From The Editor | January 20, 2025

The Intelligent CDMO: A Vision with Agentic AI

louis-g-photo-edited

By Louis Garguilo, Chief Editor, Outsourced Pharma

AI Robot Handshake collaboration-GettyImages-2191345833

Agentic AI is the next artificial-intelligence iteration for the biopharma industry.

It propels prediction to autonomous decision-making. It can, for example, usher in transformative efficiencies to contract development and manufacturing organizations (CDMOs).

If readers have day-dreamed a more effective external-support industry for your individual drug-creation activites, this may be a a ticket to that destination.

We introduced Agentic AI in part one. Here we’ll look at how it might impact outsourcing through all stages of development and manufacturing.

Agentic AI in Drug Development and Manufacturing

Novartis and Microsoft have a collaboration (first announced in 20219) to “transform medicine with artificial intelligence.”

GSK has a devoted website with a simple url: https://gsk.ai.

These are just two examples of ongoing efforts that, as we’ve discussed previously, began as the generative AI phase of biopharma with applications devoted to discovery and then route-scouting in the development phase.

Interestingly, FDA draft guidance for comment on AI use in support of regulatory decision-making (January 2025), does not specifically mention Agentic AI

But it cannot be long before Agentic AI moves artificial intelligence through the entire drug/therapy life cycle, and into the world of outsourcing.

Here are some thoughts on how this might apply to sponsor-CDMO decision-making.

Similar to generative AI systems, Agentic AI (perhaps in a “collaboration” with generative AI) will closely monitor processing/manufacturing, but then also autonomously adjust for efficiency, quality, and yield.

Agentic AI will look to adjust whole supply chains, based for example, on market demand metrics, production capacities, new partnering opportunities, and regulatory constraints considerations.

Speaking of the regulatory environment, again similar to generative AI systems, Agentic AI will autonomously analyze regulatory updates, ensuring compliance across jurisdictions, but also, for example, employ modifications in behaviors,  processes and/or documentation.

This last point may be a last straw for some readers.

“Hold on,” you say, “no way this autonomous whatever will fly with the FDA. And if something goes wrong, guess who’s responsible, and it ain’t any AI entity.”

Those are valid, delimiting concerns. Patient safety (and liability) first; any AI after.

And consider this:

Even if Agentic AI is pursued, can you imagine a GSK.ai organization saying something like, “We employ AI to autonomously alter the production and safety profiles of our drugs”?

None of these remonstrations gets anyone labeled a laggard “AI skeptic.” We’re just beginning to work through this new technology.

But as I began in part one, "AI" is already embedding in our industry. Further AI enhancements will be natural descendants we need to take seriously.

The Intelligent CDMO

I don’t purport to understand many of the above implications, or all the realities (or unrealities) of Agentic AI. (“You aren’t kidding,” you may be thinking.)

Nonetheless, Agentic AI certainly has the potential to transform CDMOs into highly adaptive and more, well, intelligent partners.

By enabling autonomous optimization – and some decision-making – for processes, resource management, quality parameters, quantities, batch sizes, etc., as well as client collaboration, CDMOs will significantly enhance their value proposition.

Continuous monitoring by AI agents could maintain a state of perpetual compliance, allowing CDMOs to address issues before audits occur.

Think of such areas as project performance analytics. Agentic AI could analyze project outcomes and output improvements for future work based on aggregated insights.

Insights could include a mix of production analysis based on the specific capabilities and availabilities of your CDMO. For example, a silico-based intelligence at a CDMO suggesting and implementing process changes to components of a client’s monoclonal antibody production.

IT specialists like to call this automated feedback loops.

For our industry, that might entail post-manufacturing data, with AI agents autonomously collecting and analyzing feedback on parameters such as productivity and therapy performance, and utilizing this information for future optimization or next-generation therapies.

Earlier, Agentic AI could autononously iterate and optimize upstream and downstream parameters to meet yield and quality specifications.

We saluted our advanced thinking (i.e., human intelligence) when we began to focus on real-time analytics, e.g., PAT systems. Tomorrow we’ll jump for joy when these activities integrate with an automated intelligence optimizing parameters in a moment’s notice.

For advanced therapies like cell and gene, AI agents could adjust production workflows for individualized therapies, ensuring compliance with regulatory requirements.

AI agents will monitor production data, detect deviations in real time, and either self-correct or alert human operators.

Consider a world benefitting from autonomously generated regulatory documentation (CMC? Ah, no big deal).

Who knows? Maybe there’s a future where the FDA has its own AI agent reviewing (and suggesting alterations to) your submitted applications.

Consider These Use Cases

Consider for the moment a tech transfer.

At the CDMO, Agentic AI could autonomously analyze client-provided data, and draw conclusions based on work completed/underway to date; advise the sponsor on the requisite materials and availability/ pricing; and immediately draw up realistic timelines given the CDMO’s capacity/equipment/workforce availability.

We can take a step back: How about Agentic AI generated RFPs before those CDMO communications arrive?  

Picking on GSK again, Agentic AI could recommend CDMOs suited for scalable production of biologics or ATMPs, using predictive models to gauge each (potential) partner’s capacity and compatibility versus GSK’s demands.

We do have to interject a word here on the communication front.

Nary an editorial or article on the subject of sponsor-external parner equation will not include admonishing to focus on the human-to-human personal connections – the superglue, we are told and have experienced – keeping sponsor-provider relationships together.

Will we enter an era of “My AI agent is talking to your AI agent”? Would that increase and qualitatively improve a CDMO-sponsor relationship?

Perhaps we’ll give birth to a future of intelligent collaboration portals driven by Agentic AI.

These might go well beyound simply improving supply-chain communications – although that alone is a worthy goal – but also lead to better thought processes and decision-making that improves the entire life cycle of our products, and brings forth better medicines for our patients.

We are an industry of stringent testing and GMP standards. Of double-blind placebo-controlled clinical trials. Of stiff regulatory oversight.

We should be able to figure out how to take in the newest technnologies knocking on our doors.

Let’s hope that here in biopharma land, these types of advancements breathe fire into what for years we’ve been calling Pharma (or Industry) 4.0.

---------

Also see:

Autonomous Decision-Making: Are We Ready for Agentic AI