PDA/FDA Week 2026: Full Notes, Recap, Takeaways and Summary

This year, the PDA/FDA Joint Regulatory Conference felt back in full swing.

Looking back over the last several years, 2019 is my reference point for a full conference week. COVID disrupted that rhythm, and by 2023 it felt good to be back. The event still used the joint PDA/FDA name in 2024, while 2025 ran as the PDA Regulatory Conference. In 2026, the joint name was back. PDA: 2024, 2025, 2026.

KeyPedia notes: 2025 recap, March 2026 PDA Week.

TLDR

If there are three short, simple takeaways from this conference, they are that the “new” FDA is modernizing, embracing new technology, and adopting the policies it is directed to implement.

1. Onshoring and America first. Whether you like the policy direction or not, the repeated emphasis on domestic manufacturing was clear. PreCheck and the ANDA prioritization pilot were two concrete examples.

2. Communicate. Use new technology, and communicate how you use it. FDA will not prescribe every implementation. There is guidance, and companies need to explain their intended use, validation, and controls. That was the message I took from the discussion.

3. FDA is using AI. Expect more emphasis on efficiency. Even the government is trying to get more done with its resources. Looking back through nearly 30 years of GAO work, many of the concerns about inspection coverage, data, and follow-up are familiar. FDA, GAO, 1998.

These are my notes and observations from the sessions I attended on September 14 and 15. Most center and office remarks focused on specific initiatives rather than raw postmarket trends and statistics.

Monday, September 14, 2026

Monday plenary, September 14, 2026.

P1: Human error and procedures

Amy Peterson, Lilly

Lilly and Takeda had a strong presentation presence and a focus on technology demonstrations in the sessions I saw throughout PDA/FDA. These were valuable examples of approaches built around technology and integration.

Amy's presentation focused on the predictability of errors and how procedure design can reduce the factors behind them. I particularly enjoyed the slide on Rasmussen's skill, rule, and knowledge framework, and the discussion that followed about technology, upskilling, and the demands different tasks place on people.

Amy Peterson's human-performance presentation, September 14, 2026.

Panel discussion

Janeen Skutnik-Wilkinson moderated the panel with Patrick Boyle, Amy Peterson, Ernest Bizjak, and Jeffrey Meng. Agenda.

One topic of discussion was checklists. Their effectiveness depends on what they are used for and how they are used. Amy emphasized explaining why the checklist exists. Ernest discussed site walks and observing how people work. Janeen brought up organizational constraints and pressures. Jeffrey asked how often companies involve the end user in drafting and designing the procedure.

Amy's IKEA example made the distinction between a checklist and a manual easy to understand. When assembling furniture, you need the assembly steps. The history of the wood belongs elsewhere. Task analysis matters when deciding what the person doing the work needs in front of them.

The inspection discussion also connected a failure to follow a procedure with broader questions about the adequacy of the procedure or investigation. The provisions discussed included 21 CFR 211.22(d), 211.192, and 211.100(a). Jones's later FDA presentation lists these same provisions among common quality-system deficiencies.

My takeaway from the checklist discussion: checklists are useful, but they need to reflect how people actually work. The design of the instruction and the way it is used both matter.

P2: Center and office updates

Paul Balcer, moderator

Michael Davis, CDER

As mentioned in the TLDR, Davis discussed PreCheck, the ANDA prioritization pilot, and advanced manufacturing across science, policy, assessment, and inspection. These programs made the emphasis on domestic manufacturing concrete.

PreCheck has two phases: early facility readiness, followed by application submission. The goal is earlier technical and regulatory engagement around new U.S. manufacturing facilities. FDA selected seven companies after receiving more than 80 requests to participate. Jonathan Patroni of Kyowa Kirin gave a useful industry perspective on the program's selectivity. FDA's announcement confirms both the numbers and Kyowa Kirin's participation. FDA.

