Pharmacy General Intelligence: Q4 2025 Update
Welcome to the last update for 2025 tracking the evolution of AI through the lens of Pharmacy General Intelligence (PGI). As a reminder, PGI focuses specifically on AI's potential to perform at or beyond the level of a pharmacist, which could lead to AI agents integrating into pharmacy workflows for tasks like medication verification, dose adjustments, and patient counseling notes.
Q4 2025 was filled with policy, industry, and technology shifts and additionally, Q4 will be remembered as the time that AI started to get political. Maybe driven by the elections in November or other factors, for the first time, beliefs and opinions on how and when AI should be used have started to be expressed by constituents from both parties. Interestingly, many of the issues being brought up have bipartisan alignment from members who would normally be on completely opposite sides of an issue. Let’s move into examining what happened, and how it will influence the development of PGI.
Policy Shifts:
Perhaps the most direct and influential policy shift that happened was the National Policy Framework for AI Executive Order.1 This order directs the Attorney General to challenge state AI laws.1 This could have a major impact on the progression and restrictions around AI as its objective is to set legislation at the federal level without state involvement.
In addition, the Genesis Mission was unveiled which gives the Department of Energy the task of creating a closed loop AI experimentation platform to utilize data and supercomputers for acceleration of innovation in biotechnology and nuclear energy.2 CMS is pushing for changes around healthcare data by pushing major players in the EMR space to share data and adopt the CMS Interoperability Framework.3 I have written about how important the access to patient data will be for development of PGI (or any medical AI) and this framework along with the FHIR API data specification could end up being the data pipeline that feeds AI systems.
There was an article published during Q4 that explores how pharmacy regulators prefer principles-based guidelines over strict rules for AI.4 It identifies core principles for responsible AI use, including transparency, redundancy, and auditability, while noting the logistical difficulty of enforcing patient "informed consent".4 This is like the stance that nursing organizations have taken with their AI policies and guidelines.4
Two final bits of policy that came up during Q4 are tensions over health AI oversight between HHS and the Coalition for Health AI (CHAI).5 There are differing opinions on whether AI oversight should be performed by federal agencies or private groups.5 This will be a new topic to watch as it will determine who would be responsible for oversight of PGI. The SANDBOX Act was also unveiled as it would allow developers to waive federal regulations during testing phases of AI products.6 This could potentially lead to faster development while at the same time increasing the risk of adverse effects from untested AI products.
Industry Trends
The effect of AI on healthcare continues to be monitored and is starting to be better evaluated as systems have been in place long enough for data to have been collected. The first source looks at automation’s effect on healthcare jobs by 2030.7 The report states that there will be a massive shift in healthcare activities with only 34% of healthcare activities being performed by humans in 2030.7 This is an incredibly ambitious statistic and brings up the balance between expectation and reality. Most likely, this change will be a bit more tempered as the counter argument stats that healthcare jobs so far have avoided mass AI-related job cuts like other industries.8 The authors state that this may change in the future, but for now, it is much more difficult to replace the humans performing healthcare activities than in other industries.8
While the rate of integration into healthcare is up for debate, a common discussion point is that Hospital AI Governance lags behind the rapid rate of adoption by facilities.9 About a third of health systems have implemented policies for model inventory and sign offs based on the article.9 While some hospitals struggle to create this oversight, clinicians see the opportunity for AI to assist with prior authorization and other administrative burdens.10 Some of the most widely deployed and utilized AI systems revolve around utilizing AI to decrease the clinical time spent on prior authorizations.
