For pharmacy and clinical informatics leaders, the “build-versus-buy” conundrum has become the default starting point for AI technology decisions. In medication workflows, however, that framing is incomplete. Whether an organization builds internally, buys from a vendor, or extends its EHR, there are foundational considerations that apply across all approaches. These include what medication intelligence foundation the AI is grounded in; who is accountable for keeping it current, validated, and patient-specific; and whether that intelligence stands up to clinical, operational, and governance scrutiny.
The stakes are high
Inaccurate or incomplete grounding data puts patient safety at risk. One national analysis estimated that inappropriate medication-alert overrides may have contributed to roughly 5.5 million preventable alert override events and between $871 million and $1.8 billion in costs for treating adverse drug events (ADEs). Research drawing on the Agency for Healthcare Research and Quality (AHRQ) Network of Patient Safety Databases found that incorrect medication and incorrect dose accounted for 67% of known ordering errors and 80% of events causing severe harm or death. Nearly 80% of those errors were classified as definitely or likely preventable.
Given these findings, it quickly becomes clear that medication AI is about more than just digital transformation. It also sits squarely in the domain of patient safety and clinical liability.
These high stakes land in a system already under pressure. Health system leaders are being pushed to move beyond piloting AI and into rapid deployment. But the practical questions are immediate: will AI reduce manual work, improve safety, and scale without creating a maintenance burden? With technology changing at seemingly warp speed, the risk of building the plane in flight is real, especially if the foundational medication intelligence beneath it is shaky.
In a late 2025 poll, 74% of revenue cycle leaders said poor data quality is a significant or critical barrier to AI adoption. Other research has pointed to the same finding from a different angle: even small amounts of poor-quality data can substantially degrade AI performance and limit how well models generalize. At the same time, EHR heterogeneity continues to undermine data quality across the clinical data life cycle. Without a strong grounding layer, even sophisticated AI models produce inconsistent recommendations and incorrect alerts, jeopardizing patient safety and outcomes.
The wrong first debate
This is why build-versus-buy is often the wrong opening debate. Whether a health system builds internally, buys from a vendor, or extends an EHR module, every AI-enabled medication workflow still depends on trusted medication intelligence that is clinically validated, context-aware, and updated as evidence changes. As Virginia Halsey, FDB’s senior vice president of strategy and product management, said, “As AI capabilities expand, analytics grow more sophisticated and digital health solutions move closer to clinical decision-making, medication data alone is no longer sufficient. Advanced medication intelligence, powered by trusted, patient-specific drug knowledge, is now foundational to modern health technology platforms.”
Once that grounding requirement is clear, the question becomes: what does it look like in practice when AI is built on a trustworthy medication intelligence layer?
What grounded agentic AI looks like in practice
At the infrastructure level, FDB MedProof MCP offers one answer. We describe it as a standardized, AI-native integration layer that grounds agents in clinically validated medication knowledge, improves safety, reduces the risk of hallucinations and supports deployment across clinical and patient-facing workflows via the Model Context Protocol (MCP).
FDB purpose-built this solution so that customer agents are always grounded in trusted, continuously updated medication intelligence. Agentic AI in medication workflows is only as trustworthy as the medication intelligence anchoring it.
Medication intelligence lightens pharmacists’ cognitive burden
Within daily pharmacy operations, FDB VerifyAssist applies the same grounding principle to hospital medication order verification workflows. FDB has reported that medication order verification can take 30% to 40% of a hospital pharmacist’s time. FDB VerifyAssist is designed to reduce that burden by as much as 30% per drug order by surfacing the most important, drug-specific verification criteria against each patient’s current clinical data, including factors such as recent labs and age. For a chief pharmacy officer, the design ties AI directly to cognitive load reduction, verification efficiency and pharmacist capacity, rather than to a generic promise of automation.
PGx raises the bar
Pharmacogenomics (PGx) demonstrates why the grounding challenge continues to grow as care becomes more personalized. FDB Pharmacogenomic Clinical Decision Support draws on evidence from FDA labeling and leading pharmacogenomic guidelines, including those provided by the Clinical Pharmacogenetics Implementation Consortium (CPIC) and the Dutch Pharmacogenetics Working Group (DPWG), and is subject to ongoing review by FDB pharmacist-informaticist teams. PGx proves the intelligence layer handles the hardest, fastest-evolving clinical data, not just commodity drug facts, to drive success throughout the patient’s healthcare journey.
Broader implementation research reaches a similar conclusion: PGx is difficult to operationalize because evidence continually evolves, interpretation is complex and effective use depends on integrated clinical decision support (CDS) rather than clinician memory alone. As therapy becomes more personalized, the bar for the medication intelligence layer keeps rising.
Questions for leadership
For the next governance, pharmacy and therapeutics (P&T) or digital health meeting, four questions matter more than a generic build-versus-buy debate:
- What medication intelligence source is grounding each workflow, and how often is it updated?
- Can it support patient-specific context, including labs, allergies, drug interactions and PGx factors?
- Who owns validation, change management, and accountability when evidence changes?
- Will it reduce low-value manual work, or simply shift maintenance elsewhere?
The bottom line is this: does selecting a vendor constrain or expand future projects? For organizations asking these questions, a tested clinical drug knowledge infrastructure often separates AI pilots from successful, scaled deployments.
The most important decision in 2026 reaches beyond build or buy. The deeper question is whether your organization has a medication intelligence foundation strong enough to support agentic AI, streamline pharmacists’ workflows and keep pace with more personalized medication decision support as it continues to evolve.
FDB’s clinical drug knowledge intelligence and infrastructure are purpose-built for this moment — connecting validated medication intelligence to the agentic AI workflows health systems are deploying today.
About the Author
David P. Delaney, MD, is President of FDB (First Databank), a leading provider of drug knowledge and clinical decision support solutions trusted by healthcare organizations worldwide. He is responsible for enterprise strategy, growth, product innovation, and operating performance. Learn more about Dr. Delaney and connect with Dr. Delaney on LinkedIn.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