Davis also discussed Operation TrialBlazer and its effort to accelerate first-in-human trials, including an Expedited IND Pilot.

Michael Davis, FDA. Slide 11.

Michael Davis, FDA. Slide 12.

Timothy Schell, CVM

I thought Tim had one of the clearest goals. My notes capture the ambition as halving the time to approval and doubling approvals. His published slides frame the goals as more approvals and faster access to treatments. The three-year plan shown in the deck applies specifically to the MUMS program for 2026–2028.

He also discussed CVM's reorganization and a lifecycle approach to animal foods. The organization chart shows separate offices for Animal Foods and Veterinary Products Surveillance and Compliance. The Animal and Veterinary Innovation Centers program supports long-term academic partnerships through cooperative agreements, with five-year projects subject to funding availability.

My reservation about increasing approvals is that FDA cannot determine the number on its own. The number and readiness of applications also matter. I would like to understand how much of the proposed increase comes from faster review and how much depends on the development pipeline.

The discussion of animal testing was also interesting. Tim distinguished testing for human products from testing veterinary products in the animals they are intended to treat. I appreciated the measured approach to the public debate and the scientific purpose of that testing.

Timothy Schell, FDA. Slide 7.

Elizabeth Miller, OII

For me, this was the most important update from KeyPedia's perspective, given our work with postmarket data. Miller began with OII's organization chart and addressed the idea that FDA was returning to a universal or generalist investigator model. My notes were clear that product-area expertise would remain.

She discussed the practical challenges of developing and retaining investigators, including training time, retirements, and the loss of institutional knowledge. The new Office of Inspectional Performance Optimization was another important part of the update. It appears in the published organization chart.

Miller described a meaningful increase in unannounced foreign inspections. That direction follows FDA's announcement on May 6, 2025, under then-Commissioner Martin Makary. Our paper, The Foreign Inspection Gap, examines the historical differences in FDA GMP oversight of U.S. drug imports from 2014 to 2024 and provides context for that discussion. FDA announcement, our research paper.

My question for Miller was how OII would measure success beyond simply doing more inspections, and whether the new optimization office would be responsible for those measures.

Elizabeth Miller, FDA. Slide 3.

Karim Mikhail, CBER

Karim's remarks focused on talent, hiring, and his satisfaction with progress, with much of the update centered on staffing and capacity.

Center and office roundtable

Paul Balcer moderated the roundtable, with questions selected from the audience. PDA.

Topics included investigator training, the Federal Law Enforcement Training Center, and preparing staff for the field more quickly. The discussion also covered how OII keeps pace with advanced manufacturing, and when older guidance needs to be reconsidered in light of current expectations for safety and efficiency.

Emerging technology moving into surveillance was another theme. We have separately covered examples in the HHS AI use-case inventory.

Center and office updates panel, September 14, 2026.

C1: FDA and industry using AI

Shawn Larson, moderator

The session focused on real-world uses of AI, governance, and the review of inspection documents. Agenda.

Hesha Duggirala, FDA

This presentation was all about Elsa. Hesha explained its purpose and design and addressed questions companies are asking, including whether AI is autonomously targeting their facilities for inspection. Her slides say no.

She distinguished Elsa from other tools available to FDA staff, including ChatGPT at HHS and Claude at HHS. Elsa uses Claude Sonnet and Gemini models in an environment described in the presentation as FedRAMP / FISMA High. That description concerns Elsa's deployment, not a claim that every FDA AI tool uses the same models.

“Document collection is not the same as AI review.”

Hesha Duggirala, presentation slide 5.

I thought this was one of the clearest points of the session. Investigators request and collect documents. AI can help them review large amounts of information with less repetitive work, but the investigator still evaluates the evidence and owns the conclusions.

