Building the Autonomous Pharmacy: A Road Map for AI-Enabled Medication Management

For more than a decade, health systems have talked about the autonomous pharmacy—a fully digitized, self-optimizing medication-use process in which error rates approach zero, and pharmacists work at the top of their license—as an end state. Getting there, however, has proven slower than the vision suggested. A new white paper argues that generative and agentic artificial intelligence (AI) may finally be the catalyst that moves health systems past the fragmented, manual workflows that have kept most organizations stuck at the limited or intermediate stages of automation. A 5-level model for classifying AI-enabled pharmacy systems may help health systems advance to new levels of pharmacy autonomy.¹

The Original Framework: 5 Levels of Pharmacy Autonomy

The Autonomous Pharmacy Framework, first published in 2021² and organized around 5 components—enterprise structure, information technology (IT) infrastructure, automation, data intelligence, and human activity—describes 5 levels of organizational maturity, from nonautonomous pharmacy to fully autonomous pharmacy.

At Level 1, nonautonomous pharmacy, automation is minimal to nonexistent. The paper states, “Data are managed largely on paper or in disparate spreadsheets, and pharmacists are heavily engaged in direct distribution, with technicians and nurses spending much of their time on manual drug management tasks, including purchasing, locating, and counting.”²

Level 2, limited autonomous pharmacy, introduces some automation, including barcode tracking, although data remain managed disparately across sites with only partial visibility. Pharmacists are still mostly focused on distribution and verification during this stage, whereas nurses and technicians remain manually responsible for most drug management. Automated support is utilized lightly.²

By Level 3, intermediate autonomous pharmacy, most processes are automated, and barcode tracking is applied widely. Data become integrated across the enterprise and are mostly visible. Pharmacists are somewhat freed from medication distribution, with some direct patient care responsibilities emerging, whereas technicians shift toward procurement and controlled substances, and nurses still rely on unassisted dispensing.²

Level 4, highly autonomous pharmacy, extends automation with few remaining gaps across processes. Data visibility is nearly complete, offering real-time insights and workflow optimization. Pharmacists become routinely involved in direct patient care, population health, and clinical programs, whereas technicians maintain the automation and workflow app, and nurses spend most of their time on direct patient care rather than administrative tasks.²

Level 5, fully autonomous pharmacy, represents complete process automation, with each dose tracked as a node on the network, full data visibility, real-time workflow optimization, and predictive intelligence. The white paper states, “Pharmacists realize the full scope of their role in direct patient care and clinical program optimization. Technicians ensure optimal function of the automation and workflow app, and nurses focus on direct patient care.” The framework ties these levels to measurable outcomes across 5 performance elements—safety, financials, efficiency, patient adherence, and people—with the fully autonomous end state targeting zero medication errors and waste, as well as 100% data visibility, time spent on clinical activity, and regulatory compliance.²

A New Way to Measure AI Maturity

The white paper’s central contribution is the Medication AI Agent Progression Framework, a 5-level model that classifies AI-enabled systems by their degree of automation, data intelligence, and required human involvement, which is distinct from, but complementary to, the original Autonomous Pharmacy Framework published in 2021.^1,2

Each level is defined along 3 dimensions—automation, data intelligence, and human activity—that shift in tandem as systems mature.¹

• Level 1, Rule-Based Automation: Covers fixed-rule tools with no ability to learn or adapt, such as drug-drug interaction alerts, allergy warnings, and Excel “if/then” formulas, where automation is limited to single systems and humans must still decide and act.¹
• Level 2, Intelligent Automation: The white paper suggests using focused machine learning models, such as predictive sepsis models and automated inventory management, that recognize patterns across multiple, but still separate, systems and adapt over time with continuous feedback, although humans still make the final call.¹
• Level 3, Agentic Workflows: Combines machine learning with formal workflows so systems can take human-like actions within single systems via application programming interfaces, according to the white paper. This includes placing calls or escalating alerts, reducing but not eliminating the need for human decisions on simple tasks.¹
• Level 4, Semiautonomous Agents: The paper describes this as a multimodal perception, such as computer vision and ambient listening (eg, clinical copilots and fall-prevention tools), enabling systems to interpret goals and plan tasks across multiple systems, while humans still oversee consequential decisions.¹
• Level 5, Fully Autonomous Agents: A theoretical level with no real-world examples yet, the authors wrote; however, if it is fully integrated across systems with complete contextual understanding, independent reasoning, and learning in real time like a workforce peer, human workers will review the agent’s output rather than direct it.¹

“I think organizations are still in the process of understanding and adopting the framework. Most organizations are likely to lie somewhere between the 2.5 to 3.5 range out of a 5.0, but organizations that are building new consolidated service centers are moving toward highly automated systems,” author Janjri Desai, PharmD, MBA, DPLA, vice president of pharmacy services at Stanford Health Care, explained in an interview with Pharmacy Times.

