Elizabeth Hickman, the Chief Executive Officer at AustinPx, brings over two decades of deep-seated expertise in the biotech and pharmaceutical sectors to the table, particularly within the complex field of bioavailability enhancement. Her career has spanned leadership roles at major industry players like Catalent and West Pharmaceutical Services, providing her with a unique vantage point on the evolution of drug development. As we navigate a transformative era where computational power is redefining the laboratory, Hickman serves as a crucial bridge between digital innovation and the physical realities of the manufacturing bench. This discussion explores the shifting landscape of R&D in the wake of a massive surge in AI investment seen throughout the first half of 2026, the persistent “developability” bottlenecks that computational models cannot yet solve, and the critical need for formulation flexibility when dealing with the increasingly complex small molecules emerging from generative platforms.
With the first half of 2026 seeing a massive surge in AI-driven platform deals, how is this rapid acceleration in drug discovery fundamentally changing the pressure on downstream development teams?
The sheer velocity of discovery we are witnessing right now is breathtaking, but it creates a paradoxical situation where the front end of research is moving much faster than the physical reality of the lab can often keep up with. When we look at the steady run of new AI deals centering on small molecule discovery, we see an industry that is incredibly bullish on the idea that in silico models can reshape the entire R&D landscape. However, this greater discovery speed significantly raises the stakes and the costs of getting development wrong once a candidate finally enters the formulation and manufacturing phase. We are finding that while AI can expand the chemical space and surface candidates with incredible target potency, it does not necessarily make those molecules any easier to develop into a stable, deliverable medicine. In fact, the bottleneck has simply shifted; we are now faced with a higher volume of complex assets that require immediate, high-stakes decisions regarding their physical viability before they can even dream of entering the clinic.
When these generative AI platforms prioritize target potency and selectivity, what are the specific physical and chemical liabilities that tend to emerge, and why do they pose such a challenge for traditional manufacturing?
The reality of generative AI today is that these platforms are often heavily biased toward maximizing target potency, which frequently pushes the resulting molecules into a very difficult territory defined by high molecular weight and high logP values. We are seeing a trend where candidates arrive at our doors with significantly reduced aqueous solubility and high melting points, which are classic physicochemical liabilities that make traditional formulation almost impossible. These “in silico darlings” may look perfect on a computer screen, but they often possess poor solvent solubility or narrow processing windows that restrict our available formulation paths. Because AI models aren’t always optimized for developability, we are forced to deal with assets that are increasingly “greasy” or stubborn, requiring us to rethink how we build a stable dosage form that can actually generate meaningful exposure in a patient. It is a constant tug-of-war between the elegance of the molecule’s design and the grueling reality of its physical properties.
As we move further away from standard compound profiles, how should early-phase teams re-evaluate their legacy screening assumptions to avoid prematurely abandoning promising drug candidates?
Many of the legacy screening assumptions that the industry relies on were built during an era when we were working with a much narrower, more predictable range of compounds. If a team applies those rigid, old-school metrics to the complex molecules coming out of modern AI platforms, they run a very high risk of deprioritizing a “diamond in the rough” simply because it doesn’t fit a traditional mold. In this high-speed environment, the job of an early-phase team isn’t just to act as a gatekeeper that filters out what is “easy” to handle; it is to act as a creative problem-solver that asks if a strong molecule can be supported through advanced formulation design. We have to be careful not to let an asset fail just because its melting point is too high or its solubility is too low for a standard approach. Instead, we need to lean into biopharm assessment and manufacturing planning much earlier to see if there is a viable path forward that bypasses these traditional roadblocks.
What are the fundamental questions that sponsors must answer during the early development stage to ensure a candidate is operationally realistic and not just scientifically interesting?
To keep a program from stalling out or wasting precious material, sponsors need to secure clear, data-driven answers to three foundational questions almost as soon as a candidate is identified. First, we have to determine if the molecule can reach the necessary therapeutic exposure at a practical, human-sized dose, which is often the biggest hurdle for poorly soluble APIs. Second, we need to know if the candidate can remain physically and chemically stable throughout the rigors of processing and long-term storage. Finally, we must confirm that there is a manufacturing path that is truly scalable and repeatable, rather than just a one-off success at the laboratory bench. When these questions are ignored or delayed, we see programs run into massive delays where teams are forced to revisit work that should have been settled months prior, leading to wasted resources and lost time in the race to the clinic.
In an era of high-tech prediction, why is it still necessary to anchor development in a sequence of in vitro screening and in vivo confirmation?
AI is a powerful tool for narrowing down options and prioritizing where we spend our energy, but it cannot yet replace the physical proof that comes from an integrated decision loop. We still need that rigorous sequence of in vitro screening, in silico PBPK modeling, and ultimately, in vivo confirmation to see how a molecule actually behaves in a biological system. These steps allow formulation scientists to define the practical limits of absorption and dissolution, creating a feedback loop that informs whether a specific formulation approach is actually moving the needle on performance. This process ensures that we are making decisions based on how the molecule performs in the real world, which leads to a far more efficient use of limited API and prevents those devastating late-stage surprises that can sink a promising drug. There is no substitute for the tangible data that comes from watching how a formulation interacts with a living system; it is the ultimate reality check for any computational prediction.
When a candidate falls outside the “comfort zone” of standard formulation, what specific advanced strategies can teams use to maintain the molecule’s value while overcoming its physical limitations?
When we encounter molecules with high melting points or limited thermal tolerance, we have to move beyond default paths and embrace what I call “formulation latitude.” One of the most effective tools in our arsenal is the use of bioavailability-enabling approaches, such as advanced amorphous solid dispersions, which can provide a practical path to improved absorption for those notoriously stubborn, poorly soluble compounds. We also look at processing flexibility, such as solvent-free fusion processing, which can open doors when traditional routes are blocked by heat or solvent sensitivities. By expanding the formulation design space—utilizing a broader range of excipients and tailored multi-component systems—we can find that delicate balance between performance, stability, and manufacturability. This flexibility is what allows us to keep the most valuable molecules moving forward, even when their physical properties would have caused them to fail under a more traditional, rigid manufacturing framework.
What is your forecast for the evolution of the CDMO sector as AI-driven discovery continues to flood the pipeline with increasingly complex small molecules?
My forecast is that the industry will see a dramatic shift where the “D” in CDMO—development—becomes the primary differentiator and the most critical stage of the entire drug lifecycle. As AI continues to accelerate the front end of discovery, the pressure on development will grow exponentially, and the organizations that thrive will be those that can master the transition from a digital sequence to a physical product without losing the molecule’s original potential. We will move away from a world of “one-size-fits-all” manufacturing and toward a highly specialized era of tailored formulation strategies that are integrated directly with computational outputs. Success will no longer be measured just by how fast we can find a molecule, but by how effectively we can navigate the complex physicochemical liabilities of these new assets to deliver a repeatable, scalable, and clinically viable medicine. The future belongs to those who can marry the speed of AI with the sophisticated, flexible engineering required to solve the developability challenges of the next generation of chemistry.
