.png)
Over the past months, Aizon's Co-Founder and Chief Compliance Officer Toni Manzano and VP of AI Consulting Services Kevin Baughman delivered a series of three webinars on agentic AI in pharma Quality:
- Accelerating Batch Release with Agentic AI
- APQR on Demand: How Agentic AI Turns Months of Work into Minutes
- Power to the Experts: Agentic AI for Pharma Quality & Compliance
Each session looked at a different Quality challenge. Toni set out the regulatory and methodological context, and Kevin showed how Aizon Agentic Studio addresses it. Taken together, the series points to one conclusion: the tools to transform Quality are already here, and the question now is how to use them well.
Here are five lessons that came up again and again.
1. Regulators are not the bottleneck
One of the most common concerns from attendees was whether regulators would allow AI in GMP environments. Toni's answer was the same in every session: "Please believe me, regulators are not blocking anything."
In fact, he argued, this is the first time in history that regulation has moved ahead of the industry. Regulators are setting out how to use AI responsibly, not whether it can be used. They have been calling for modernization for years, and they increasingly use AI themselves.
The approaches do differ. The draft Annex 22 excludes generative AI from critical GMP decisions, while the FDA takes a risk-based approach regardless of model type. But the direction is clear, and Quality teams that wait for absolute certainty risk waiting too long.
2. The real barrier is access to data
What is actually holding many Quality teams back is something else: data.
"Data itself is not information," Toni reminded attendees. A thousand pH readings mean nothing without the batch, the phase, the equipment and the timestamp behind them. Without that context, most manufacturing data stays locked away, unused and unanalyzed.
The live polls confirmed this:
- Most attendees estimated that between 25% and 75% of their manufacturing data goes unused.
- In the APQR session, 65% named collecting and preparing data as the biggest time sink in the process. Only 14% pointed to analyzing it.
- In the batch release session, most attendees said gathering the data for a single root cause analysis takes days.
In many companies, every batch comparison, export or report goes through IT, which quickly becomes a bottleneck. Toni described the shift that's needed as moving from an IT-centric model to a data-centric one: IT ensures data is clean, contextualized and trustworthy, and the experts who need it can use it directly.
That's the foundation Agentic Studio is built on. It works on top of Aizon's GxP data fabric, where data from historians, LIMS, ERP, MES and batch records is already contextualized around the batch. Agents work from validated, structured data rather than raw exports.
3. Agentic Studio turns Quality experts into builders
An agentic system plans, acts, checks its own results and iterates. In each session, Kevin showed what that means in practice, with Quality applications built in Agentic Studio from plain-language prompts:
Batch release. From a single prompt, Agentic Studio built a batch release dashboard comparing the latest batch against top-performing historical batches. It flagged the parameters that were out of specification and recommended holding the batch. Agents can also review batch records against SOPs and highlight the areas to investigate, turning thousands of pages into a prioritized list in minutes.
APQR. The annual product quality review was designed to drive continuous improvement, but as Toni put it, it often becomes "a bureaucratic check." Gathering the data alone can take 70 to 120 hours per product, pulled from five to seven different systems. Kevin showed a complete APQR application, built in Agentic Studio from one detailed prompt, that drafts every section of the review in minutes: materials, in-process controls, OOS and CAPA, process changes, stability and environmental data. Individual sections can then be refined in natural language and published with access controls. As Kevin noted, the long-term goal is a continuous product quality review rather than an annual one.
Expert-built tools. In the final session, Quality SMEs saw how to build their own applications in Agentic Studio, with no coding and no ticket queue: a batch comparison cockpit for a CAPA investigation, an SPC control chart application, and an environmental monitoring view that predicts excursions before they happen.
As Kevin put it, Agentic Studio is "a massive accelerator." It gets an application most of the way there in minutes, so experts can spend their time refining and reviewing it, not building it from scratch.
4. Humans stay in charge
None of this means handing quality decisions to a machine. Toni used an analogy familiar to anyone in the industry: when a CMO prepares a product quality review, the sponsor remains accountable for it. In the same way, the agent prepares the draft, and the expert stays accountable for the decision.
He went a step further, preferring to talk about "human oversight" rather than "human in the loop." Experts should be in control from the very beginning, defining the intended use, the rules and the guardrails, not just reviewing outputs at the end.
Agentic Studio is designed around that principle. Guardrails limit what an agent can be asked, ground its outputs in validated data from the data fabric, and check its results. Applications built by SMEs follow a clear path from prototype to validated tool, running on the Aizon platform with 21 CFR Part 11 controls and a full audit trail. Combined with sound risk assessment under ICH Q9, that's what builds trust with quality teams and inspectors alike.
As Toni put it, humans are very creative, but repeating the same thing 24/7 in a systematic, objective way is work that is best left to machines, so people can focus on judgment.
5. The effort is an investment, not a recurring cost
The final lesson is about expectations. "This is not magic," Toni told the APQR audience. "The only magic is the effort."
That effort goes into contextualizing data around the batch, using standards such as ISA-88 and ISA-95, and into defining templates and rules that reflect your SOPs. Once the foundation is in place, the same data and Agentic Studio can support APQR, continuous process verification, real-time monitoring and much more. As Toni and Kevin put it in the Q&A, the effort is one-time and the benefit keeps growing.
And the purpose behind it never changes. Whatever tools Quality teams adopt, Toni reminded attendees, the final driver must always be the patient.
Thank you to everyone who joined and asked such insightful questions!
Watch the full series
All three webinars are available on demand:
