In 2011, I attended the fourth iteration of the Massachusetts Life Sciences Innovation Day (MALSI) in Boston. I was early in my career, working with a postdoctoral fellow to commercialize the technology we later patented. We were looking for funding. (We didn’t get it.)

The event was held at the Harvard Club, which was a stubbornly unmodern venue, and presumably kept that way for prestige purposes. It was my first real exposure to the life sciences industry.

Two talks stayed with me.

One speaker had just sold his startup for roughly $500 million to a large pharma company. The other, a pharma executive, confidently declared that research could essentially stop. Big pharma now had all the information it needed to develop drugs from here on out.

Fifteen years later, at Health.Tech in Basel, I heard something eerily familiar, although it was phrased differently.

Today’s message isn’t that research is done. Instead, leadership repeatedly emphasized that finding targets and designing drugs is no longer the bottleneck. The real challenge, we are told, is development, execution, clinical translation, regulatory navigation, and scale.

Meanwhile, with the upcoming patent cliff, mergers and acquisitions are as relevant as ever.

So, what role do data, bioinformatics, software development, and AI play in this cycle?

Clinical Development is the Bottleneck:

FDA approvals have been flat for some time, and the average has hovered very closely to around 48 approvals per year, and 46 in 2025. A recurring theme was that since target identification, and how to develop drugs against those targets is now made easier by AI, as well as the evolving nature of research and high-throughput screening, there is an abundance of potential drugs to develop.

Furthermore, it is increasingly possible to predict toxicity earlier in development, which patient populations are targeted, as well as the long-term profitability of those drugs.

However, clinical trials remain a human-driven process. In fact, about 90% of the budget for a clinical trial goes to human capital and not tech. 

Pharma leaders are now hoping that agentic AI systems can speed things up with CRAs still holding the keys but managing these agents to rapidly increase productivity.

Agentic AI and the self-driving trial:

As always, there is an awareness that clinical trials are slow. Their slowness comes from the fact that enrollment, oversight, documentation, and lack of specialists all contribute to decreasing the rate at which a trial can proceed.

The hope is that agentic AI will speed up these processes, and there are several potential levels of AI involvement. Level 1 is merely assistive, think chatbots helping CRAs to write documentation and e-mails. Level 2, agents taking over more tasks from the CRA, but humans remain in the loop. Finally, at level 3 agents are almost entirely running the show with humans checking to make sure that things are proceeding.

Level 1 is already very much a part of clinical trials right now. Level 2 seems rather feasible (but is task dependent). While a self-driving clinical trial is not likely at this point.

What remains important is how these systems are validated and implemented. Perhaps the greatest challenge is cultivating a data-savvy workforce that understands both the limitations of agentic AI and the realities of clinical execution.

This is where cultivating hybrid profiles becomes especially important, think part data scientist and part clinical research associate. However, the current reality is that the workforce doesn’t yet exist in abundance. A point, which we will circle back to.

Data Fragmentation is an issue where AI can help:

Clinical research organizations often deal with a large variety of data types. However, how these modalities are combined and used to model benefits and risks for individual patients is murky. One speaker suggested that a variety of models should be used for things such as efficacy, safety, subgroups, etc., and combined into an overarching benefit-risk model to help assist clinical trial outcomes; however, as with many of these ideas, the devil is in the details. Building systems like this is not simply a modeling problem. It requires careful validation, integration across multiple data sources, and governance to ensure the outputs are reliable and interpretable. Organizations that can bring together the right scientific, clinical, and data expertise will be far better positioned to make use of these approaches.

Manufacturing next-gen medicines:

Another subject that was brought up is the idea that as new medicines are created, current manufacturing capabilities are too slow to respond. A potential solution to this is to generate digital twins of future manufacturing plants to simulate how well a planned manufacturing plant could adapt and change in response to future demands. Although this idea sounds great, how it is executed is somewhat unclear currently. What seems more certain is that companies will need to begin experimenting with these approaches and develop the internal expertise required to evaluate and operationalize them effectively.

Additional observations:

There was a lot to take in, and I couldn’t capture all the details at the conference. However, I’d like to add some other important takeaways.

• Mergers and Acquisitions are likely to accelerate over the next couple of years.
• AI can potentially help in speeding up processes related to due diligence, but most aspects of BD execution remain human-driven.
• Oncology and CNS are the top investment areas; and despite the obvious need for new antibiotics, they remain undervalued.
• Wearables continue to be exciting, especially regarding how they can be integrated into clinical and disease monitoring contexts.
• Neural implants continue to excite, but those hoping to have their brains uploaded to the cloud need to wait much, much longer.

What does this all mean for pharma/biotech?

Well, although there is a tremendous and reasonable amount of excitement about how models and agentic AI can speed up clinical trials, or how digital twins could improve manufacturing, and some of this hype is reasonable, it is largely revealing that the development of medicine continues to be a slow and costly process.

I think it’s fair to say that there will be efficiency gains because of these technologies, but it remains to be seen if it will mean that fewer drugs will fail upon development. There is reason for optimism, but the magnitude of the effect remains to be seen.

The main point of action is big pharma needs to build a hybrid workforce, it needs people who understand medicine as well as tech and since this profile remains rare, it means there should be investment in building hybrid profiles, and that simply takes time and openness to bring in people who may not have the perfect profile but have a lot of potential.

To circle back to our main question, we can ask: what role do data, bioinformatics, software development, and AI play in this cycle?

The answer is, they can potentially help speed up drug development but only with the right expertise, and developing that expertise requires time, compromise and patience.

Sean O’Toole