invented test scenarios and other “wackiness.” To reduce that risk, Merck is adding guardrails, using one AI system to check another and assigning confidence scores. The overall direction is optimistic, but the message is cautious: enterprise agentic AI works best when the plumbing, security, and oversight are already in place.
Key points
- Merck says its agentic AI results depend on infrastructure built before the current AI push.
- The company says one drug-discovery cycle was reduced by 33% with AI assistance.
- AI-generated marketing drafts are reportedly 99% compliant, cutting review time from months to days.
- Merck is also using agents for app modernization tasks like documentation, Terraform, and code refactoring.
- The company says guardrails and cross-checking are needed because AI still makes mistakes.
