Our ambition is to look at each and every step along the way, from discovering a drug all the way through to making sure it is developed in a safe way, manufactured properly, supplied effectively, and ultimately reaches patients in a more personalized way. We’re trying to use AI to significantly reduce drug development timelines. So however long it takes today, which is often upwards of 10 years to go from identifying a drug candidate to actually getting it to a patient, we’re working to get these miracles to patients faster.

That means embedding AI into different parts of our organization. We look at AI in three parts.

First, there’s “expert AI.” These are more specialized areas of AI designed to help experts leap ahead in how they operate. The most obvious applications for us are in science,  how we can give scientists tools that help them do much more in terms of analysis, like hit selection and drug discovery. But it’s not just limited to research and development. You can also think about use cases in manufacturing plants, where we’re using AI to increase yield.

Second, we have “snackable AI.” The idea here is that AI is at everyone’s fingertips, helping them in their day-to-day operations. Take budgeting, for example. Instead of following a conventional annual or monthly process, can we use all our historical data and future projections to create an AI-augmented version? The concept behind snackable AI is to change workflows and how each of us approaches our work.

Third, GenAI has been buzzy because it is creating a new lever for us. We’ve identified key areas to explore, and we’re going after bold ideas where entire processes can be reimagined. Instead of running a bunch of small proofs of concept and hoping some scale up, we selected meaty areas of the business and asked, “If we reimagine this process from scratch with GenAI, what would it look like?”

One big area is content generation. Since we operate in many markets, generating medical information and making sure all the checks and balances are in place is critical. People need to understand what our medicines do and how they are different.

It is a massive undertaking, especially when you consider all the documentation we need to produce. Every manufacturing site has a wide range of regulatory documents. Moving people away from just filling in blanks and generating documents, toward actually pulling insights from those documents and changing the way they work, is what we are aiming for.

Our GenAI journey has focused on finding those key areas where we can engage the broader organization. If you think of AI as a new source of light, we’re not just asking who has the best lightbulb. We’re asking how this new source of light can change the way we live our lives.

Agentic AI is going to help us go even further. It will let us embed more GenAI into workflows, enable agents to interact with one another, and help us identify the right spots to have humans in the loop.

We started thinking about it almost two years ago and we launched a program around it. We were looking at it from the perspective of: How do we make sure that we truly understand where AI is being applied, so that we can bring responsible AI to everyone at Sanofi?

It’s rooted in different elements. One is that you are ultimately accountable for the outcome the AI produces. Whether you are the developer creating it or the person utilizing the AI insights, you, as an individual, are accountable for those outcomes.

If you start with that fundamental principle, you can move into the next layer. In that case, the AI needs to be transparent and explainable. It needs to be eco-responsible, fair and ethical. It needs to be robust and safe.

Once we had those tenets in place, it became about understanding, just like with any other technology, that depending on the level of risk and how you’re utilizing it in the business, the question becomes: What technical controls, process controls, governance controls, and training controls need to be in place so that people can use AI effectively?

For us, it wasn’t about creating something overly heavy or blocking everything along the way. It was about truly being thoughtful about the risks we’re taking with AI and then managing the level of oversight based on that level of risk.

The jury’s still out in terms of how AI regulations are going to form. But between different regulatory bodies and across different countries, you’re going to start to see a pattern emerge. Whether it’s through an industry-based approach or a geographic one, some structure will take shape.

But the crux of it is that AI will be linked to its actual application or where it’s being used, and we should regulate it or treat the risks accordingly. If you look at our industry specifically, the FDA just released some guidance on how AI is going to be used in processes like clinical development and drug approval. You’re starting to see regulators step in, embrace the opportunity, and still manage the risks underlying it.

CDO Magazine appreciates Dimitrije Jankovic for sharing his insights with our global community.