What It Takes to Get There

Progressing through the framework, the authors argue, depends on 5 layered infrastructure requirements: data infrastructure and integrity, regulatory and ethical oversight, stakeholder engagement and education, performance monitoring, and active management of an “agentic workforce” alongside human staff at the higher levels.¹

Data governance and Health Insurance Portability and Accountability Act compliance form the foundation. Regulatory readiness is flagged as one of the biggest open questions because higher-level workflows, such as medication auto-verification, would require changes to law and accreditation standards before they could be implemented. The ethical considerations are also significant: The paper calls out algorithmic bias, transparency, informed consent, and accountability for AI-related errors as issues that must be addressed alongside the technology itself.¹

“The least-understood issue by health system leadership right now is accountability for AI-related errors, especially under the rapid adoption of agentic workflows that we see ourselves heading toward. While humans are largely in the loop with agents today, it is far too easy to ‘set it and forget it’ in terms of deployment, leaving unclear who is responsible when performance drifts.… Ethically, this is challenging as accountability is diffuse from the start.… This creates risk for errors, automation bias, inequitable performance, and weakened clinical judgment (deskilling and never skilling) over time,” said author Seth W. Hartman, PharmD, MBA, vice president of clinical applications and chief pharmacy information officer at UChicago Medicine. “Our responsibility and the message to share is clear.… Our ethical obligation is to not just adopt AI quickly, but to ensure the safe performance, transparency, and responsibility of each party within the process to detect and prevent failure of the system.”

A Practical Path for Health Systems

Recognizing that most pharmacy leaders are not starting from a blank slate, the paper offers a 4-step adoption framework: problem identification, solution scoping and comparison, implementation and education, and performance monitoring and sustainability. The authors caution against a mismatch between the scale of a problem and the sophistication of the AI solution applied to it, as overengineering a fix can generate as much organizational resistance as underinvesting in one.¹

Vendor partnership also gets particular attention. Because pilots are often not feasible for Level 4 and 5 technologies given the infrastructure they require, the authors recommend a detailed, side-by-side vendor comparison to build the business case, tied explicitly to efficiency, access, or other core organizational priorities rather than AI adoption for its own sake.¹

As generative and agentic AI mature, the gap between where most health systems sit today and the fully autonomous end state may finally start to close, but only for organizations willing to invest deliberately in the data, regulatory, and workforce foundations the paper describes. For pharmacy leaders, the takeaway is less about chasing Level 5 and more about honestly assessing where their organization stands now and building the infrastructure to advance one level at a time. As health systems weigh where to invest next, the Medication AI Agent Progression Framework offers a shared vocabulary for measuring how far a pharmacy has actually come and how far it still has to go, something the field has lacked.¹

“Directors who are exploring adoption of AI within their organizations should focus on 3 key things: developing a strong partnership with their IT team, possibly through implementation of a pharmacy technology and informatics program to bridge the gap between clinical workflows and IT capabilities; taking the time to outline and understand the key problems within their organizations that make sense to leverage technology and digital solutions to solve and recognizing that not all problems need AI or agentic solutions; and not rush or feel pressured to implement AI for the sake of implementing AI,” Desai concluded. “There are an overwhelming number of platforms and AI capabilities that exist, and it’s most important to be thoughtful and methodical in the approach to implementation. If there is no pressing need, wait and watch can also be a sound strategy given the pace of innovation, and it may result in the selection of the most sustainable solution.”

REFERENCES

1. Desai J, Hartman S, Flynn A, Sisodiya D, Wagner K, Desai N. Advancing the Autonomous Pharmacy: Leveraging AI to Transform Medication Management. Accessed July 23, 2026. https://static1.squarespace.com/static/66a2a3645c1c775b50b81de7/t/69b9cd90b1098674062c2482/1773784464042/omcl-wp-advancing-the-autonomous-pharmacy-v1.4-030326-FINAL.pdf

2. Sullivan M, Tryon J, Maehlen H, et al. The Autonomous Pharmacy Framework: A Transformative Vision for the Medication Management Process. Accessed July 23, 2026. https://static1.squarespace.com/static/66a2a3645c1c775b50b81de7/t/67a28520b464f1202addbd16/1738704160325/AutonomousPharmacyFramework_WhitePaper.pdf