It turns out that AI really is doing the work. At least a growing chunk of it.
Artificial intelligence is spreading through the global workforce like wildfire, dwarfing the adoption pace of the internet, with a new Anthropic report citing Gallup research that 40% of U.S. employees now use AI at work. While the field is volatile, the steady drumbeat of AI adoption has little comparison among nascent adoption curves. The growth trajectory surpasses even the internet’s early trajectory, which took around five years to hit similar penetration levels.
“We hear ‘AI supports, not supplants’ at every conference, but some roles will be downsized.”
While many professionals use generative AI as a collaborative assistant, a new report from AI research firm Anthropic shows that businesses are deploying it with a more disruptive goal: automation. The study found that 77% of enterprise AI use (at the API level) involves delegating complete tasks to the company’s AI models while the firm said 79% of conversations on its Claude Code development platform were automation-based.
Marie Flanagan
This pattern extends into pharmacovigilance. As automation expands in intake and routine case handling, the bottleneck is hybrid talent that can bridge AI, data and regulated safety workflows. Labor-market analyses indicate that AI skills carry sizable pay premiums. But at the same time, automation is making some simple tasks such as data-entry obsolete. “Some roles will be downsized. In PV that’s intake and data management, but we’re retooling people for higher-judgment work like medical review and signal detection. Human oversight isn’t going away,” said Marie Flanagan, Regulatory and AI Governance Lead at IQVIA Safety Technologies.
In such of AI/PV unicorns
The evolution of these hybrid roles is something Flanagan knows firsthand. “I’m in one of those hybrid roles myself,” she explains. “My background isn’t AI, and I’ve been clear about that. I’d rather learn AI than learn PV from scratch, and I think many would agree.”
While tech companies debate ethical AI frameworks in abstract terms, pharmacovigilance operates in a world where a single misconfigured prompt could compromise patient safety. Flanagan describes this as a third circle beyond coding and biology in a Venn diagram: the trust layer that binds technical capability to regulatory compliance. At IQVIA Safety Technologies, this translates into cross-functional teams aligning around concrete production questions before any AI model touches real safety data.
“77% of enterprise AI use involves delegating complete tasks to AI models”
“There’s a real need for a multidisciplinary approach in drug and device safety, especially because regulators now expect cross-disciplinary validation: PV experts, data scientists, AI scientists, validation specialists, etc.,” she said. “With generative AI, you need this team from design through development, deployment, and routine use. One person can’t embody it all.”
AI filters the noise but can’t judge the signal
Signal detection represents both the greatest opportunity and the steepest learning curve for AI in pharmacovigilance. With unstructured real-world data growing exponentially, GenAI excels at filtering out the noise, potentially eliminating large swaths of irrelevant information and surfacing patterns across years of safety databases, literature and registries. Yet the technology hits a hard wall at summarization.
“I see eye-watering salaries for these roles, but the supply lags industry needs. There isn’t one course that delivers what PV needs right now.”
While some engineers dismiss “prompt engineering,” in a regulated field like pharmacovigilance the prompt is part of the system’s behavior and must encode domain constraints (e.g., seriousness criteria, MedDRA/WHO-ART terms, required citations/traceability). Flanagan offers an example. “One of our AI agents is an in-product assistant. Some say, ‘It’s just prompting; how hard can it be?’ But prompting in a regulated industry is daunting. You need substantial domain knowledge to engineer safe, effective prompts. People in our industry are cautious because at the end of the line is patient safety.”
The paradox of AI adoption in pharmacovigilance is that the solution creates its own crisis. “Signal detection is a huge opportunity, met with trepidation,” Flanagan said. GenAI’s ability to scan social media, patient forums, medical literature and electronic health records means it will surface exponentially more potential safety signals than traditional monitoring ever could. “There’s a lot of unstructured real-world data, much of it noise. Distilling large volumes into something meaningful is a great use case.”
When more detection power unearth more problems
“If you imagine a Venn diagram of coding and biology, I’d add a third circle: governance. Governance is the trust layer that ties it together.”
Signal detection scales the work as much as it speeds it. As Flanagan puts it, GenAI can “aggregate and summarize at speed,” but she draws a line: “GenAI should distill and summarize—not form hypotheses or make critical judgments about potential safety signals.” The staffing pressure is real: “Today there are 50,000 PV professionals globally… Without AI, you might need 100,000 people in five years,” she said.
Flanagan is clear about the culture gap: “Some people hope if they don’t look directly at it, it might go away and not take their job.” She argues that change management—not tooling—is why many pilots stall: “The cultural mountain shouldn’t be underestimated.” Her prescription is practical: “If someone asks how to reskill, I say: volunteer for an AI pilot or integration project—get hands-on.” The team model is equally explicit: “There’s a real need for a multidisciplinary approach… PV experts, data scientists, AI scientists, validation specialists,” with regulators “effectively saying: recruit or upskill teams to enable the critical thinking a regulated PV environment demands.” And the glue is governance: “If you imagine a Venn diagram of coding and biology, I’d add a third circle: governance. Governance is the trust layer that ties it together”—from design through deployment and routine use.
Filed Under: Pharmacovigilance
![[Adobe Stock]](https://www.drugdiscoverytrends.com/wp-content/uploads/2024/11/AdobeStock_844651123-copy-268x170.webp)