Hospital leadership continues to access their strategies with some hospitals looking at becoming a leader by developing their own platforms for clinical data and expertise tied to improving patient outcomes.11,12 Within the industry there is an opportunity for health systems to be able to implement and quantify the impact that these new AI systems on patient heath outcomes. At the same time that leaders want to implement these systems, the workers often feel left out of their employer’s strategy.13 The EY Agentic Workplace Survey dives into the relationships and key findings from interviewed employees and their attitudes around AI at their employer.13 Some of the key findings from this survey are that there is misalignment between leadership and employees when it comes to AI success and that the employees are enthusiastic about AI, but want better training and assistance with implementing it into their role.13 Finishing this section, reports from HLTH 2025 further support everything discussed so far indicating that while the trust and maturity of AI has increased, the industry has not yet established strategies to measure deployment success and outcomes.14
Technological Advancements:
On the technology side, we are seeing an increasingly rapid release of new models like GPT 5.1/5.2 and Gemini 3. As models progress, data feeding them becomes more important in the healthcare space. For the first time, the models are reaching something that has been theorized but is just now becoming reality. AI assistance in research is not only assisting with the research, but identifying novel outcomes. The first instance of this which I have seen in the health care space was Google’s Gemma AI for Cancer Therapy discovering a novel drug target to make tumors visible to the immune system.15
As exciting as these advances are, the more prevalent use of AI has also identified the shortcomings.16 The last article warns of "semantic drift," where an AI correctly identifies a general rule but fails to account for medical exceptions.16 It describes a case where an AI system incorrectly suggested a dose reduction for oxacillin in dialysis patients—a mistake that could be fatal if not caught by a human-in-the-loop. This case is particularly interesting because it shows the complexity of what must be evaluated by a clinician when ordering the medication. This could easily translate into the verification process where the medication cannot simply be verified without a “meta” understanding the medication and its patient specific use.
Challenges and Opportunities
The challenges and opportunities are like what has been discussed throughout 2025. Getting the right data to the AI system is still one of the biggest challenges but the policy around this is looking to lower this barrier. One of the biggest changes I have noticed in 2025 is that the discussions are often less about the capabilities of the model, and now pointed more towards the rest of the system around the model. Specific to PGI, as I outlined in another blog post, the probability engines have improved to the point where given the right information, they can match or exceed humans. I predict that 2026 will continue to focus on the flow of data and how that data can be served efficiently to AI systems.
Looking Ahead
2025 was a move from AI being an assistant system to agents. AI is now not only expected to return answers, but also have the ability to act and perform actions. This will continue to develop in 2026 and EMRs are stating that AI will become more integrated into their platforms. Look for more ambient AI (providing assistance without asking) and agentic processes (AI that completes a task). The example of Google’s Gemma AI for cancer identifying a novel drug treatment target is likely the first of many examples where AI will assist with novel discoveries. This will likely create a year with more breakthroughs that could influence PGI. Thank you for reading and as always would love to hear your thoughts and feedback.
1. The White House. Presidential Actions: Ensuring a National Policy Framework for Artificial Intelligence. Dec 11, 2025.
2. The White House. Fact Sheet: President Donald J. Trump Unveils the Genesis Mission to Accelerate AI for Scientific Discovery. Nov 24, 2025.
3. Beavins E. CMS showcases early look at national provider directory. Fierce Healthcare. Nov 14, 2025.
4. Gregory PAM, Austin Z. Responsible Adoption of Artificial Intelligence (AI) in Pharmacy Practice: Perspectives of Regulators in Canada and the United States. Pharmacy. 2025;13(152).
5. Diaz N. Tensions rise over health AI oversight and regulation. Becker's Hospital Review. Oct 8, 2025.
6. Beavins E. Health AI leaders split on utility of AI regulatory sandboxes as Ted Cruz says state AI moratorium still on the table. Fierce Healthcare. Sep 29, 2025.
7. Bruce G. Automation’s effects on healthcare jobs by 2030: 7 notes. Becker's Hospital Review. Nov 24, 2025.
8. Bruce G. Healthcare avoids AI job losses — for now. Becker's Hospital Review. Dec 16, 2025.
9. Diaz N. Hospitals face AI governance gaps heading into 2026, report finds. Becker's Hospital Review. Nov 11, 2025.
10. Minemyer P. Providers believe AI is the key to easing prior auth burdens: survey. Fierce Healthcare. Oct 7, 2025.
11. Mayo Clinic Platform. Homepage. 2025.
12. Diaz N. Trump AI order prompts reassessment of strategy at health systems. Becker's Hospital Review. Dec 11, 2025.
13. EY. Unchanneled worker enthusiasm squanders agentic AI’s promise. Oct 23, 2025.
14. Beavins E. As AI pushes further ahead of governance strategies, only some vendors are stepping up. Fierce Healthcare. Oct 29, 2025.
15. Azizi S, Perozzi B. How a Gemma model helped discover a new potential cancer therapy pathway. The Keyword (Google). Oct 15, 2025.
16. Nodesian. Technically Accurate, Medically Fatal: The AI Error We Caught in Real-Time. Medium. Nov 2025.