The data question needs care. FDA's published privacy assessment says FDA data is not used to train the underlying models. It also says prompts, uploaded material, document libraries, and generated outputs are retained under applicable records schedules, with access controls for individual workspaces and shared libraries. I would not describe that as a guarantee that information is never stored or that the underlying infrastructure has not changed. Elsa privacy assessment.

Hesha also set out questions industry should be able to answer in writing: what does the tool do, what data does it act on, and what decisions does it inform? That leads directly into intended-use validation, controls on inputs and outputs, change control, and documented human review.

The presentation's limits were equally clear: Elsa is not a regulatory decision-maker or an authoritative source. Outputs require verification. The AI Internal Council governs FDA's internal use of AI, which is separate from FDA's oversight of industry uses.

Hesha Duggirala, FDA. Slide 7.

Hesha Duggirala, FDA. Slide 15.

Hesha Duggirala, FDA. Slide 17.

Karthik Iyer, Lilly

Karthik focused on AI for inspection readiness, the design decisions that come before implementation, and building for the longer term. His examples included connecting internal and external findings, preparing for PAI/PLI, and balancing the volume of information against the decisions people need to make.

He also acknowledged pressures from vendors and leadership that may not fit a company's long-term plans for its tools. That led into the panel discussion.

AI panel: Tara Gooen Bizjak, Hesha Duggirala, Jeffrey Meng, and Karthik Iyer

I found this panel engaging because industry quality leaders and regulators could discuss how the tools were actually being used. The discussion of Lilly's return on investment focused on time saved and the value of getting to market sooner.

I also think there is a workforce expectation here. Younger or less experienced staff may expect these tools to be available. That creates a practical reason to give people appropriate tools and clear controls around their use.

One of the most pointed examples in my notes came from Tara: imagine an investigator looking through roughly 2,000 deviation entries. Before Elsa, the investigator reads the list and highlights possible issues. With Elsa, the tool helps review the list and identify trends. The investigator still has to examine the underlying records, ask questions, and establish the observation. That is consistent with the human-review process described in the FDA slides.

Jeffrey also emphasized that Form 483 observations are the investigator's observations, rather than final agency determinations.

FDA and industry AI panel, September 14, 2026.

Tuesday, September 15, 2026

P3: GMP compliance trends

Francis Godwin, FDA

Godwin was one of the clearest and most knowledgeable presenters I heard. He treated the session as an investment in educating industry and used concrete enforcement examples to do it.

One example concerned microbiology data integrity at an ophthalmic manufacturer. The case involved discarded plates, backdated documentation, manipulated camera timestamps, and an endotoxin failure that had not been documented or investigated. The firm's response then attempted to characterize a promised third-party assessment as an internal audit.

One phrase I wrote down from Godwin's panel discussion was:

“Then trust is destroyed.”

Francis Godwin, as captured in my contemporaneous conference notes, September 15, 2026. This wording has not been checked against a recording.

He distinguished weak controls around otherwise reliable data from evidence that the data had been changed or manipulated. My notes connect that distinction with VAI versus OAI, but it should not be treated as an automatic classification rule. The larger point was the loss of confidence in what the firm says once the evidence itself is unreliable.

The case also addressed controls in microbiology laboratories. The response FDA sought included third-party oversight and photographic evidence with electronic controls to prevent manipulation. Those were measures discussed for this specific case.

Francis Godwin, FDA. Slide 27.

Francis Godwin, FDA. Slide 28.

Francis Godwin, FDA. Slide 29.

Francis Godwin, FDA. Slide 30.

Kathleen Jones, CBER

Kathleen went into pre-license inspections, biologics license applications, and DMPQ's role, as well as the distinction between a product-specific inspection and broader surveillance work with OII.

One example that stood out in my notes involved alarms during pre-license inspections. Silencing the alarm does not resolve the underlying condition. Her published slides call for maintained equipment records and for alarms and deviations to be investigated and resolved.

She also discussed how inspection outcomes affect recommendations on approval or a complete response letter, and again mentioned PreCheck and earlier engagement.

