What does agentic AI mean for pharma operations?

AI has already found its place in late-stage pharma R&D operations, as both a practical solution to processing soaring workloads, and a means of transforming the work that companies do.

With growing international geopolitical tensions, and renewed pressures on price and margins in key markets, pharma organisations need to optimise and innovate their delivery of routine tasks wherever possible. AI has significant appeal in this context. Generative AI (GenAI) has already served a disruptive catalyst over the last three years, through its ability to take and combine existing knowledge, distil key facts and new insight, and present these in intuitive new ways. That potential is now being further magnified with agentic AI.

From GenAI to agentic systems

In regulatory affairs and drug safety/pharmacovigilance, a range of pharma companies have now actively leveraged GenAI to streamline labour-intensive activities including marketing authorisation application preparation, product change control/regulatory impact assessment management, adverse event case processing and safety reporting, with a positive impact on operational cost-efficiency. Anecdotally, benefits have included accelerated task execution, honed accuracy and consistency in process output, and large-scale resource optimisation – as capacity has been freed up for more challenging work.

In deploying AI across these specific use cases, function heads have learned a lot about the potential, paving the way for the technology’s latest incarnation – agentic AI. This involves the autonomous coordination of goal-driven AI “agents”, creating unprecedented potential to redefine the way organisations operate and the value they deliver. This is due to the technology’s ability to apply its own reasoning.

With agentic AI, there is much greater autonomy in what AI does and how. Prompted with the desired outcome, individual specialist agents each invoke their own intelligence, experience and reasoning to fulfil their part in the most effective way possible. All of this is coordinated and governed by an “orchestrator”.

The ability to reason, anticipate, generate insight and knowledge and make better decisions is ideal in the pharma industry, because it is so inherently data-rich, process-heavy and outcome-critical. Agentic AI could help to challenge current processes and –help to inform alternative ways of working, or additional ways for teams to add value.

The current state of play

Where pre-agentic AI is enabling new cost-efficiency in R&D functions such as regulatory affairs and drug safety/pharmacovigilance, in discrete use cases, the vision for agentic AI is more ambitious, influencing the nature of the work performed by Safety, Regulatory and adjacent teams.

An emerging example of agentic AI’s potential picks up from efforts to streamline Medical Dictionary for Regulatory Activities (MedDRA) coding of adverse events, with considerable potential to transform the value of pharmacovigilance. Where, so far, AI has helped boost efficiency and accuracy around the classification of adverse event data, with the potential to invoke additional reference cross-checks, or expedite next actions, combining autonomous MedDRA coding with proactive signal triage could take out manual bottlenecks.

Where designated agents detect an unusual combination of coded terms, for example, they could then raise an automated “probable signal” alert along with priority recommendation for human review. The agentic system could also pre-populate a signal report draft (including proposed case lists, timeline and supporting evidence snippets). Benefits here would include a reduction in the time to first credible signal, while experts are freed to concentrate on higher-value work – such as ambiguous/novel cases and investigation design. Harnessing agentic AI could also help enhance human decision-making by honing recommendations for review around emerging high-risk clusters – e.g. for flagging to epidemiology/medical affairs for risk-mitigation actions (e.g., targeted communications, batch holds, enhanced monitoring).

Regulatory opportunities

In a regulatory context, opportunities for agentic AI include reinventing the global management of product regulatory compliance. Autonomous, “regulation-aware” and appropriately structured dossier assembly and submission orchestration is within reach now. Orchestrated AI agents could continuously ingest clinical data packages, study reports, CMC documents, eTMF pointers and legacy submission artefacts.

Agentic systems can also perform automated regulatory gap-analysis versus target-region requirements, draft region-specific CTD/eCTD modules (with citations and traceability to source documents), and orchestrate the technical packaging (file naming, folder structure, etc). Including human expertise remains important, but progression towards greater AI autonomy is about routing suggestions for human review when potentially ambiguous scenarios arise and an expert check is needed.

The agentic system might also generate a short “decision rationale” and a list of recommended human checks, and run a rules/validation pass (file integrity, cross-reference checks, local appendices). This could inform autonomous routing of items to subject experts (e.g., CMC, clinical, labelling) with suggested edits and severity scores – providing human reviewers with a near submission-ready dossier.

Ultimately, shorter regulatory cycle times would help accelerate go/no-go decisions and, thereby, patient access; sponsors meanwhile would be able to iterate protocols more swiftly. Agents’ gap-analysis outputs could be fed upstream to clinical operations and protocol teams, too, enabling trials to be designed that need fewer regulatory clarifications over time.

Harnessing agentic AI to best effect

Any plan to deploy agentic AI assumes that the organisation has a strategic rather than tactical vision for AI; ideally one that capitalises on the technology’s cumulative benefits across more than one use case.

This requires a more embedded and systematic approach to deploying the technology. Agentic AI proffers the benefits of autonomous reasoning and decision making, as well as continuous adaptation, in reaching defined goals. The total benefits should multiply as respective agents continue to hone what they do, based on their own deductions or new insights.

Beyond core requirements around good data (AI output can only ever be as good as the information it is given to work with, and how readily this can be combined), teams interested in exploiting agentic AI must consider how to establish and foster trust around AI reasoning.

Where AI systems are assigned greater autonomy across extended workflows, potential risks could be more than just incorrect outputs – e.g. unintended data movement, loss of operational control, misaligned decision-making, and blurred accountability. At the same time, however, companies need to avoid being overly prescriptive and limiting in their attempts to establish good governance (ideally, the latter should serve as a facilitator as well as a mitigator of risk).

A principles-based approach, rather than one that is hard-wired around specifics, is recommended here, and will help process stakeholders to think through scenarios and goals that agentic AI can help address. Companies can then supplement these principles with their preferred service-design methods – perhaps journey maps laying out how agentic workflows will behave and evolve over time.

Ideally, trust will be cultivated by the infrastructure and its design, controls and transparency, rather than rely on static checklists that might date quickly. This also leaves scope to adapt the degrees of autonomy assigned to individual AI agents, encouraging more trusted reliance on the technology over time so that the benefits are properly felt.