Kwame Zaire is a prominent figure in the pharmaceutical manufacturing landscape, bringing a wealth of expertise in production management and the integration of advanced electronics and equipment. As a thought leader in predictive maintenance and safety, he has spent years advocating for data-driven quality control. Currently, the industry is witnessing a seismic shift as major players like Lonza work to transform over 40 years of manufacturing history into a “cognitive CDMO” model. This initiative involves synthesizing data from 140 commercial drugs and 32,000 batch records into a functional intelligence layer. In this discussion, we explore how the intersection of artificial intelligence and human expertise is redefining productivity and setting new benchmarks for process optimization in biotechnology.
Transitioning decades of legacy records into a functional intelligence layer is a massive undertaking. From your perspective in production management, how does a company actually turn thirty-two thousand batch records into a predictive engine?
It begins with a grueling but necessary “cleaning up” phase where we look at the raw history of 3,300 clinical molecules and 1,500 tech transfers. You have to imagine the sheer volume of data points—everything from reactor temperature fluctuations to the specific feeding profiles of raw materials used over the last 40 years. By digitizing and structuring this vast library into what we call an intelligence layer, we are essentially giving the factory a memory that spans decades. It is no longer just about archiving what happened in a lab in the past; it is about using that historical context to identify the specific process-signal combinations that lead to success. When you have the highest number of executed batches in the world, that data becomes your most competitive asset, allowing you to learn from every single outcome to improve future decision-making.
The concept of a “cognitive CDMO” suggests a facility that can learn and recommend interventions. How does the “human-in-the-loop” philosophy function when the AI is the one identifying complex signals?
The “human-in-the-loop” is our most critical safety and quality gate, ensuring that every AI-driven recommendation is anchored in real-world domain expertise. Our Manufacturing Science and Technology experts act as the final evaluators, taking the actionable insights generated by the system and deciding how to apply them to a specific batch. For instance, while the AI might detect a subtle signal in the feeding profile that correlates with a higher titer, it is the human expert who understands the biological nuances and ensures the intervention is responsible and maintains integrity. This approach prevents the technology from becoming a “black box” and instead turns it into a powerful tool that augments the brilliance of our workforce. We want our people focusing on high-level scientific problem-solving, not getting tied up in the transactional nature of manual data sorting.
We have seen reports of AI-guided process optimization increasing output by 15% for large pharmaceutical customers. Can you describe the sensory or operational changes that occur on the factory floor when achieving these kinds of gains?
Seeing a 15% increase in output is a transformative experience for a manufacturing team because it validates the shift toward proactive quality management. In that specific case, we analyzed over 100 historical batches, looking at reactor conditions and production outcomes to find the precise variables that were driving yield. On the floor, this feels like moving from a state of constant reaction to one of total control, where variability is minimized before it can even impact the batch. There is a palpable sense of confidence when the AI can predict performance and recommend targeted interventions in real-time. It eliminates much of the guesswork that has traditionally haunted complex biologics, allowing us to hit higher titers and more consistent yields with a level of precision that was previously unreachable.
Lonza has established a dedicated “AI Factory” to scale and govern these technologies. How do you integrate advanced tools like agentic AI and machine learning with decades of traditional domain expertise without disrupting current production?
The AI Factory acts as a centralized engine that builds and scales these tools while ensuring they are governed by the same rigorous standards we apply to our physical manufacturing. We are integrating generative AI and agentic AI into the workflows to help select experiments and anticipate risks during the development phase, which significantly accelerates tech transfers. By reusing knowledge across different molecules, we can avoid starting from zero every time a new project enters the pipeline. This integration is handled systematically, so the machine learning models are “trained” by our domain experts, ensuring that the software understands the pharmaceutical context of the 140 drugs we’ve brought to market. It is a seamless layering of digital intelligence over physical infrastructure that actually makes the production environment more stable, not less.
What is your forecast for the evolution of the “cognitive CDMO” from 2026 to 2028?
From 2026 to 2028, I expect we will see the total normalization of predictive manufacturing where the “intelligence layer” is a standard requirement for every client partnership. We are moving toward a future where the CDMO doesn’t just provide stainless steel tanks and cleanrooms, but provides a predictive roadmap for every molecule that significantly reduces the risk of batch failure. I forecast that the gains we are seeing now, such as that 15% yield increase, will become the baseline as our AI models ingest data from even more executed batches. Ultimately, the industry will shift toward a model where scientific knowledge and operational data are so deeply connected that the factory itself becomes a learning organism, constantly refining its own processes to deliver life-saving medicines faster and more efficiently than ever before.