The presentation boiled the assessment down to seven major systems and three critical elements: SOPs, training, and records. Are the procedures adequate and followed? Is the work documented? Can the team manufacture consistently? That last question gets to the core of the GMP side of approval. I would have liked more discussion of collaboration around combination products.

Kathleen Jones, FDA. Slide 5.

Kathleen Jones, FDA. Slide 7.

Inspection panel

Ivy Sweeney, Francis Godwin, Kathleen Jones, and Dillard Woody joined the discussion, with Tara Gooen Bizjak moderating. Agenda.

The panel covered unannounced inspections, trust in data, and the root cause behind CAPA. Godwin discussed the problem of a CAPA that leads into still more CAPAs without resolving the underlying issue.

The sponsor and CMO/CDMO relationship was another important topic. An observation involving another product on a shared line can still matter to the sponsor. My notes emphasize both parties' responsibilities and the information sharing that needs to be addressed in the quality agreement. The sponsor may need the Form 483 or supporting information to understand the issue and the response. Information sharing becomes difficult in these situations, but that is also where it matters.

Consultants and independent third parties also came up. The panel emphasized qualifications and scope. Experience with one type of product does not automatically establish expertise with another. If a firm's response relies heavily on a consultant, the person's qualifications and the actual assignment need to support that response.

GMP compliance-trends panel, September 15, 2026.

B3: Internal audits

Helen Motamen, Sanofi

Helen discussed the position of power an auditor holds, and the importance of humility, integrity, and resourcefulness. I particularly liked her point about entering a room quietly, observing, and understanding the work before jumping in.

She also covered planning factors: product risk, time since the last audit, staff turnover, facility size and age, emerging issues, and inspection history. Time on the shop floor and observing gowning connected back to the first plenary. Humanizing the process matters in the audit as much as it does in the checklist.

Local and corporate audits should both happen, with corporate audits offering a way to compare issues across the network. The discussion also covered how new technology changes what auditors need to understand and review.

Yvonne Lozano, AbbVie

Yvonne followed Helen and continued the discussion through root-cause investigation. It fit well with the regulatory remarks earlier that morning. She covered tools such as five whys, fishbone analysis, and fault trees, and the need to distinguish a symptom from its cause.

“Link observation to root cause to actions.”

Yvonne Lozano, presentation slide 10.

That captured the point I wrote down: corrective actions should address the identified cause.

Audit panel

Jonathan Chapman, Maan Abduldayem, Helen Motamen, and Yvonne Lozano took questions, with Karyn Campbell moderating. Agenda.

The discussion covered the differences between local and corporate audits, and the challenges of classifying findings and communicating them.

One of the most engaging parts concerned how much of the follow-up the auditor should handle. Should the auditor stay involved through the response, or hand over the observation? There are benefits and burdens to each approach, especially in an internal audit. The discussion considered different levels of involvement while maintaining independence from the CAPA.

Maan flagged retraining as a weak response when it does not address the cause.

Another useful question was how to capture audit findings that could also become regulatory findings. Whether the company calls a finding critical, major, minor, or something else, the issue still needs to be captured and resolved. Helen emphasized raising and remediating findings. Jonathan noted that a documented gap could become a regulatory finding, which makes identifying it worthwhile. Communicating the issue and the response across sites is part of that work. Yvonne's slides also address when corrections should extend beyond one site.

Internal-audits panel, September 15, 2026.

Super summary and recap

PDA/FDA brought a wealth of knowledge this year. Beyond the sessions and hearing directly from regulators, the room was full of professionals and managers who need to bring that knowledge back into their organizations.

Compared with the last few years, I heard less fear and more understanding of AI and chatbot tools. Both internally built systems and SaaS tools remain useful. Alongside that interest, there was a stronger appreciation of the need for appropriate, reliable data and a clear understanding of how the tools are used.

PDA/FDA 2026, Washington, DC.