Introduction: The Infrastructure Imperative

Agentic AI is not a plug-and-play revolution. Merck and Mastercard are demonstrating that real results—33% faster drug discovery cycles and 80% faster compliant marketing materials—come only after building the underlying data and platform infrastructure. Sean Finnerty, Merck's VP of Digital Platforms, warns that one-off AI deployments create technical debt that stifles innovation. This intelligence briefing dissects the strategic implications for enterprises racing to deploy AI agents.

Why Infrastructure First Wins

Merck's infrastructure supports 2,500 AWS accounts, multiple Azure subscriptions, and new GCP integrations, alongside 47 edge locations and hundreds of databases storing petabytes of structured and unstructured data. This plumbing enables agents to access the right context securely. Mastercard's Chief Data Officer Andrew Reiskind echoes the need for robust data pipelines to handle complex dispute workflows that mix deterministic and probabilistic decisions. The lesson: without a unified data layer, agentic AI will fail at scale.

Strategic Consequences for Enterprises

Who Gains?

Cloud hyperscalers (AWS, Azure, GCP) and data platform providers (Databricks, Redshift) are clear winners as enterprises invest in multi-cloud architectures. Merck's approach of letting workloads run on any cloud—"run your adjacent workloads wherever you want"—locks in these vendors. Early adopters like Merck and Mastercard gain competitive moats through faster innovation cycles and cost savings.

Who Loses?

Legacy IT vendors offering monolithic solutions lose relevance. Companies that skip the plumbing phase will face ballooning technical debt, as Finnerty warns: "If we do one-offs, we're gonna end up with thousands and thousands of things that are ultimately just gonna be debt." Manual process workers in compliance and dispute resolution face displacement as AI achieves 99% accuracy in first drafts.

Second-Order Effects

Agentic AI will accelerate industry consolidation. Only firms with deep pockets for infrastructure (Merck's multi-cloud setup) can compete. Regulatory scrutiny will intensify as AI handles compliance—mistakes like serving gluten to a celiac patient (Reiskind's analogy) carry reputational risk. Cross-agent protocols (MCP, A2A) will become standard, creating new interoperability standards.

Market and Industry Impact

Enterprise AI spending will shift from model training to infrastructure. The cost-benefit analysis Reiskind advocates—breaking problems into constituent pieces—will become a core competency. Expect a surge in demand for data engineering talent and platforms that simplify context delivery. The "human-as-governor" model (Finnerty) will dominate regulated industries, with AI handling 99% of tasks but humans overseeing critical decisions.

Executive Action

  • Audit your data infrastructure: Can it support thousands of agents with secure, contextual access? If not, invest in multi-cloud plumbing before deploying AI.
  • Decompose workflows into structured and unstructured components. Use cost-benefit analysis to determine acceptable risk levels (e.g., 1% error tolerance).
  • Adopt cross-agent protocols (MCP, A2A) early to avoid vendor lock-in and enable future scalability.

Why This Matters

The window to build agentic AI infrastructure is closing. Early movers like Merck and Mastercard are already reaping efficiency gains that competitors cannot match without foundational investments. Every month of delay compounds technical debt and competitive disadvantage.

Final Take

Agentic AI is not about the model—it's about the plumbing. Enterprises that treat infrastructure as a strategic asset will dominate; those that chase quick wins will drown in debt. The signal is clear: build the pipes first, then let the agents flow.



Source: VentureBeat

Rate the Intelligence Signal

Intelligence FAQ

Infrastructure first. Without robust data plumbing (multi-cloud, edge locations, unified context), agentic AI creates technical debt and fails to scale.

Merck cut one discovery cycle by 33%, potentially shaving a year off time-to-patient.

Reputational risk from errors—like serving gluten to a celiac patient. Acceptable error tolerance (e.g., 1%) must be defined upfront.