Fireside Chat: Julie Kim

Session Focus

Kim had recently stepped into the CEO role at Takeda. The conversation covered her priorities, the fifteen-year path behind the company’s new narcolepsy treatment, how AI is used across drug development, and where scientists remain irreplaceable.

Two Horizons

Kim set out a two-horizon view for the company before officially stepping into the role, at the year-end earnings call, because Takeda sits at a point of transition in both pipeline and leadership.

Horizon one focuses on successful launches of late-stage assets, continued pipeline advancement, support for core in-line brands, and transforming the organization. She named three: MIMRYLA for polycythemia vera, ORZAFO for narcolepsy type 1, and zasocitinib, whose filing the FDA had accepted days earlier.

Horizon two concerns achieving sustainable growth and scientific leadership in the company’s therapeutic areas.

A 245-Year Company

Takeda was founded in Osaka and celebrated its 245th anniversary in June, with headquarters still in Tokyo.

Kim’s framing of her responsibility: making sure Takeda is around for another 245 years or so.

Her vision has three audiences. Patients with a disease in one of Takeda’s therapeutic areas should want Takeda as the company providing their medicine. Clinicians and healthcare partners should want to research and partner with Takeda. Governments should view the company as a partner in the community.

She noted the company’s US position, which represents half the business, and that Takeda is the largest life sciences employer in Massachusetts.

Fifteen Years to a Narcolepsy Medicine

Cudkowicz raised the near-simultaneous announcements of the Lasker Award for orexin biology and the FDA approval of Takeda’s narcolepsy treatment.

Kim first corrected a common misunderstanding. Narcolepsy type 1 is a rare sleep disorder with a host of symptoms well beyond what is commonly viewed as cataplexy or sudden sleep onset.

What makes ORZAFO significant is that it is the first medicine treating the underlying cause of narcolepsy type 1.

The timeline she described is the substance of the story. The originating discovery dates to roughly 1998 or 1999. Takeda began work turning it into a medicine in 2011, about 15 years ago.

The first molecule demonstrated that an orexin receptor 2 agonist worked, but was not commercially viable because it required continuous IV infusion.

The next was an oral formulation, commercially viable, but a safety signal emerged during clinical development. With other molecules in development, the company first had to determine whether this was a class issue or a molecule-specific issue. Laboratory work confirmed it was molecule-specific, which allowed the program to continue.

The third, TAK-861, became ORZAFO and has now received approval in the US, Japan and China.

Cudkowicz drew the lesson: the value of a company with a long history and the ability to pivot, learn and keep pushing forward.

On what the discovery opens up, Kim pointed to two orexin receptors supporting different indications. Beyond narcolepsy type 1, the company is looking more broadly across sleep-wake, and phase three data suggests potential routes in cognition and fatigue.

AI Across Development

Kim said Takeda uses AI at every step from candidate identification through regulatory submission, reworking how research and development are done to incorporate it.

A significant part is eliminating information gaps, with large volumes of information available instantly. That morning she had been with an R&D team where a scientist described work that used to take literally weeks now being accomplished in 10 minutes.

She was equally clear about the limit. AI cannot yet eliminate the need to test new medicines in people, so that part cannot be compressed, while all other aspects of drug development can be significantly affected.

On clinical trials specifically, she named protocol design as a very challenging part of development that AI makes considerably easier, both in identifying which endpoints to use and in feasibility testing. Her example of the latter: a protocol that appears optimal may, in the field, reduce the eligible population to a very narrow set once all requirements are applied.

Scientists and Machines

Asked about the relationship between machine learning and research staff, Kim framed it as a team with distinct roles.

Her phrase for her scientists was that they have big, beautiful minds, and the goal is leveraging those minds for work only they can do. Routine tasks such as repeating assays can be automated, potentially through physical AI in future, without reducing the need for scientists.

Her argument rested on the narcolepsy story she had just told. Biology is really hard, and the program involved multiple failures before reaching a molecule that succeeded. AI could have helped with some of that, but the creativity of scientists and the understanding of biology that carries an asset from molecule to medicine remain necessary.

Her conclusion: there will always be a role for scientists in drug discovery and development, and those roles will simply look different.

Key Takeaways

1. A 245-year-old company frames strategy in centuries, which shapes how transitions are handled.

2. The narcolepsy program took 15 years and three molecules, failing first on formulation and then on a safety signal.

3. Distinguishing a molecule-specific signal from a class effect is what allowed the program to survive.

4. The approved medicine treats the underlying cause rather than symptoms, and opens routes into cognition and fatigue.

5. AI spans candidate identification through regulatory submission, compressing some tasks from weeks to minutes.

6. Human testing cannot be compressed, which bounds what AI can do for development timelines.

7. Protocol feasibility is an underrated AI application, catching designs that would narrow the eligible population too far.